I am thinking about this because I am currently on a family trip and I have found that in the country I am in right now there are all sorts of problems from inability to access Wikimedia Commons (which yes is a separate project) to slow speeds when accessing Wikipedia.

In addition, almost every browser has an in-built or promoted VPN including Microsoft Edge Secure Network, iCloud Private Relay, Mozilla VPN, etc. And there are free VPN providers such as ProtonVPN that are extremely handy on trips.

I am wondering if we can maybe discuss automatic IP block exemption and potential criteria for automatic, indefinite grants. Maybe:

1. Having a confirmed email address that is not in use on another account;
2. Account at least one year old;
3. Account has made at least 10,000 edits;
4. Not more than 10% of all edits within the last year being reverted;
5. No history of sockpuppetry, meatpuppetry, or misuse of VPNs for nefarious purposes on Wikipedia or another Wikimedia project (which probably there should be a way to cancel the autopromotion).

Aasim (話す（はなす）)19:49, 24 July 2025 (UTC)

The first 2 criteria are completely useless. Apart from that you've probably roughly described one set of criteria that admins should be using to grant temporary IPBE (we often grant IPBE using lesser criteria). An admin should still be looking over the details, like they do for pretty much every other user right. Once you meet the threshold you're automatically good forever? Not a chance. "No history of sockpuppetry", "nefarious purposes", and dodgy logged out editing, are all impossible to detect automatically. These are judgment calls, for which admins and checkusers are paid the big bucks. I also maintain that we should only be granting permissions to people who actually need them. You can define 'need' in various ways, but 'going to use the permission' would definitely fit in there. --zzuuzz^(talk)20:24, 24 July 2025 (UTC)

Also, for the case of "I hit a block" - bypassing the block isn't always the right solution. Blocks can be bad, underlying block reasons can change -- as such fixing the block may be the best solution. (That doesn't cover the case ofI should be allowed to use any VPN or Proxy I want because I've been around a while but is another factor.) —xaosflux^Talk20:54, 24 July 2025 (UTC)

I don't see why the current IPBE process isn't good enough (while promoting security against proxy vandals).Aaron Liu (talk)21:13, 24 July 2025 (UTC)

It is going to be very difficult toWP:GAME 10,000 edits as rate limits make it all but infeasible save for bot accounts (which are already IP block exempt from all but Tor exit nodes). And those 10,000 edits have to be either 1. while manually IP block exempt or 2. via non-proxies or open proxies that have not yet been blocked. If we are really concerned about gaming we could measure account age from first edit. Self reverts would count towards that 10% limit. The more edits and older the account must be, the less likely we will have serious gaming of existing processes.

We can lower the revert threshold to 5% or 1% if we are concerned about gaming. Or double the requirements to 20,000 and 2 years old, at which point fewer than 10,000 Wikipedians would qualify. The fifth point would be a matter of enforcement.

We could also have procedural revokal based off of inactivity, so that there must be a minimum activity requirement to maintain the automatically granted role.

This is just a brainstorm to try to address a legitimate issue with browser VPNs (which many may be unaware that they have them on, yet alone how to turn them off).Aasim (話す（はなす）)12:17, 25 July 2025 (UTC)

The legitimate issue with browser VPNs is already solved by linking to instructionsin the proxy block message template, which everyone sees when they encounter this form of block, for how to disable almost all of the popular ones, or how to whitelist Wikipedia. We don't need to make holes in our security features for people who won't read instructions, and for the editors who have a legitimate use case for proxy use, listen to the functionaries here telling you that it is practically no administrative burden at all to review IPBE requests. Automatic IPBE grants are asolution in search of a problem.Ivanvector (^Talk/_Edits)15:20, 25 July 2025 (UTC)

I am also trying to figure out how VPNs are a unique problem to Wikimedia projects. I am not saying that open proxies aren't problematic, I just think the approaches to them might be a little draconian. VPNs have only gotten more popular in the past decade, because they are extremely useful. I also understand security is important but there has got to be a more intelligent way that doesn't require widespread human intervention where bot downtime potentially leaves hundreds of IP addresses that are currently being used for coordinated disruption open (as many other websites that are not MediaWiki wikis do).

A verified email and phone number can make things really hard; with filtering out of VoIP and throw-away email/phone number services it can be very hard to circumvent account bans. But that might also raise privacy concerns.

Another thing websites and forums do for anonymous users is collect email addresses (that again are not from throwaway providers) and require verification codes to protect against abuse. This might work well for temporary accounts as Wikimedia eventually moves to phase out IP addresses entirely for most purposes. Apple and Cloudflare have been trying to introduce private access tokens to try to encourage legitimate users, and maybe MW can recognize private access tokens and allow the user through despite a VPN block (but then the edit would be attributed to the access token).

I do wonder if this could become a major problem in the future if certain browsers make VPNs mandatory, but that is the discussion happening onm:Apple iCloud Private Relay. If we want to kick the can down the road sure but then there might be entire ecosystems unable to edit Wikipedia. IP addresses might always remain useful especially for autoblocks and non-VPN use cases, but for VPNs something more intelligent to identify different users is needed.Aasim (話す（はなす）)16:21, 25 July 2025 (UTC)

While I believe the linked expression is very overused, I agree with Ivanvector and the usage here. We should wait for the problem to arise first because the security benefits against sockpuppeting are much higher than the bits of convenience automatic IPBE might provide. If things gets to the point where we should have automated granting, we'd see a far cloggier IPBE request queue. I'd rather not risk having the very rare 10k-edit sock elude us than satisfy the rare active editor who can't wait to get their request approved until we see a mensurable need for this thing.

The additional things you propose in this comment won't add anything as they're reusable. Many sockpupeteers used to be legit editors who would've needed to pass anything required of that account, and nothing stops a sock from being verified with the same phone number.Aaron Liu (talk)01:56, 27 July 2025 (UTC)

The idea of a verified phone number on most services is to make it harder to ban evade. Discord already does this; server bans are by username and by IP, but Discord also says that requiring a verified phone number to be granted the role can make evasion very difficult, presumably because we can make it so. Discord does filter out phone numbers that belong to VoIPs and we can do the same thing as well.Aasim (話す（はなす）)13:33, 28 July 2025 (UTC)

I don't think automated detection of using the same phone number as another account is a good idea as phone numbers are very much transferable; i.e. one can hand in their phone number back to their carrier, which can give that same phone number to me, which is how I end up confusedly replying "new phone who dis" to my inbox.

So if you want to do that you would have to expose phone numbers to CheckUsers. Which is an even bigger security thing whom I do not see the rare convenience of this proposal eclipsing. Yes, we trust CheckUsers, but still I doubt it's something Wikipedians are comfortable with, especially in places like China where WMF office-actioned a dozen trusted users for possible governmental collusion.Aaron Liu (talk)16:40, 28 July 2025 (UTC)

It is also important to remember that the United States is not representative of the whole world, not all phone numbers work the same as they do in the US, and not every editor (let alone every potential editor) has a mobile phone of their own.

In the UK I can eaisly get a working, second-hand phone for less than £5 and PAYG and no-contract sim cards for the same or even lower price. If I spent more than 1 minute looking I could probably get it even cheaper than that. It's not free and there is hassle involved, but the barrier for me to be able to verify with multiple phone numbers is extremely low at the same time the identical restriction places an (almost) insurmountable barrier on some people verifying a single account.Thryduulf (talk)18:27, 28 July 2025 (UTC)

I don't see how that changes anything. Is pay-as-you-go supposed to stop you from reselling the sim card to someone else?Aaron Liu (talk)19:24, 28 July 2025 (UTC)

No, PAYG it makes iteasier to resell the the SIM card as there are no contracts involved. You just buy the SIM card and put credit on it.Thryduulf (talk)20:12, 28 July 2025 (UTC)

So that makes the situation I described more common. I don't think you understood, so I'll repeat what I said:

Transferred phone numbers can cause one account to have the same phone number as another, meaning you can't rely on just detecting duplicate phone numbers to find block evasion. And blocked users will just bypass this measure by verifying all their accounts with the same phone number. Unless you give CheckUsers access to phone numbers, which I doubt is a cost the community will accept just for the sole convenience of allowing some editors to access their accounts faster.Aaron Liu (talk)20:28, 28 July 2025 (UTC)

I think we're arguing the same thing (against the proposal) but from opposite angles.

Your argument seems to be that accounts legitimately operated by multiple people could have the same phone number, and I don't disagree with that but it isn't directly relevant to my argument. I'm saying that (a) for some people getting a single phone number is a barrier high enough that requiring a phone number to edit would prevent their participation here; and (b) for other people getting multiple phone numbers is such a low barrier that requiring each account to have a unique phone number would prevent almost no barrier to socking (especially if backed by corporate resources).

The proposal only works if there is a 1:1 relationship between person and phone number. You are (I think) arguing against it because a single phone number could relate to multiple people. I'm arguing against it because a single person could easily have multiple phone numbers. Together that means the relationship is actually many:many.Thryduulf (talk)22:55, 28 July 2025 (UTC)

Now I'm curious - why would someone want to sell a SIM card second hand? Also aren't eSIMs making traditional SIMs obsolete? It is not like you can't get a phone that has no SIM tray unless you are talking the US model of iPhone. But given that most phone numbers are semi permanent (my phone number has not changed since my teenage years, and my parents' have not changed since they got cell phones in California) blocking by phone number makes it really difficult to evade a block as it would necessarily mean going through the costs of acquiring a second phone number.

If there is a way to identify pay as you go phone numbers we can block those as well and have human review based on the circumstances. Nowhere am I suggesting that we don't eliminate the existing channels for manual grants of IP block exemption.Aasim (話す（はなす）)21:59, 28 July 2025 (UTC)

You don't want it anymore—because you're moving somewhere, because you don't like how it looks, or because everyone's blocked you, etc—and you want to get back some of your costs.

eSIMs can be resold too.

I know that manual grants will still coexist. My point is that introducing phone number verification for this is a very bad idea because they touch the private information nerve and far from guarantee uniquely identifying the person behind them. I've also talked about giving CheckUsers phone number access already.Aaron Liu (talk)22:22, 28 July 2025 (UTC)

... okay, the instructions are in{{blocked proxy}}, butnot in{{blocked p2p proxy}} where they're more relevant. We should fix that.Ivanvector (^Talk/_Edits)15:22, 25 July 2025 (UTC)

I'm not sure what problem this is intended to solve, and how it is in keeping with core security policies.No open proxies has been aglobal policy since 2006. It is not just an English Wikipediapolicy. I do a lot of IP block exemptions (it is probably the largest percentage of my logged admin actions), am probably more liberal with it than most people, and I cannot remember the last time that I granted indefinite IPBE. If people need it, they will likely get it. It is not really a hardship to make their case. I've been working on some guidance for admins deciding whether or not to grant IP Block exemptions, and pretty much the first line is "don't grant it indefinitely".Risker (talk)22:38, 24 July 2025 (UTC)

• I remain convinced thatall automatic rights grants are a bad idea, with the exception of autoconfirmed (I have complicated opinions about extendedconfirmed). They invitepermission gaming, and yes we absolutely have bad actors dedicated enough to wait out these arbitrary conditions so that they can keep using their open proxies. It really takes no time at all to evaluate an IPBE request, and as Xaosflux said sometimes granting the permission is a less ideal solution than modifying an overly-aggressive block, or one where there is too much collateral. I pretty much grant IPBE to anyone who bothers to ask - it shows they can read instructions. I also never grant it indefinitely, I thought that was already forbidden by policy. Even my own alt doesn't have indefinite IPBE.Ivanvector (^Talk/_Edits)12:15, 25 July 2025 (UTC)

I understand the desire behind this, but I don't think this is warranted. VPNs are security theater and it doesn't really matter if Wikipedia doesn't allow it. Seeing as its been global policy for nearly 20 years (with good reason--would make Checkuser useless) I don't see a strong need to locally reverse it. Its already open to all editors who ask, and its not very hard to ask. Just like with all rights, if you don't need it, then you don't have it.JackFromWisconsin (talk |contribs)01:31, 27 July 2025 (UTC)

VPNs are security theater Are you sure? They have their use cases, such as when you want to hide the geolocation of your IP address, when you want to protect yourself / your IP address from legal action, and when you need to bypass geo-restrictions. I'll bet mainland Chinese editors of Wikipedia, for example, would not consider VPNs to be security theater. –Novem Linguae(talk)11:12, 14 August 2025 (UTC)

Sure, but most people who use VPNs are notgay pirate assassins.--Ahecht (TALK
PAGE)20:16, 8 September 2025 (UTC)

• I agree with others that this seems unnecessary. The only time it would be useful would be if a long-term, active editor editor is unaware of the proxy policy until the first time they try to edit using a proxy (plausible), and that first proxy edit attempt happens in a context where they are unable to disable the proxy briefly while they request a block exemption (less plausible). So it's a rare situation to begin with, and even in that situation, we only lose a few hours/days/weeks of their editing until they get to a context where they can request an exemption. It just seems like the auto block has large benefits and small costs, so it's unnecessary to change it. --LWG^talk18:56, 28 July 2025 (UTC)

  @LWG Or the much more common scenarios of a long-term, active editor needing to edit from a mobile device away from WiFi coverage, only to discover that they subscribe to one of the many mobile operators whose entire IP range is blocked, or they move to an area only serviced by an ISP that is subject to a rangeblock, or they need to edit from a corporate network that run all traffic through filtering software hosted on AWS, Google Cloud, or Azure.--Ahecht (TALK
  PAGE)20:05, 28 July 2025 (UTC)

  Ahecht right, but how often are those edits going to be so important that it's a big deal to take a wiki break for a couple days while you request a block exemption? --LWG^talk22:12, 28 July 2025 (UTC)

• Support PerWP:IPBE, "Administrators and bots are always exempt from such blocks..." and so it's quite reasonable that other long-standing trusted users should also be exempt.Andrew🐉(talk)20:47, 29 July 2025 (UTC)

  I hope you are aware this is the idea lab. If you think it is a good idea maybe you can help work it into a proposal that has a chance of passing.Aasim (話す（はなす）)22:31, 29 July 2025 (UTC)

  The people with power – the admins – don't want to pass this because they are already exempt, it would weaken their power a bit and so there's nothing in it for them. I expect that it would have to happen as part of a WMF initiative affecting user accounts like the newTemporary Accounts. Don't hold your breath.Andrew🐉(talk)23:03, 29 July 2025 (UTC)

  You know, as an admin I am offended that you think so little of me.Donald Albury23:20, 29 July 2025 (UTC)

  Do you have any evidence at all for this assumption of bad faith of all administrators?Thryduulf (talk)23:46, 29 July 2025 (UTC)

  How does that address what you replied to?Aaron Liu (talk)00:02, 30 July 2025 (UTC)

  Please read from the top of this pageThis page isnot forconsensuspolling. Stalwart "Oppose" and "Support" comments generally have no place here. Instead, discuss ideas and suggest variations on them. Others have stated potential problems with the idea that might need to be addressed to have a better (albeit very low) chance of passing.Aasim (話す（はなす）)12:26, 30 July 2025 (UTC)

• Please seeWhy everyone should use a VPN for further evidence that this is a normal and respectable way of accessing the Internet.Andrew🐉(talk)16:12, 9 August 2025 (UTC)

  Are paid VPNs blocked as well? I was under the impression that "open proxies" in "No open proxies" only meant relays anyone could access for free.

  The policy isn't because Wikipedians find VPNs non-respectable either. Wikipedia also blocks T-Mobile cellular users and T-Mobile is a perfectly normal carrier. It's solely for protecting against sockpuppetry.Aaron Liu (talk)17:16, 9 August 2025 (UTC)

  what?? I've never had trouble editing from a tmobile connection (practically all my mobile edits)... (t ·c)buidhe03:13, 21 August 2025 (UTC)

  WP:T-MOBILEAaron Liu (talk)03:43, 21 August 2025 (UTC)

  Paid VPN services are blocked too I believe. The whole reason they are blocked is because one can again quickly change their IP by logging out and logging back in. But for workplace VPNs, the IP address range is a lot narrower if not only a single address.Aasim (話す（はなす）)04:38, 21 August 2025 (UTC)

  The definition is inopen proxy: if you can get to it from anywhere on the Internet, it's an open proxy.IznoPublic (talk)00:01, 26 August 2025 (UTC)

• Now that we know the WMF hands over editors' IP addresses (among other things such as email addresses) if they lose a lawsuit (per this comment[1]; discussion:[2]), maybe this idea isn't so bad after all.Some1 (talk)18:30, 9 August 2025 (UTC)

  I don't see how that changes things. The downsides of sockpuppetry still outweigh the benefits, benefits which are far smaller than you might think as we do have a working manual review system.Aaron Liu (talk)21:03, 9 August 2025 (UTC)

• Support because I also meet all of these conditions and have to switch off my VPN every time I edit, which is fine at home, but impossible while traveling.Nswix (talk)12:02, 24 August 2025 (UTC)

  Again this is a place for ideation rather than consensus polling. Please read the top of this page.Aasim (話す（はなす）)15:52, 24 August 2025 (UTC)

I have an idea for a new section for the mainspace of Wikipedia.

Wikipedia:Articles for improvement is an obscure initiative that seeks to highlight subpar articles with the aim of improving them. The project is currently nearly dormant.

I propose reviving the project on the main page, as a template (like TFA or DYK). The idea was previously discussedhere with a rough consensus in favor, with dissents that pointed out issues such as vandalism or ambiguity in the proposal. I created a (nonfunctional) mockuphere.

I would appreciatea) discussion about the proposal, so that we may iron out any issues or ambiguity, andb) someone technically proficient at Wikipedia helping me create a functional mockup.

Many thanks in advance!Bremps...20:31, 25 August 2025 (UTC)

Yes. We'll get a flood of incompetent edits, but (1) most of them will be from well-meaning people and (2) it will likely attract new people onto the project. I've long said the Main Page needs much more reader engagement on it, and that the WMF should advertise "learn to edit" rather than "give us your money". This is a step in the right direction.Cremastra (talk ·contribs)20:58, 25 August 2025 (UTC)

For historical context: when the articles for improvement section was on the main page in the past, there wasn't a flood of edits. Edits were primarily from the people already involved with the initiative.isaacl (talk)22:22, 25 August 2025 (UTC)

What format did it appear on in the main page? There may need to specifically be a note that YOU, the READER, can add sources and edit the text.CMD (talk)02:52, 26 August 2025 (UTC)

For reference, this is how themain page appeared on May 4, 2013.isaacl (talk)16:25, 26 August 2025 (UTC)

Thanks. There is definitely scope to test different formats.CMD (talk)11:55, 27 August 2025 (UTC)

I notice that the section was comparatively very small, and easy to overlook for casual readers not familiar with Wikipedia, which is the opposite of what we want to achieve. By making it larger and adding an encouraging editing tip, it could be more effective.ChaoticEnby (talk ·contribs)13:44, 27 August 2025 (UTC)

Indeed. It also provided no explanation of what exactly was meant to be done or how to edit.Cremastra (talk ·contribs)14:14, 27 August 2025 (UTC)

This is why the pairing I'm having in mind could have good synergy – teach users one specific aspect of editing (adding references, removing promotional wording, writing out-of-universe, ...), and give them the chance to apply it on some relevant articles for improvement.ChaoticEnby (talk ·contribs)14:21, 27 August 2025 (UTC)

As a (very) frequent reader, but an under-experienced, sporadic editor of Wikipedia, I'd like to offer my opinions on this subject, as I often find myself wanting to contribute in simple ways, but not sure where to look.

For one thing, I rarely find myself on the main page anyway. Anytime I want to simply look something up here, I type "wq" (search keyword) then, "thing I want to look up" into the address bar in Firefox". And while I can see the benefit of advertising on the main page to attract new people, is that really the issue? It already says, right at the top: "Welcome to Wikipedia, the free encyclopedia that anyone can edit." Is there really anyone who has heard of Wikipedia, and is NOT aware that they can open an account, and start editing? "Oh anyone can just write whatever they want on there!" is what my own beloved mother, and too many other folks, say to disparage the credibility of information on Wikipedia. So I think it's not a matter of making that aspect more visible, though a reminder probably won't hurt.

The problem I see, at least for people like myself, is the difficulty I find in just navigating the massive labyrinth of behind the scenes stuff, and easily finding stuff to do, as an editor. Um.. okay, I feel kinda stupid now, because I kinda just discovered the "Main Menu" button in the top left, ah geez.. I mean okay that should have been super obvious, but, for some reason I've always focused so much more on the buttons on the top right okay? that HUGE distracting globe icon just goes to the main page, so I guess I've kinda been ignoring the smaller menu button next to it, which I didn't even realize has direct links to some of the stuff I've been struggling to find...

Okay but as for AFI stuff specifically, it would be nice if I just had an obvious button for that, showing a list of AFI's, ranked by parameters that I was able to customize. or maybe an option in that drop-down menu at the top right? possibly between Watchlist and Contributions. Or perhaps some sort of gadget, enabled by default in the preferences, and customizable, that would notify me of things in need of doing. And perhaps an ability to quickly, and easily flag articles for an AFI list.

I guess what I'm really trying say here, is that instead of putting AFI on the Main Page, it should be somehow integrated directly into the registered users' GUI, tailored by default to simple editing tasks for new users, but customizable to fit the needs of more advanced editors. Some thing that would generate a list of editing tasks that I could just go straight to work on.OwlParty (talk)10:15, 7 September 2025 (UTC)

I think what you describe is already very close to another feature we currently have! It's atSpecial:Homepage (not the same as the main page!), and provides users with a "dashboard" of articles they can improve, among other things. You can take a look at it and tell us whether that's what you had in mind.ChaoticEnby (talk ·contribs)13:05, 7 September 2025 (UTC)

@OwlParty asked:Is there really anyone who has heard of Wikipedia, and is NOT aware that they can open an account, and start editing?

Answer: Yes. In fact, professionaluser research on Wikipedia readers indicates that most readers don't seem to know this. And they're horrified when you tell them. (Also, you don't have to create an account to edit.)WhatamIdoing (talk)00:38, 9 September 2025 (UTC)

Wow, I had no idea.OwlParty (talk)02:02, 9 September 2025 (UTC)

I think input from the participants in the articles for improvement initiative is essential. I don't see any discussion of this atWikipedia talk:Articles for improvement, but perhaps I overlooked it?isaacl (talk)22:20, 25 August 2025 (UTC)

Good point. I just sent out a noticehere.Bremps...23:14, 25 August 2025 (UTC)

Hmmm... Right now, the sections on the main page are purely reader-serving, and don't intend to attract editors. I think encouraging more editing is a good thing, but should be done carefully, as to not obstruct the encyclopedia's purpose.

I'm not sure if AFI is a good idea for new editors. Usually the articles need prose writing and research, and have tags and large structural issues. Something small like copyediting or adding short descriptions for example are easier places to start. I feel as this idea will likely attract large amounts of editors, and dissuade most of them, when they find that their edits have been reverted for OR, RS, etc. —PerfectSoundWhatever (t;c)23:28, 25 August 2025 (UTC)

I'd support some sort of ongoing funnel of readers and new editors toWP:TASK._signed,Rosguill^talk16:28, 26 August 2025 (UTC)

I suggest joining the discussions atWikipedia talk:Growth Team features about suggested tasks on each user'shome page. (The username link at the top of the page links to the user's home page automatically for new users; users with older accounts can enable direct access to their home page atPreferences →User profile →Newcomer editor features →Display newcomer homepage.)isaacl (talk)17:00, 26 August 2025 (UTC)

I think it's reasonable to have these concerns. I have them myself. If the trial run devolves into anarchy or stagnation we may have to pull the experiment. But I think we should still err on innovation given that we're trying to keep Wikipedia dynamic to keep up with the times.

From a cost-benefit standpoint, the worst that can happen is that a page is no better off than when it was showcase. The best that can happen is a large recruitment drive for an encyclopedia that's been taken over by ChatGPT in traffic.Bremps...02:35, 27 August 2025 (UTC)

Definitely agree – now is the time to innovate and try new things, and I don't see any negative effects on our reputation that "here are articles you can improve!" could have (compared to, say, the Simple Summaries trial). If anything, even if it doesn't work out, it still showcases Wikipedia as more dynamic and inviting for readers/potential editors.ChaoticEnby (talk ·contribs)11:41, 27 August 2025 (UTC)

I think this is a good idea. This is a part of the Wikipedia community that deserves a bit more love, honestly.Danubeball (talk)02:38, 26 August 2025 (UTC)

I agree with the above, that the project's effectiveness could be improved.Sahaib (talk)06:44, 26 August 2025 (UTC)

I would definitely agree – one other thing I had in mind was "this week's editing tip", and both could maybe go hand in hand? Learning how to deal with a specific type of issue, and having an article to apply your knowledge on.ChaoticEnby (talk ·contribs)15:14, 26 August 2025 (UTC)

I like the idea.Bremps...02:36, 27 August 2025 (UTC)

Note: I have requested technical help atWikipedia:Village_pump_(technical)#Requesting_help_for_creating_mockup_of_proposal.Bremps...16:30, 29 August 2025 (UTC)

I'm concerned about the lack of comment from those who currently support the articles for improvement initiative. Unless there is buy-in, or a new influx of volunteers to help support the new process (in collaboration with the existing process), I'm reluctant to direct a group of volunteers that they must now spend time supporting a main page section.isaacl (talk)16:41, 29 August 2025 (UTC)

The only WikiProject members that seem active are Danubeball, Sahaib, and PerfectSound who've already commented above as well as @SVcode and @Clovermoss.Aaron Liu (talk)18:30, 29 August 2025 (UTC)

My interest in the wikiproject has come and gone over the years. It's definitely an interesting idea but actually getting everyone interested in a specific article doesn't always work. If there's interest in spotlighting articles that need some work on the main page, maybe implementing something likeWikipedia:Community portal would be better? I was drawn to that when I was a newbie.Clovermoss🍀(talk)18:51, 29 August 2025 (UTC)

As part of "buy-in", I think there should be some discussion among these volunteers about what needs to be done on a regular basis to keep a main page section going and the queue full, which was a problem last time. (Although some agreement on success criteria might be nice, in practice that'll just be rehashed in future anyway, should someone propose removing the section.)isaacl (talk)21:29, 29 August 2025 (UTC)

A queue implies options, which is good. I think it'd be very chaotic to point at any article in particular and then encourage the thousands of people looking at the main page to improve it, especially without guidance.Clovermoss🍀(talk)06:57, 30 August 2025 (UTC)

By "queue", I was referring to a queue of what will appear on the main page in order to ensure that it is updated in a timely manner, which did not always happen last time and resulted in the section being pulled. However, note the initiative was called "Wikipedia:Today's articles for improvement" because it did shift from initially suggesting one article into suggesting several, as can be seen in the example I previously linked from 2013. However, this change was seen as diffusing focus, which I believe is why there is only one suggested article per week in the current format.isaacl (talk)17:12, 30 August 2025 (UTC)

I'd expect the article to receive more watchers/RC patrollers as well.Aaron Liu (talk)23:03, 30 August 2025 (UTC)

Who here would be willing to participate in the queue process? I would be, see my comment below.Bremps...17:19, 1 September 2025 (UTC)

Regarding buy-in, I would be an avid participant in the process. In terms of the workload, it would be less than, say,WP:DYK orWP:ITN, as we are promoting articles that are by definition not good. The key issue in my opinion is article selection, which will be more contentious because the template is going on the main page.Bremps...16:46, 1 September 2025 (UTC)

Just came across an IP editor who is trying to get some articles into mainspace:123 This editor has been making good faith contributions and I would like to help them, I just can't. Their IP keeps changing all the time so it is impossible to have any kind of communication with the editor.Communication is required. The editor won't even see that their drafts have been declined or why.TurboSuperA+_[talk]06:24, 5 September 2025 (UTC)

It's an unfortunate loss that the IP in question has a dynamic account, but one incident is not a strong case for such a large change. Some IPs are relatively stable, and some IP editors do check back on places, allowing for some communication. I'll add some welcome templates, just in case they work.CMD (talk)06:36, 5 September 2025 (UTC)

I wonder if IPv6 addresses could get a unified Talk page for the subnet. Someone correct me if I'm wrong, but I think that usually all IPv6 addresses on the /64 subnet are assigned to the same household.TurboSuperA+_[talk]07:01, 5 September 2025 (UTC)

Yes, this is true; a /64 is one household.

Also, I managed to create a talk page for a /64 subnet (User talk:2600:4041:5CE9:B300:0:0:0:0/64) somehow, so it's definitely possible, but I'm not sure if the IP editor actually got the message I left there.SuperPianoMan9167 (talk)22:07, 5 September 2025 (UTC)

I might also have created a talk page for a nonexistent user, but that seems unlikely since the talk page does indeed show up in block logs for the IP range.

When temporary accounts are deployed, I think that much of the issues currently affecting IP users will be addressed (for example, the confusion caused by dynamic IP allocation and the inability to ping IP editors).SuperPianoMan9167 (talk)22:11, 5 September 2025 (UTC)

(Sorry for the broken-up replies) @TurboSuperA+: try creatingUser talk:2001:1970:49DE:8C00:0:0:0:0/64 directly and see if the user gets the message.SuperPianoMan9167 (talk)22:14, 5 September 2025 (UTC)

So there is a talk page for the /64! Thanks for this.TurboSuperA+_[talk]07:35, 6 September 2025 (UTC)

IP4 editors often have a stable IP address, but IP6 from mobiles tends to change much more. I assume that the imminent rollout oftemporary accounts will reduce this issue and allow more viable feedback to the user's talk page —GhostInTheMachine^{talk to me}07:04, 5 September 2025 (UTC)

That doesn't explain how the temporary accounts will work. If the temporary account is tied to an IP address, then doesn't that mean that the username will change when the IP changes?TurboSuperA+_[talk]07:07, 5 September 2025 (UTC)

The system adds a cookie so that the returning user can be identified as aprevious IP editor —GhostInTheMachine^{talk to me}07:19, 5 September 2025 (UTC)

The temporary account is tied to a browser cookie and not an IP address. The same temporary account can have edits made using multiple IP addresses (which will solve a lot of the confusion regarding IP editors on a /64). The only caveat is that the temporary account, being temporary, only lasts for 90 days or until the cookie is deleted.SuperPianoMan9167 (talk)21:57, 5 September 2025 (UTC)

Isn't one of the primary rationales for draft space existing to allow unregistered editors to create drafts? In any case, unimproved drafts are deleted after six months, so there is no inherent cost to the project to allow such creation.BD2412T22:20, 5 September 2025 (UTC)

There is always an inherent cost in volunteer time. The theory is that it is made up for in the volunteer time that can be contributed by unregistered editors. My off-hand suspicion is that for any non-llm drafts that theory holds up quite well, for example the drafts created by the IPs here feel like they have potential to form content even if not as standalone articles.CMD (talk)00:29, 6 September 2025 (UTC)

Anonymous ip should not be allowed too post at all IMOH.Jp33442 (talk)00:32, 6 September 2025 (UTC)

That goes against the whole point of Wikipedia. As it's a "free online encyclopedia that anyone can edit." –The Grid (talk)14:19, 9 September 2025 (UTC)

the drafts created by the IPs here feel like they have potential to form content even if not as standalone articles.

Exactly!TurboSuperA+_[talk]07:36, 6 September 2025 (UTC)

Allowing unregistered editors to created articles is the reason thatWikipedia:Articles for creation exists.

The Draft: namespace exists because a WMF staffer thought it would be a good place for collaboration. It turns out that he was wrong.WhatamIdoing (talk)20:18, 6 September 2025 (UTC)

You've said that about the draft namespace before. As I believe I've already pointed out to you, the draft namespacewas proposed by an English Wikipedia user (then TheOriginalSoni, now just Soni) at the proposal village pump and received community consensus.isaacl (talk)04:01, 7 September 2025 (UTC)

The idea did not originate in that RFC.WhatamIdoing (talk)06:05, 7 September 2025 (UTC)

The point is that it was proposed and discussed within the English Wikipedia community. It wasn't created because one WMF staffer thought it was a good idea.isaacl (talk)17:58, 7 September 2025 (UTC)

Solely and exclusively because? Of course not. (Also, he's a long-time enwiki editor.) The relevant point for this discussion is that AFC predates and is independent of the Draft: namespace. The Draft: namespace was not created for AFC, nor AFC for the Draft: namespace. The fact that the Draft: namespace ism:where articles go to die is less relevant, though no less true.WhatamIdoing (talk)01:13, 9 September 2025 (UTC)

Sure; your summary in your reply to BD2412 is just very reductive of the history, seeming to lay all the responsibility on one person and their opinion.isaacl (talk)02:32, 9 September 2025 (UTC)

WP:POINT applies here. –The Grid (talk)15:58, 9 September 2025 (UTC)

The editor won't even see that their drafts have been declined or why. They can see that for all of the drafts that have been declined by looking at the draft, for exampleDraft:Visarion II, Metropolitan of Cetinje. Generally, the draft's talk page should be the place where any discussion about the content is held. —Kusma (talk)09:46, 7 September 2025 (UTC)

With temporary accounts soon to be implemented, it seems that the communication issue will be resolved (as they are not tied to individual IPs but to browser cookies), so I don't think there will be much to worry about.ChaoticEnby (talk ·contribs)13:36, 7 September 2025 (UTC)

It is an established fact that gun sales in the United States increase after mass shooting events.[3][4][5][6] Including firearm brands and models in articles about mass shootings and shooters is akin toadvertising.

I think we should have aMOS:FIREARMS section inWP:MOS that would say that inclusion of gun brands and models is strongly discouraged unless absolutely necessary and relevant to the article (as determined by consensus on the article talk page).TurboSuperA+_[talk]04:43, 3 September 2025 (UTC)

At a quick glance at those citations, I see a clear correlation between violence and gun purchases, but I don't necessarily see a clear causation, and also don't see anything linking specific makes/models. What am I missing in these citations, or do I misunderstand your thesis? With regards more broadly if we're contributing or advertising in a promotional sense, I would fall back on not-censored -- we do not shy away from topics people find objectionable, even if they might seem to shine a good or bad light on a specific topic. We're interested in verifiable facts, not necessarily shaping WP in alignment to our personal beliefs or morality.TiggerJay (talk)05:24, 3 September 2025 (UTC)

• Bad idea perWP:CREEP andWP:RGW. Those studies furthermore have pretty limited scope and unclear significance and even some self-contradictory parts. There's also nothing I see that has any specifics on linking make/model sales to those used in specific incidents, or that anyone needs to look at Wikipedia for that information.35.139.154.158 (talk)05:25, 3 September 2025 (UTC)

Truthfully, I see where you're coming from, but TiggerJay is correct,WP:NOTCENSORED is a real reason to not preemptively exclude such information. That being said,WP:NPOV (specifically,WP:DUE, but alsoWP:BALASP) should be considered when attempts are made to say exactly what brand/model of firearm was used for the mass shooting of the week. Also considerWP:IINFO,WP:VNOT andWP:ONUS. Taken together, unless there's significant RS coverage of these kinds of details (and not just a passing mention in a minority of sources), these are all good reasons to consider excluding (and they're alreadyWP:PAG, so no need for more instructions). If there's disagreement, start anWP:RFC to help determine whether there is consensus to include or exclude. —Locke Cole •t •c05:35, 3 September 2025 (UTC)

I agree totally with LC above, already covered by WP:DUE etc if properly applied. It really doesn't matterwhy people are obsessed with such details - if they can't provide evidence of the level of WP:RS coverage needed, out it goes.AndyTheGrump (talk)06:19, 3 September 2025 (UTC)

This sort of very low-level content prescription isn't usually in scope for the MoS, so I thinkWP:CREEP applies here.Locke Cole is quite correct -- if a detail isn't generally reported in high-quality, reliable sources, we shouldn't be reporting it anyway perWP:DUE. On the other hand, if itis generally included in those sources, we shouldn't exclude it just because we find it distasteful or worry about how people may react to it. The closest similar guidance (regarding whether to include or exclude certain factual details) isMOS:DEADNAME, which follows more-or-less the same principle -- a former name under which someone was widely known in reliable sources should be included; one under which they weren't covered in RS shouldn't.UndercoverClassicist^T·C06:43, 3 September 2025 (UTC)

I see little value in routinely including commercial brand names of firearms in articles about mass shootings but I think the information about the model and the caliber of ammunition used is relevant. For example, there has been extensive discussion for years about howAR-15–style rifles have certain features that make them common weapons of choice in these crimes, and many readers will wonder whether or not one was used in a given mass shooting. Many other models of firearms have also been used in these crimes. Similarly with caliber. The.223 Remington and the very similar5.56×45mm NATO are the most common cartridges used with AR-15s but many other calibers and cartridge are also used with AR-15 models that are available with a variety of chambers, and I think that information, when well referenced, is of encyclopedic value as well.Cullen328 (talk)07:26, 3 September 2025 (UTC)

I agree with Locke Cole, and with S Marshall below. @Cullen328 when you sayI think the information about the model and the caliber of ammunition used is relevant. that's fine but irrelevant. What information is relevant varies by the individual article based on what information reliable sources determine is relevant. For example, while I haven't followed the news coverage of theAnnunciation Catholic Church shooting that closely, I don't recallany mention of the ammunition at all, let alone type and brand, and nothing more detailed about the weapon used than rifle, shotgun and pistol. I'm sure it's out there in some reliable source, but unless there is extensive coverage of either of those facts it's simply not DUE - and DUE is the only policy or guideline I think we need for situations like this.Thryduulf (talk)10:32, 3 September 2025 (UTC)

Thryduulf,WP:DUE deals with how to handle a range of viewpoints and the words "view" and "viewpoint" are used consistently in that section of policy. The example used is theflat Earth notion. The type of weapon used in a crime is an objective factual matter and is not a "viewpoint". There are not a range of viewpoints about which firearm was used in a crime. If reliable sources report that competent legal authorities verify that an AR-15 was used in a crime, there are not a range of viewpoints arguing that it was really a bolt action deer hunting rifle or a Colt .45 revolver instead. Applying WP:DUE to objective factual matters is a misreading of policy.Cullen328 (talk)16:37, 3 September 2025 (UTC)

DUE applies to how much weight to give to factual matters as well as viewpoints.Thryduulf (talk)17:39, 3 September 2025 (UTC)

Looks like, until somewhat recently,WP:MINORASPECT was a subsection ofWP:DUE, while now it's a subsection ofWP:BALANCE. That may be a source of some of the confusion here.Anomie⚔17:44, 3 September 2025 (UTC)

• This is too prescriptive, too detailed, and too low-level. The manual of style already overreaches way beyond "style" in far too many places, and it needs to be cut back, not expanded.—S Marshall T/C07:44, 3 September 2025 (UTC)

  - But, as always, if you want it done right, BOLDly do it yourself. Seems to me that simply stating meta/tangential/strictly-off-topic opinions like yours is rarely very effective, even when a handful of editors offer two thumbs up. A more direct, more focused approach is necessary; without that, good ideas can bounce around non-fiery Wikipurgatory indefinitely. We've all witnessed that.

  WikiProject Manual of Style is "defunct"(sounds vaguely like "defucked"); I would support aWikiProject Manual of Style reform (full-bodybattlearmor required). I would not only support its creation but would become an active member unless and until I decided I was wasting my time. But that in itself is an idea lab discussion unless one can BOLDly create a WikiProject. ―Mandruss ☎ IMO.10:23, 3 September 2025 (UTC)

• I don't see any need for specific guidance. It is already policy that article content is subject to consensus on the talk page.Phil Bridger (talk)10:13, 3 September 2025 (UTC)
• Unnecessary as inappropriate inclusions of this kind of information can already be removed if consensus to do so exists, or if an argument for removal can be made fromWP:DUE orWP:OR. If the content survivesWP:DUE objections and can be cited to significant coverage in reliable sources, thenWP:NOTCENSORED. --LWG^talk15:32, 3 September 2025 (UTC)
• This is a common problem in mass shooting articles, but I don't think a new policy is needed. Exact details about make, model, calibre etc should be included only if they have significant coverage in reliable sources. I've seen numerous mass shooting articles where users added this type of info, but a lot of it was poorly sourced orWP:OR.--♦IanMacM♦^{(talk to me)}16:48, 3 September 2025 (UTC)

  I agree that this has a distinct whiff of US politics about it, but I think that where the information is well-sourced, too much detail is a "problem" that we don't need to worry about here. This is an encyclopaedia that has articles about individual episodes of US TV shows; individual townships in Where the Heck, Utah (pop. 99); individual species of extinct coleoptera; more than 4,000 words about sexuality in Star Trek; and quite a well-written one about that precognitive octopus. The fact that Johnny R. Fruitcake used a 9mm Penis Extender 9000 for his shooting spree in the Tiny Tots Preschool is hardly the most urgent detail to remove.—S Marshall T/C17:26, 3 September 2025 (UTC)

  I often remove things that are not the most urgent things to remove. And I have little problem with a community consensus to remove them. ―Mandruss ☎ IMO.18:32, 3 September 2025 (UTC)

• I agree with the OP. These articles can get way into the details of the guns and ammo. It's usually sufficient to simply say "an automatic rifle" etc.. without needing the brand and/or model - unless that is a factor. I have observed when gun information is generalized, invariably somebody will re-add the specific brand and model of the gun. For some editors, it is a focus of their attention. --GreenC18:09, 3 September 2025 (UTC)
  • I do not agree with the OP about the MOS (yet). That's a very high level of consensus that takes time to achieve. The place to start is with an essay. Write a good essay describing the problem, why it's not a good idea, alternative suggestions. Create a capletter redirect likeWP:FIREARMS then start using it in edit summaries. See if it catches on and other people start using it and start contributing to the essay. When you run into conflicts,start RfCs. This is very important. You need a record of dispute over the issue. Once you have 5 or 10 RfCs that have shown clear community consensus, now you got something that has weight, This is when it gets added to MOS. The pathway I describe is exactly what happened with the essayWP:FRAUDSTER which was then added intoMOS:CONVICTEDFELON. Why? Because we demonstrated the problem, described it, linked to the essay repeatedly, showed overwhelming consensus in many RfCs, and that was it. It got added to the MOS. — GreenC18:29, 3 September 2025 (UTC)
• I agree with the OP, GreenC, and a number of others. For those saying local consensus is enough, I say how's that working out so far? When something hasn't worked very well after years of "trial", try something else that stands a reasonable chance of working better. Repeat as necessary, or until you run out of somewhat viable ideas. Our respective personal crystal balls have unacceptably high error rates that are largely masked because we accept what they tell us (i.e., safely maintain status quo) far more often than we put them to the test (i.e, try a change and see; the sky shall not fall if the trial fails). ―Mandruss ☎ IMO.18:27, 3 September 2025 (UTC)
• This information should be included if and only if reliable sources cover it in reasonable detail. If it isWP:DUE, it should be included, if it is not, it should not. Don't make a rule about it (WP:MOSCREEP). —Kusma (talk)18:35, 3 September 2025 (UTC)

• In most cases spelling out makes and models is irrelevant trivia,WP:UNDUE, and often a form of POV pushing. I support the OP proposal for an addition to the MOS.Iggy pop goes the weasel (talk)19:09, 3 September 2025 (UTC)

  I think it's been years since I saw one of these disputes, but the last time I did, I remember seeing editors assert that there was already a rule against this. They might have been referring toWikipedia:WikiProject Firearms#Criminal use.

  AIUI for most articles about crimes (including but not limited to mass shootings), editors preferred naming a class of firearm (e.g., "Handgun" or "Homemade firearm" or "AR-15–style rifle") rather than a specific brand name or model number.WhatamIdoing (talk)17:19, 4 September 2025 (UTC)

• The mention of specific firearms in mass shooting articles, and mention of mass shootings in specific firearms articles, have long been contentious. See e.g.The Verge in 2018. My perception, similar to that story, is it's predominantly been the pro-gun folks trying to erase connection between "mere tools" and what those tools have been used for. Personally, I don't get the push to censor the weapon used. I get that in an article about a firearm, there needs to be some substantial coverage focused on that weapon in relation to one or more events to satisfy WEIGHT, but the weapon used in a mass shooting/mass killing event seems like a pretty basic piece of information, and I don't see a good enough case to violate NPOV here.
  I do get the impulse to look for ways to reduce harm, but if we're going to pull an IAR with regard to NOTCENSORED, how about:

• juststop creating biographies of mass shooters? I noted a while backthat our article on Anders Behring Breivik is200,000 bytes, going into intricate details of his beliefs and activities to an extent that dwarfs articles about far more important historical figures.
• have a higher bar to include the name of suspects/shooters in event articles?
• stop racing to create articles on mass shootings and then relying onWP:RAPID (as opposed toWP:DELAY) to prevent it from being deleted?
• get rid of the mass killing leaderboard articles?

Do we even have any evidence of a mass shooter going to Wikipedia to find out which gun was used, and then using that gun? Because we do have evidence of mass shooters explicitly wanting a Wikipedia page (e.g.[7] [[8]). Simply conceding the "guns are just an incidental tools that have nothing to do with a mass shooting" narrative seems like a step backwards, not forwards. —Rhododendrites^talk \\17:33, 4 September 2025 (UTC)

I'm very much on the side of not including excessive detail of gun brands and models in articles on most mass-shootings, but this has absolutely nothing to do with NPOV, NOTCENSORED, reducing harm ortrying to erase connection between "mere tools" and what those tools have been used for. indeed I'm firmly pro-gun control (an approach similar to the one we have in the UK would be a Good Thing imo) and argued for the inclusion of the suspect's name in theKilling of Austin Metcalf article. My objection is based entirely on the detail being, in most cases, undue weight. In some shootings it makes a difference whether a handgun or rifle was used, and some characteristics of those weapons can be relevant (automatic vs manual for example). However, in the absence of detailed commentary in reliable secondary sources about things more specific than that, such detail in our article is unencyclopaedic.Thryduulf (talk)18:56, 4 September 2025 (UTC)

Your argument is functionally indistinguishable from the one I mentioned. You're both creating a higher bar for WP:WEIGHT as it applies to a particular data point, and writing off objections with a hand-wave at what's "encyclopedic". —Rhododendrites^talk \\19:08, 4 September 2025 (UTC)

No, I want thesame bar of WEIGHT to be applied as it is for other aspects that are not central to the article.Thryduulf (talk)19:29, 4 September 2025 (UTC)

Great! But the bar forWP:WEIGHT is not "detailed commentary". Just using the article you mentioned for convenience, although it's not about a shooting, where is the detailed commentary of his birthdate, GPA, his height, the name of the jail he was taken to, the amount of the bond? More likely, these bits of information are merely mentioned in multiple sources, establishing weight. What you are arguing is that information about a gun must meet ahigher bar of not just being included as part of the story in multiple reliable sources about the subject but "detailed commentary". —Rhododendrites^talk \\19:54, 4 September 2025 (UTC)

The bar for WEIGHT is frequently some sort of detailed commentary.

Here's an example:James W. W. Birch was murdered in 1875. The Wikipedia article says they "speared him to death". It does not say "they speared him to death using a long single-edged sword with a steel blade and a curved black handle with a tooled leather grip and smallguard" – even though that is true and verifiable.

Under what circumstances would we provide a detailed description of the murder weapon? Most of the time, the answer to that question is when the reliable sources at least give us a hint that a detailed description of the murder weapon is relevant somehow. The sources in the Birch article apparently didn't indicate that any details about the sword were relevant (it's in a museum; the details are easily available), so we didn't include it. But if a source had written something like "Since the assassins used a single-edged sword, the murder took twice as long" or "Birch tried to dodge away from the blade, but the sword was so long that the end hit him anyway", then presumably we would have.

Why should the rules be different for a 21st century firearm vs a 19th century sword?WhatamIdoing (talk)20:15, 4 September 2025 (UTC)

Most of the time, the answer to that question is when the reliable sources at least give us a hint that a detailed description of the murder weapon is relevant somehow - Being mentioned by multiple sources in the context of of the event is what we need, not a special "detailed description". i.e. "we tend to mention things that other sources decide is worth mentioning" not "we tend to mention things only when other sources provide 'detailed commentary' about those things". That would be closer to the bar for us the include, well, a detailed commentary.
This section isn't about whether to provide a long explanation based on brief mentions. It's about including the name/type of gun at all. So yes, if in the course of covering a stabbing, multiple sources described a murder weapon as a spear, I would expect us to say it's a spear and not to censor that fact pending "detailed commentary" of the spear. —Rhododendrites^talk \\22:18, 4 September 2025 (UTC)

Being mentioned by multiple sources isn't sufficient. Think about it: Donald Trump wears red ties. Multiple sources mention this.[9][10][11][12][13][14][15][16][17]. But the articleDonald Trump does not mention this. Why not, if "Being mentioned by multiple sources in the context of a description of the event is what we need"?

I think that Wikipedia editors are meant to provide an encyclopedic summary, which does not necessarily include providing exact details for some things. That is, I believe we should say that the murder weapon was a sword (because it was a sword, not a spear), but not provide a detailed description of the murder weapon. Leave out the trivia; just say "three firearms" or "a handgun" or "anAR-15–style rifle". Don't say "Smith & Wesson M&P15 Sport III with laser sight, adjustable stock, and free-float handguard in black" – unless the sources indicate that there is something specific about one of those details that actually matters.WhatamIdoing (talk)22:42, 4 September 2025 (UTC)

Trump doesn't make a good analogy for much because of scale. I'd reply by just pointing up to my GPA, et al. examples above. We include a ton of detail based on it coming up in a variety of sources about the subject. The % of articles about a shooting that mention the type of gun is going to be an awful lot higher than the % of sources about Trump that mention his red tie.

I mostly agree withLeave out the trivia; just say "three firearms" or "a handgun" or "anAR-15–style rifle". Don't say "Smith & Wesson M&P15 Sport III with laser sight, adjustable stock, and free-float handguard in black" – unless the sources indicate that there is something specific about one of those details that actually matters - for that level of detail, yes, we need a reason to include it. That's not what this section is about -- at least not from the top -- which is about naming the brand/type. So I see theCharleston church shooting article as more or less the way to go:opened fire with a Glock 41 .45-caliber handgun. Don't know why we'd need more than that, but doesn't see a reason not to include that. —Rhododendrites^talk \\13:27, 5 September 2025 (UTC)

Why do we need more than "handgun"? What encyclopaedic information does the manufacturer or calibre convey? What (according to reliable sources) difference did it make to the event and/or outcome that it was a Glock handgun vs any other manufacturer's handgun?Thryduulf (talk)13:40, 5 September 2025 (UTC)

The main answer is that it's the same as any other little detail, like I said above. It's encyclopedic because we tend to defer "what's encyclopedic" to the prevalence of details in coverage of a story elsewhere, not because "Thryduulf and Rhododendrites say it's encyclopedic". But more directly, exactly what it communicates will vary. There's a lot that's communicated in saying that someone used a Beretta Model 93R vs. a Colt Single Action, which are both "handguns". The media often mention the type knowing about metacommentary about certain gun types and mass shootings. Mentioning it being an AK-47 puts the event in the context of other AK-47-related shootings. Certain models are commonly used in high-profile crimes and have become culturally or politically significant because of the broader context of these events, even if not mentioned in detail. Censoring such information when there's obviously relevance to the event just seems like advocacy to me. I don't know if I'm saying anything new, though, so I don't plan to follow up on this further. —Rhododendrites^talk \\14:23, 5 September 2025 (UTC)

Apparently, we have decided that theGlock 41 isn't a notable handgun model; editors have decided that it deserves a mere 41 words' worth of description in Wikipedia.

I asked Mr Google how manyGhits it would give me forCharleston church shooting -wiki -wikipedia. It said 2.73 million. I then asked how many ghits for the same thing, only with"Glock 41". It said 557. That means that only 1/5,000th of sources mention the model.

Looking only at news articles, the ratio is 68,500:9 – only 1/7,600th of news articles mention the model.

Asking it to compare news sources mentioning "Glock" but not "Glock 41" and vice versa was interesting; 810 mention the company without the model number, and 271 mention the model number but don't mention the company separately.

That doesn't make me think that we've made the right choice in this article. Most sources don't mention either the brand or the model, so why are we mentioning both? Most sources that mention either the brand or the model mention only the brand and not the model, so why are we making the less common choice? That doesn't really look like "following the sources" to me.WhatamIdoing (talk)22:34, 5 September 2025 (UTC)

• This seems like a solution in search of a problem to me, in a way that fundamentally comes down to "I don't like it."⇒SWATJester^{Shoot Blues, Tell VileRat!}22:29, 4 September 2025 (UTC)

• Doesn't the firearm brand and model fall under 'trivia' in some way? Unless the brand/model is widely reported in RS, it seems the same to me as including the color or brand of the perp's shirt or shoes, etc.Some1 (talk)18:08, 5 September 2025 (UTC)

  I agree, but I think that the difficulty is in whether "widely reported" means "mentioned in a large fraction of sources" (10 mention it, out of 30 total sources) vs "mentioned in a tiny fraction of a very large number of sources" (10 mention it, out of 300 total sources).WhatamIdoing (talk)22:36, 5 September 2025 (UTC)

• I agree that this information is generally undue. Unless the type of firearm is specifically notable in and of itself (i.e. calls to restrict the usage of said firearm, orWP:SIGCOVspecifically about the firearm type), this is undue excessive information for the average reader.WP:TECHNICAL applies here, aside from any concerns over copycat shootings. Thisshould be covered byWP:DUE (policy) andWP:FANCRUFT (essay), but having something spelling this out in full will prevent editiors from adding this minutiae to an unfortunately commonplace and ever-growing category of articles without consensus.GnocchiFan (talk)10:02, 7 September 2025 (UTC)
• I was one of the editors involved in some of the firearms/crime inclusion/exclusion debates from about 10 years back. I think it's interesting that some argue that pro-firearms people want to include make/models in the articles about the crimes. That was never my impression but that may indicate just how big the gap in perception was/is. We are all looking at the same picture yet seeing something that is so different. Certainly a good argument for AGF! My concern then was the feeling that an articleabout the firearm was often becoming an article about crimes. This is around the time I challenged the idea that WEIGHT has reciprocity. That is if A has weight in context of subject B, does B have weight for inclusion in the article about A? My view is no and at least one well attended RfC I was involved with supported that view. As for including model name/number in an article about the crime, my feeling is that should be decided by weight in sources about the crime. In general, I agree with those who feel such details aren't, in most cases, encyclopedic and inclusion shouldn't be assumed. I'm sure there are cases where more detail is warranted and I'm sure RS can help us establish weight for inclusion. I would also suggest that we shouldn't have any blanket MOS on this content. I don't think the MOS should, as a general rule, govern article content outside of perhaps the intro material in things like BLP. How such content is displayed, sure, but not what is included/excluded and certainly the deeper into an article the less the generic MOS guidelines would apply. That said, absent a specific source review/details, I would generally favor leaving make/model out of articles about the crime.Springee (talk)13:32, 7 September 2025 (UTC)

  We are all looking at the same picture yet seeing something that is so different. This is unusual? Otherwise, good comment. ―Mandruss ☎ IMO.14:58, 7 September 2025 (UTC)

• The answer is "it depends on the case" and there is no use for hard rules here. I think it is useful to include, personally, provided reliable sources say it. Gun brand and model are often politicized and gun companies have often been sued for alleged responsibility, which generally invites curiosity about that. If it is not clearly said by any source obviously don't extrapolate past that into original research.PARAKANYAA (talk)06:35, 10 September 2025 (UTC)

Moved fromWikipedia:Administrators' noticeboard § AI Moderator proposal

 –voorts (talk/contributions) 12:47, 5 September 2025 (UTC)

On Reddit, there is a very useful extension called AI Moderator (AI Moderator | Reddit for Developers). It allows users to configure rules (eg "remove the post if it is spam") and the AI will evaluate the comment/post to see if it meets the criterion. I am thinking of bringing a similar system here, with the following notes:

• This is distinct from the WMF-developedmediawikiwiki:Moderator_Tools/Automoderator and ClueBot in a key way - the former is binary "revert/don't revert", while this proposal allows authorised wiki users to set up custom rules that need not be pure "vandalism". I think this can be very useful for some of the chronic LTAs that tend to obfuscate whatever they want to say.
• On the Reddit system, only Gemini/OpenAI are supported. The proposed system will only use open-source AI (eg the 1b/4b Gemma models, to keep speed reasonable) running on Cloud VPS. I am expecting up to 60 seconds' delay before the system reverts a flagged edit, but this is subject to change.
• This proposal complements and does not replace the current edit filter system.

I'm hence requesting feedback from the community on whether this is something they'd like to have on this wiki, or whether I've missed something (eg someone has already developed a similar tool). Please ping me in a response.Leaderboard (talk)08:53, 5 September 2025 (UTC)

No, we don't want AI bullshit-bots 'moderating' anything around here. We have enough problems with their hallucinatory garbage already.AndyTheGrump (talk)09:02, 5 September 2025 (UTC)

@Leaderboard What problem faced by the English Wikipedia would this solve? I don't see the need for this; our talk pages are generally quite civil, and when they are not, the best solution is usually warning the offender, not removing the post outright.Toadspike[Talk]09:21, 5 September 2025 (UTC)

@Toadspike A few that come to mind:

• LTAs that obfuscate or otherwise exhibit behaviour that is difficult to handle using edit filters (eg MidAtlanticBaby). AI tends to be significantly better from my experience, and this should reduce the need to constantly update the filter and should also reduce false positives.
• Complex requests that cannot be easily handled by a filter or where a high risk of false positives exist. For example, determining the intent of a comment or edit to a page, or verifying a source for an article that a user added.
• As Tamzin said, it can be useful as a first-stage log.

Leaderboard (talk)09:45, 5 September 2025 (UTC)

While there's obvious pitfalls, I could see some use in integrating an LLM into AbuseFilter for log-only filters. A filter that can track, for instance, "Did this edit change a date of birth without changing/adding a citation?" or "Does this edit look like it's trying to convey the same message as some recurring LTA rant?" would be useful. Anything "AI moderator"-like, though, is a complete nonstarter. The actual moderation decisions need to be made by humans. --Tamzin^{[cetacean needed]}(they|xe|🤷)09:30, 5 September 2025 (UTC)

I agree to this. There is no way we get the consensus for AI actioning on stuff, but it works be useful if does some work of scouting for bad edits and notifying volunteers (interested ones) through either a separate platform on toolforge or integrated with edit filters.~/Bunnypranav:<ping>09:40, 5 September 2025 (UTC)

My perspective is different, especially given that the wiki is already making use of (or was - not sure) similar automated tools, such as ClueBot. Having tested this with multiple subreddits (not the same as this wiki, but still), I can say that AIModerator should be effective as a first-stage defense for many types of common undesirable edits. Yes, there can be false positives, but that's true with an edit filter or ClueBot as well.

And also: the same development process as for an edit filter applies here - eg first by logging the edits, then warning the user, and then disallowing. So there's enough opportunity to test the effectiveness of the filter.Leaderboard (talk)09:49, 5 September 2025 (UTC)

I should add that this proposal is unquestionably misplaced on this noticeboard: if AI is ever to be introduced for any function similar to that proposed, it will require theassent of the broader community, after first (a) having clarification as to what problem it is supposed to be solving, (b) seeing evidence that it is capable of solving (or ameliorating) said problem, and (c) an appropriate period for discussion.AndyTheGrump (talk)09:46, 5 September 2025 (UTC)

There are similar initiatives (eg.Wikipedia:Edit check/Tone Check), but AndyTheGrump is right that this is the wrong location, and I would further add that any proposals should be much more specific.CMD (talk)09:48, 5 September 2025 (UTC)

@Chipmunkdavis and @AndyTheGrump Apologies, can you then move this to the right place? I was under the impression I had to solicit feedback herefrom past experience.Leaderboard (talk)09:51, 5 September 2025 (UTC)

Based on how I am reading it, VPP and AN were the two best options for your Leaderbot proposal regarding temp rights. This case is different as it doesn't really deal with anything that is related to administrators in general.

(While I am commenting, you might want to reviewWP:AI since it covers past AI proposals and more.) --Super Goku V (talk)10:43, 5 September 2025 (UTC)

@Super Goku V "This case is different as it doesn't really deal with anything that is related to administrators in general" - actually it does as admins (and other trusted users to be defined) are the ones going to be configuring the system.Leaderboard (talk)11:00, 5 September 2025 (UTC)

No they aren't. Not withoutprior community approval of any such addition to admin (or other 'trusted users') rights/responsibilities. Leaderboard, I'm going to be blunt here. You have only made 500-odd edits to Wikipedia, less than half of which are in article space. You clearly have little idea as to this place works. You appear to be making a proposal of which you have yet to either define properly nor demonstrate an actual need. You are now apparently telling us who exactly gets to 'configure' this AI-bot. That isn't how it works. In my opinion, you would do well to drop this proposal, along with any other bot proposals, until you actually have the necessary experience as a normal editor to understand what we actually need, and how to appropriately ask for approval.AndyTheGrump (talk)11:17, 5 September 2025 (UTC)

@AndyTheGrump Are you serious? I may not be familiar with en.Wikipedia policies, but I'm a fairly experienced global editor (check my CentralAuth), admin on multiple wikis, and also have an approved bot on this wiki. I also regularly work with wikis where I have little to no experience. Just because I come up with an idea does not mean that it's going to happen or that I am "unqualified". After all, I've been clear from the beginning that this is a proposal, and things like these, especially when they involve significant development work, absolutely do not need to be "properly defined" at this stage. Your criticism would be better placed if I were to just go toWP:BAG and ask for bot rights. P.S: other users, like Anomie, have provided appropriate criticism and suggestions without feeling like I'm being insinuated, and that's what I'm looking for.Leaderboard (talk)13:39, 5 September 2025 (UTC)

Just confirming,Leaderboard hasover 17,000 edits distributed across dozens of wikis (but mostly from their top half dozen) over an eleven year period, and has sysop rights on multiple projects.BD2412T18:52, 5 September 2025 (UTC)

@Leaderboard: The strong negative reaction your proposal is receiving here (and the feelings of insinuation) might be explained by the fact that the English Wikipedia appears to be the Wikimedia project that is the most vocally critical and skeptical of generative AI and LLM usage. To give an example,WikiProject AI Cleanup, a WikiProject I am a member of that aims to find and clean up unreviewed LLM-generated content, only exists in two languages: English and French. The English WikiProject has173 participants, but its French counterpartfr:Projet:Observatoire des IA only has 17. (Both figures are as of writing this comment.)SuperPianoMan9167 (talk)22:33, 5 September 2025 (UTC)

If you didn't know, enwiki also has a recently-createdspeedy deletion criterion,G15, that specifically applies to unreviewed LLM-generated content. Many proposals about AI-powered features on enwiki have received considerable pushback; there's a list of current and past discussions that you can readhere if you're interested.SuperPianoMan9167 (talk)22:43, 5 September 2025 (UTC)

@SuperPianoMan9167 And that is OK. Hence why this discussion - I also plan to do this with other wikis to see what their stance on this is.Leaderboard (talk)04:24, 6 September 2025 (UTC)

@Chipmunkdavis this is a very early-stage proposal so the lack of specifics is natural. We'll get there as I get feedback - feel free to ask any clarifying questions in the meantime.Leaderboard (talk)09:52, 5 September 2025 (UTC)

Wikipedia:Village pump (idea lab) is probably a better place for this discussion. — rsjaffe 🗣️09:56, 5 September 2025 (UTC)

I agree, this noticeboard is more for recent disruptions on the project.98.235.155.81 (talk)10:03, 5 September 2025 (UTC)

Tone check does not use a LLM. The term "AI" is heavily overloaded.* Pppery *_{it has begun...}16:13, 5 September 2025 (UTC)

The tone check page claims it uses a SLM, which doesn't seem much of an overload from LLMs at all.CMD (talk)00:22, 6 September 2025 (UTC)

Regarding Tone Check, I wonder if the "AI Moderator" can review the tone of editors' talk page comments and give them a "friendly reminder" to be more civil if their comments are deemed overly aggressive or unfriendly. Not saying it's a good idea, but it's an idea.Some1 (talk)18:21, 5 September 2025 (UTC)

I find the idea of an "AI moderator" policing the tone of human beings to be intrinsically offensive.Stepwise Continuous Dysfunction (talk)23:25, 5 September 2025 (UTC)

I am deeply against the idea of such a chatbot "moderator". Also why is this posted in this forum at all?Simonm223 (talk)11:22, 5 September 2025 (UTC)

@Simonm223 your question is being discussed directly above.Thryduulf (talk)11:35, 5 September 2025 (UTC)

Absolutely not. This would just create massive amounts of problems and not solve anything. We already have more than enough AI nonsense on this site as it is.2A0E:1D47:9085:D200:EF8D:3F1F:719E:C21C (talk)11:37, 5 September 2025 (UTC)

As seen above, I doubt there will be much acceptance at all of a new LLM-based AI actually reverting things.User:ClueBot NG does use machine learning, but it's a well-established and carefully developed system specific to Wikipedia maintained by people who have been active in the community for a long time, not a "LLMs are the new Blockchain, let's use them in everything" idea. You might do better to look into a classification system that humans can use as input, similar tomw:ORES and the successor that's being worked on, rather than trying to jump straight to an LLM-based bot.Anomie⚔12:33, 5 September 2025 (UTC)

It's worth pointing out that some smaller wikis do usemw:Extension:AutoModerator which works using the ORES system.* Pppery *_{it has begun...}16:13, 5 September 2025 (UTC)

The only response this deserves isTrout_signed,Rosguill^talk16:17, 5 September 2025 (UTC)

I utterly disagree, based on the actually sensible discussions occurring in this section of what sort of parameters we could use as guardrails for such a tool.BD2412T18:41, 5 September 2025 (UTC)

To the extent that this is a proposal to use a LLM to populate a list of potentially problematic edits for a human sysop to review, I think it's a really good idea and I hope you go ahead. We might eventually get to the stage where human sysops experienced with the AI's output are prepared to recommend a trial of a bot to run it (or to exapt ClueBot NG to do so).—S Marshall T/C16:54, 5 September 2025 (UTC)

A simple question: why would we need a separate bot for this anyway? Wouldn't it make more sense (if it is actually feasible, solves a real problem, and is accepted by the community) to add such AI-driven functionality to ClueBot NG?AndyTheGrump (talk)17:06, 5 September 2025 (UTC)

Same as why I am not recommending that such functionality be added to the AutoModerator project by WMF. That is, I see them as complementary - handling related but distinct tasks.Leaderboard (talk)18:08, 5 September 2025 (UTC)

ClueBot NG is a trained neural classifier, and its codebase appears stable and largely unchanged in a while. I'm not even sure if it's regularly retrained. LLM-based AI should probably be a different bot altogether, as it's sufficiently different technology. It definitely shouldn't run under the same bot account.ProcrastinatingReader (talk)10:25, 6 September 2025 (UTC)

@Leaderboard, what if you run a trial of the tool where no edits are made by the bot and its flagged changes are examined?JayCubby18:11, 5 September 2025 (UTC)

@JayCubby At a later stage. This will only be once the specification has been designed (after getting feedback from different communities, like what we're having now) and I develop a working POC (proof-of-concept).Leaderboard (talk)18:14, 5 September 2025 (UTC)

I see no reason why this would be trustworthy, and three or four reasons why it wouldn't be ethical. Even using this to create a list of edits for humans to review would bias our moderation priorities towards what the slop machine says is important. That's not a burden we need.Stepwise Continuous Dysfunction (talk)23:24, 5 September 2025 (UTC)

No, absolutely not. SCD gives a good reason. Another is simple: we're busily fighting off AI being used to edit articles.Even if this could be ethical and useful,giving AI promoters a "butyou use AI!" argument to use is not something we need to do. -The Bushranger_{One ping only}00:13, 6 September 2025 (UTC)

I find that an absurd line of reasoning. We use bots to revert spambots. It does not give spambot promoters a "butyou use bots!" argument. –SD0001 (talk)11:07, 7 September 2025 (UTC)

I think the first mistake was to use the word "moderator", especially in conjunction with the word "AI". By doing so, you're going to get kneejerk opposition. (I was expecting something far more dubious when I saw the section title.) To be honest, a variant of this idea might actually be a useful tool in combating vandalism. If it worked well, it would help us better use anti-vandalism editor labour. Technically, I'd imagine it'd need some finetuning, and for some LTAs the nuance may be too specific or the false positive rates unacceptable, but so long as it can distinguish well and the precision is sufficiently high, then it's hard to argue against adopting this technology, except perhaps on a cost basis, but those would be considerations for the Cloud VPS teams.

If you wanted to pursue it, if I were you I'd coordinate with anti-vandalism editors and/or SPI to figure out where it'd be most useful, with Cloud VPS to figure out if you have the GPU compute needed to run open source models for inference, and probably editors interested in AI/ML with some free time if you needed technical assistance on model building. VPI is probably the wrong audience for brainstorming this.ProcrastinatingReader (talk)10:25, 6 September 2025 (UTC)

My Cloud VPS allocation right now is 8 cores/32 GB RAM, which should be enough for smaller models. Things like GPU can come later, but @ProcrastinatingReader I'd appreciate if you could crosspost or otherwise notify users that are likely to be interested in this.Leaderboard (talk)11:04, 6 September 2025 (UTC)

Most LTAs trying to game edit filters end up getting reverted almost instantaneously for vandalism/disruption even if they aren't identified as socks immediately. In any case the tool should not be making the reverts by itself. I could see something on the lines of what Tamzin proposes for more complicated conditions that can't be straightforwardly handled by regex; but I'm not sure that ends with a manageably-sized list of problematic edits. My thinking tends a bit towardsWP:BROKE for now, but I doubt a trial would cause any harm either. For all I know the list of flagged edits could end up being manageably small after all.JavaHurricane16:24, 6 September 2025 (UTC)

Strong support an easily reversible 60-day trial. Work out the details of said trial (config details etc) amongst yourselves.Strong oppose crystal ball reasoning, on principle. Those balls are unreliable and overrated, basically quackery.Strong abstain (lol) if trial would not be easily reversible. ―Mandruss ☎ IMO.16:51, 6 September 2025 (UTC)

I see I'm the only one using bolded !vote format, suggesting it's improper for VPI. Not sure what to do with that. I'm just trying to voice my strong support for trials over crystal, which applies to any idea that's somewhat viable. There is way (did I mention way?) too much risk-aversion at en-wiki for the encyclopedia's good. Most reliable way to determine whether something will be a net improvement:Try It And See™ (TIAS).

Consider if you will a fictional big pharma company. Their scientists have created a new medication and the company needs to decide whether to put it on the market, ie release it for sale. It hasn't undergone any trials, since the company has twenty really smart, experienced people who are certainly capable of deciding among themselves whether it's safe. Trials would be an unnecessary waste of time. Today, those twenty are seated at a conference table, discussing the question. Is this wise?

No.

So, what, en-wiki is different because we don't develop drugs that could maim or kill people? I truly don't see why that would make a difference. In reality, the company knows they can't really know until they...wait for it...Try It And See™. They know this because they have twenty really smart, experienced people. So they schedule the TIAS to begin in 14 days and last for 60 days. Then they adjourn the meeting. ―Mandruss ☎ IMO.17:18, 6 September 2025 (UTC)

Where would you suggest that discussion take place instead of VPI?WP:LLMN?WT:EF? Or some other noticeboard or talk page?SuperPianoMan9167 (talk)19:24, 6 September 2025 (UTC)

1. I wouldn't know.2. When did I say there should be a "that discussion"? lol. 3. I laid down my argument and others are free to pick up that ball and run with it. I'm happily semi-retired so I would be a no-show. ―Mandruss ☎ IMO.19:32, 6 September 2025 (UTC) Edited per discussion. 20:12, 6 September 2025 (UTC)

I was interpreting "that" as arelative pronoun ("I knowthat I like books) instead of ademonstrative ("I likethat book"); I was talking about further discussion that would take place developing this idea. I probably could have worded it better :)

The more I learn about other languages, the more confusing and nonsensical my native language (English) seems.SuperPianoMan9167 (talk)19:44, 6 September 2025 (UTC)

LOL. That's crazy. I've never seen that happen in my 69 years. No, don't blame yourself even a little. English is a crazy language, far below the state of the art, and needs to be replaced by Esperanto. Is this the right place to propose that? :D Ok, so I'll fix that. ―Mandruss ☎ IMO.20:12, 6 September 2025 (UTC)

This sounds interesting, but we should first figure out what the parameters (and especially the success criteria) are for the experiment. If we start a trial, and have a high rate of spam getting reverted, but also some amount of good-faith comments getting reverted, then should we accept it as a success? Invariably, this is – to some extent – what we will see: some true positives, some false positives, and some false negatives.

The other difference with your big pharma example is that trials usually aren't done on the whole live population. Having a smaller test case on which we could run the trial for 60 days could be much better, avoiding the risk to either Wikipedia's reputation or editor retention.

For now, we don't have all of that. A controlled trial isn't just "let's run the proposal for 60 days", especially since we don't have a single proposal – do we directly import the "AI Moderator" tool with an open-source model? Do we fine-tune it, give it specific prompts? Do we train smaller models on specific cases (as was suggested below)? All of this has to be known in order to be able to Try It And See.ChaoticEnby (talk ·contribs)13:22, 7 September 2025 (UTC)

Here's my thoughts: this feature should probably be tested as a client-side, on-demand feature only, perhaps as agadget or user script that makes API calls to an LLM. The advantage of this approach is that those users who are interested in this kind of tool and would like to test it are able to do so, while those users who oppose the tool are not forced to use it. It's best tonot set up stuff like this server-side without obtaining broad community consensus. I do see the merits of this proposal for identifying sneaky LTAs that are able to figure out how to bypass regex in filters, although like current filters, this feature would definitely have some false positives. If this proposal moves forward, I would recommend developing it more as a way to "improve abuse filter detection with machine learning" rather than act as an "AI moderator", as calling it an "AI moderator" may drive away users who would otherwise support the idea. As already mentioned, Wikipedia already uses some machine learning for anti-abuse workflows, most prominently in ClueBotNG.SuperPianoMan9167 (talk)19:24, 6 September 2025 (UTC)

I believe an LLM is too heavy for combating LTAs. I think we can get away with a specialized model such as an LSTM or GRU model or a small transformer. Each LTA would have their own model trained on known-LTA diffs and known-not-LTA diffs, making the model a classifier. This would be better than using an LLM to distinguish e.g.Master12112wp and having it either hallucinate AGF on every single LTA edit (making it effectively useless) orWP:ABFing by classifying every piece of Hindi text as LTA (relevant lemmy post). LLMs are optimized towards the general task of predicting a plausibly-human token given previous tokens, but we can do better for this specialized task. (On the other hand, LLMs may be useful for other tasks, such as spotting NOTFORUM comments on controversial talk pages.)OutsideNormality (talk)21:59, 6 September 2025 (UTC)

I don't think having this kind of "automoderator" determining whether posts are spam or not would be helpful, as it would essentially be a moreWP:BITEy version of ClueBotNG. The latter is already only configured to revert the most obvious kinds of vandalism, and a LLM trained on a much wider definition of spam from other websites (and with the risk of errors) would be much worse, and could lead to systemic bias depending on which posts it considers "spam" or not.

However, I do agree with the idea of training smaller models against specific LTAs, as an addition to existing abuse filters – they can be more flexible in learning complex rules than regex, and can be trained for a more specific task than a generic LLM, instead of flagging any text they don't recognize as plausible. I don't think this should be extended to NOTFORUM comments, as human judgement is often key in these cases, and the risk of "grey area" comments being reverted by a model is too high.ChaoticEnby (talk ·contribs)13:13, 7 September 2025 (UTC)

The NOTFORUM comments example was just an example off the top of my head of a task that would be better done with a large language model than a small language model.OutsideNormality (talk)15:25, 7 September 2025 (UTC)

My reply about NOTFORUM was regarding language models in general, not specifically small/large.ChaoticEnby (talk ·contribs)00:34, 8 September 2025 (UTC)

Enough of shoving AI down our throats. Plus, I don't consent to any of my edits being used to train this - even if I'm in somewhat of a good standing somehow.LilianaUwU^{(talk /contributions)}20:28, 7 September 2025 (UTC)

Enough of shoving AI down our throats. Oh Liliana. Such language. This is not a group of powerful elites talking about what to do to thehuddled masses yearning to breathe free. At least, not any more than any village pump discussion. Wikipedia isself-governed.self-governed. If "your throats" don't participate in the decision-making, they have no right to complain about the outcome. None. We'll take your comment as opposing the idea, although you didn't say why, sort of making it aWP:IDLI argument. ―Mandruss ☎ IMO.20:57, 7 September 2025 (UTC)Edited after reply 02:40, 9 September 2025 (UTC)

People have legit concerns about AIs/LLMs. (For example, there is about a half-dozen comments in this discussion that touch on false positives.) I don't think this is going to reduce those concerns. --Super Goku V (talk)07:46, 8 September 2025 (UTC)

The original proposal uses a pre-trained model (1b/4b Gemma 3), so none of your edits would be used to train it(as they've already been trained on¯\_(ツ)_/¯). My small language model proposal abovewould use Wikipedia diffs as training, however. If you oppose eventhat, I'm curious as to why (and note that ClueBot NG already does a similar thing, just using a hybrid of a Bayesian classifier and feedforward neural network instead of anRNN like my proposal).OutsideNormality (talk)21:37, 7 September 2025 (UTC)

The use-a-specialized-langugae-model-as-the-next-step-of-protection-against-abuse-filter seems to be a good idea and perhaps we can take this to its own subsection. The community seems to be more on-board with this from what I have seen. You can never have enough AI.--85.98.23.90 (talk)20:52, 7 September 2025 (UTC)

To elaborate on the previous discussion, the idea of using specially trained smaller language model sounds more promising. Edit filters (being regex-based) can't capture more complex language patterns that are often useful in spotting LTAs, and training small language models for very specific tasks like this is absolutely feasible. This shouldn't be used to build a generic spam filter (given the high risk of false positives), but to perform specific tasks that edit filters can't necessarily achieve on their own.ChaoticEnby (talk ·contribs)00:45, 8 September 2025 (UTC)

I think this is a far more reasonable idea, like a much more targeted CluebotNG, and is worth exploring. There is less of a systemic risk of introduced bias at scale.fifteen thousand two hundred twenty four (talk)01:48, 8 September 2025 (UTC)

Wouldn't that be easy for an LTA to work around, just by adjusting their message in a way that goes out of the "pre-trained" dataset? I thought that this would be less likely with a full-fledged LLM.Leaderboard (talk)06:55, 8 September 2025 (UTC)

Probably depends on the issue at hand. I know one LTA who uses various IP addresses to simply delete mentions of certain politicians. That sort of thing is basically impossible to stop with current methods.CMD (talk)07:03, 8 September 2025 (UTC)

If the LTA changes their method of operation, then there isn't much that can be done in terms of automated recognition, beyond a generic "spam filter" that might have a much higher risk of false positives. On the other side of the spectrum, slightly adjusting an existing message would vary in effect depending on the specific model's ability to infer broader patterns, which is something we should be testing empirically (and for which the inference power of a LLM is likely unnecessary). A full LLM would also have to be fine-tuned for each new LTA, which would add a higher training cost to the already higher execution costs.ChaoticEnby (talk ·contribs)13:27, 8 September 2025 (UTC)

I've created a basic demo using the gpt-5 nano API, where the model has reviewed thirty edits for about two cents (would be even cheaper with batching). You can view the results atUser:Ca/Automated RCP. Even with a basic non-specialized set up like this, I think the results, while not perfect, are impressive. I think it would be useful for double-checking the RC after a day of delay after human RC patrollers have checked the frontline queue.Ca^{talk to me!}09:50, 10 September 2025 (UTC)

It's possibly the prompt could be refined, but I suspect this is far too broad, generating a lot of false positive and false negatives. (And I certainly did not expect it to twig "Los Angeles" as a typo, that feels like it should be a well-known word in any large training data.) More discrete prompts would however be more expensive, so I am not sure how this would scale.CMD (talk)11:25, 10 September 2025 (UTC)

Thanks for reviewing the output. The falsities identified can probably be mitigated by providing more context to the model. Currently only the modified line itself is fed into the model. Provision of surrounding lines, titles, and the edit summary are currently not implemented. This would fix false positives 1310564969 and false negatives 1310564324 and 1310565149. But I agree that the "Los Angeles" typo example is egregious and shows how LLMs cannot 100% replace human review. I did attempt a more sophisticated prompt, but the GPT-5 Nano doesn't seem to handle it very well.Ca^{talk to me!}12:12, 10 September 2025 (UTC)

I was readingWP:ANI and saw that an editor was accused of archiving topics on the talk page of an article too soon, to close discussion and avoid changes to an article. I don't see why editors should be archiving article talk pages at all. This is a thing that bots can do better: less wasted time by humans, no question of neutrality, less watchlist clutter. We should just establish a rule that above a certain number of topics (or the size) on a talk page, the page is marked for archiving by a bot. The bot archives the earliest topics but leaves topics where the last post by an editor wasn't visible for a certain time (2 weeks or a month?, maybe change this depending on how many topics/size there are on the talk page).Rolluik (talk)08:48, 8 September 2025 (UTC)

MostOne click archiving performed by humans is completely uncontroversial. Sometimes bots miss topics to archive, likethisinstance. Also, here's alink to the thread in question, in which I was mentioned.Graham87 (talk)10:28, 8 September 2025 (UTC)

I agree that most archiving by humans is uncontroversial, I just think a bot archiving would be even less controversial. If a bot doesn't archive a topic that it normally would archive, there is something wrong with either the bot (and then the bot should be fixed) or more likely with the talk page and then the talk page could be fixed by any editor and later archived by bot.Rolluik (talk)12:14, 8 September 2025 (UTC)

Who sets up the bots?Phil Bridger (talk)11:30, 8 September 2025 (UTC)

Often people involved in editing the page, but sometimes passers-by.Graham87 (talk)11:37, 8 September 2025 (UTC)

I meant the question semi-rhetorically. The answer is of course editors, who would be forbidden from archiving under this proposal.Phil Bridger (talk)11:45, 8 September 2025 (UTC)

The editor who programs the bot and gets approval atWP:BRFA.Rolluik (talk)12:28, 8 September 2025 (UTC)

This is a solution in search of a problem. There's plenty of reasons why someone would want to archive a thread before the bots get to it. On super active pages, it's often helpful to archive resolved threads as soon as they are resolved, so the talk page can be used to discuss non-resolved issues.

If someone is archiving threads before resolution and while active discussion is happening, the problem is with that person, not with archiving in general.Headbomb {t ·c ·p ·b}11:43, 8 September 2025 (UTC)

And that falls underWP:PREMATUREARCHIVE (although this is not the most descriptive guideline section).CMD (talk)12:22, 8 September 2025 (UTC)

I have to agree. I recognize that the OP started this discussion in good-faith, and I am honestly not intending to be disrespectful or demoralizing in response, but this is really a proposal that could have only come from a fairly inexperienced editor: there are just so many reasons why a blanket ban on manual archiving would be somewhere between generally undesirable and outright untenable, across use cases. Incompetent or bad faith archiving can be dealt with through the normal behavioural channels, and meanwhile, requiring bots would not even prevent these ills entirely, as it would be trivial to create a bot to do an end run around such restrictions for intentional but problematic editing, and unintentional issues with archiving bots can and do occur from time to time. Again, meaning no offense to the OP, but the cost-benefit of the proposed change would be a ridiculously poor ratio.SnowRise ^{let's rap}14:12, 8 September 2025 (UTC)

Bots actually can't do it better. They don't check for vandalism to the talk page and just blindly archive whatever, and if that vandalism gets archived people will yell at you forevermore if you try to then revert the vandalism that would have been 100% acceptable to revert five seconds ago. Bots also tend to jump the gun and archive topics almost immediately -- as in, I've seen it happen literally minutes after something is posted.

As such, we do not need to encourageeven more botshit.Gnomingstuff (talk)17:04, 8 September 2025 (UTC)

It seems to me that the archiving of discussions while they are still active is a problem with the individual editor. We already have avenues to deal with them. — Jkudlick ⚓ (talk)15:11, 11 September 2025 (UTC)

OneClickArchiver is an essential part of the operation ofTalk:Donald Trump. Take that away, and the locals will be revolting (many say we're alreadyrevolting). ―Mandruss ☎ IMO.15:18, 11 September 2025 (UTC)

TheBiodiversity Heritage Library is very widely used here and on other projects and is an invaluable reference, but itis currently in a bit of trouble and looking for "partnership opportunities to support its operational functions and technical infrastructure" after the Smithsonian Institution opted to "conclude its long-standing role as BHL’s host on 1 January 2026". Is the WMF able to help in any way?Cremastra (Go Oilers!)23:58, 12 June 2025 (UTC)

The people who can answer that question are unlikely to see it here. Somewhere on Meta is probably your best bet, but I don't know off the top of my head where.Thryduulf (talk)00:50, 13 June 2025 (UTC)

I like the idea! I've emailed answers@ for guidance on where the best place for this suggestion would be.HouseBlaster (talk • he/they)18:09, 15 June 2025 (UTC)

Thanks!Cremastra (Go Oilers!)18:36, 15 June 2025 (UTC)

I heard back a couple days ago; they said they would raise it internally and get back to us. I'll let everyone know when I have further updates :)HouseBlaster (talk • he/they)23:22, 21 June 2025 (UTC)

Further update: the relevant WMFer is out of office for a minute, so we are still waiting to hear back.HouseBlaster (talk • he/they)02:03, 2 July 2025 (UTC)

@Cremastra @HouseBlaster this is a great idea, I use the BHL all of the time writing species articles. Without it a lot of good information would be lost and completely inaccessible, especially for obscure species where much of the available information is in the original paper on them, which the BHL often preserves.Mrfoogles (talk)22:37, 21 June 2025 (UTC)

Dear @Cremastra, @Thryduulf, @HouseBlaster, @Mrfoogles,

Thanks so much for your message here. I apologise for the wait in answering you.

To answer your question, the Wikimedia Foundation has been collaborating closely with the Biodiversity Heritage Library. Last year, it was a partner for two hypotheses within theClosing Knowledge Gaps goal of the annual plan:

1. WE2.3.3 BHL-Wikimedia Working Group:“If the BHL-Wikimedia Working Group creates Commons categories and descriptive guidelines for the South American and/or African species depicted in publications, they will make 3,000 images more accessible to biodiversity communities.”
2. WE2.7.2 Biodiversity community collaboration:“If 3,000 well-described images of South American and/or African species are released to the wider biodiversity community through 2-3 editing events and an on-wiki worklist, 300 new images will be utilized on Spanish, French, and Portuguese Wikimedia projects.”

Previously, we contributed to the whitepaper,“Unifying Biodiversity Knowledge to Support Life on a Sustainable Planet”, and suggested the creation of a BHL Wikimedia Working Group. You can learn more about their activities inthis slide presentation by @Ambrosia10 and the Diff post"Past, present and future: a Wikimedian-in-Residence at the Biodiversity Heritage Library”, by @TiagoLubiana.

Unfortunately, the BHL Wikimedia Working Group paused its activities due to theBHL reorganization.

The Wikimedia Foundation doesn’t have a funding program targeted to allied organizations but the recently-published paper,Towards a Healthy Ecosystem of Wikimedia Organizations, asks whether they should be recognized and eligible for funding. In the meantime, Wikimedia volunteers and communities can apply for a Rapid Grant or a General Support Fund.

I understand thatWikiEdu Foundation has been in contact with the Biodiversity Heritage Library to explore the potential for collaboration.

With all that, I can say that the BHL is not alone in its Wikimedia-related work and journey.

Thanks for the opportunity to shine a light on the great work BHL has been doing on the Wikimedia projects!GFontenelle (WMF) (talk)15:06, 12 September 2025 (UTC)

I see, thank you. Is there any way the WMF could make a one-time donation to the BHL, or is that red-taped? I presume there are strong restrictions on what you can do with donor money, restrictions I fully support.Cremastra (talk ·contribs)15:55, 12 September 2025 (UTC)

It's not especially difficult from a legal sense, since BHL has found afiscal sponsor inCouncil on Library and Information Resources. What they need is a technical sponsor (someone who can handle all the computers).WhatamIdoing (talk)17:57, 12 September 2025 (UTC)

Link of specific discussion topics on a talk page can be copied easily on a PC by hovering the cursor above it.

But I've not been able to do so using a smartphone. Is there any way to do it, or should a new feature be implemented in this regard?Cdr. Erwin Smith (talk)16:40, 11 September 2025 (UTC)

If you click on the timestamp for any comment on a talk page, a link to that comment is copied to the clipboard. You can click on the timestamp for the first comment in a section as a workaround. (Perhaps you have some gadget enabled that makes copying easy by hovering? For me, on a non-mobile device I use my own user script tofacilitate copying links to headings or comments to the clipboard.)isaacl (talk)18:00, 11 September 2025 (UTC)

You are suggesting to copy the diff of the first comment in a discussion to refer to the whole discussion, which is indeed clever.

However, just like me, not everyone might have this idea. It's also not a proper fix. I would suggest adding a small⧉ (copy symbol) beside the header of each discussion for quick access to its URL on a smartphone.Cdr. Erwin Smith (talk)18:22, 11 September 2025 (UTC)

@Waddie96 could you kindly look into this ?Cdr. Erwin Smith (talk)07:16, 12 September 2025 (UTC)

@Cdr. Erwin Smith Good day, and welcome to Wikipedia! I'm unsure why you've tagged me?

Anyway, check outHelp:Links for a starting guide on links. But to link to a specific discussion on a talk page, use the base pagename, i.e.Wikipedia:Village pump (idea lab) for example, and then the section heading title, i.e.Talk Page Discussion Link for Mobile Users together with a# to show the section heading, like so:[[Wikipedia:Village pump (idea lab)#Talk Page Discussion Link for Mobile Users]] for this chat. It will look like this:Wikipedia:Village pump (idea lab)#Talk Page Discussion Link for Mobile Users.

A neater way is to use{{Section link}}:{{section link|Wikipedia:Village pump (idea lab)#Talk Page Discussion Link for Mobile Users}} to make:Wikipedia:Village pump (idea lab) § Talk Page Discussion Link for Mobile Userswaddie96 ★ (talk)07:31, 12 September 2025 (UTC)

Thanks for taking your time presenting a proper solution !

However, the goal is to make the process as easy as it is in a PC, ie, right clicking on a discussion header from the upper-left dropdown menu, and clicking copy.

That's the reason I'm suggesting a⧉ copy symbol beside each discussion header to copy the section links to perticular discussions swiftly on the go.

I lack the technical expertise to do this, so I asked for your help (given there's no objection on this suggestion).Cdr. Erwin Smith (talk)08:09, 12 September 2025 (UTC)

Go up to the talk page Table of Contents and "long press"(?) on the relevant section link? That should give you an option to copy a link —GhostInTheMachine^{talk to me}08:17, 12 September 2025 (UTC)

By Table of Contents, if you're referring to the bullet-list symbol displayed left to the article heading, then I couldn't find it.

A quick GPT search told me that it's hidden in the mobile version of Wiki. I switched to the desktop version through my browser (Brave) options, and it was still missing.

Then I switched to the desktop version through the option available at the bottom of Wiki, and finally found it, exactly where it was in my laptop.

If it is indeed hidden, then the problem can be solved simply by 'unhiding' it ?Cdr. Erwin Smith (talk)13:13, 12 September 2025 (UTC)

I also found 3 user scripts that do more nearly what you want:User:andrybak/Scripts/copy-section-link,User:Nardog/CopySectLink andUser:Polygnotus/Scripts/SectionLinks. I just tried them all. The Nardog version seems the cleanest — it changes the section[edit] to[ edit | copy ] —GhostInTheMachine^{talk to me}08:34, 12 September 2025 (UTC)

There you go, thanks GhostInTheMachine.waddie96 ★ (talk)10:35, 12 September 2025 (UTC)

Okay, I just tried them all.

Out of the 3 of them, onlyPolygnotus works in the Talk page. But even this fails to copy subsections in a discussion, which I think can be done easily through the ToC in a PC.

I thank you for your effort, really. But realistically speaking, how many people can we tell this solution individually?

I didn't raise this issue just for myself after all. Shouldn't it be far more convenient for thousands of less-informed editors like me if the Table of Contents is made available in Mobile version as well !Cdr. Erwin Smith (talk)19:13, 12 September 2025 (UTC)

You can submit a feature request to themeta:Community Wishlist. (As a workaround, my script works on talk pages, including subsection headings.)isaacl (talk)16:06, 13 September 2025 (UTC)

Wikipedia:Requests for comment/User names is a venue to request comment on editors' usernames to determine whether they comply withthe username policy. There have been three (3)archived requests in 2025, and there weresix (6) in 2024. We have too many noticeboards, discussion venues, and processes; dealing with this one seems like some low-hanging fruit.

My thought would be directing future requests toWP:AN and/orWP:ANI if community consensus is required—like all other nuanced "does this person need a block" question. Thoughts?HouseBlaster (talk • he/they)19:50, 6 September 2025 (UTC)

I'll add one other thing: a good portion of the RFCs here result in blocks for vanilla disruptive, which would also belong at AN(I).HouseBlaster (talk • he/they)19:54, 6 September 2025 (UTC)

I had zero idea that the venue in question even exists; I just assumed problematic usernames that aren't quite problematic enough forWP:UAA are discussed at AN and ANI anyway.We don't need a specialized process for everything, so I would definitely get on board with this idea.SuperPianoMan9167 (talk)20:08, 6 September 2025 (UTC)

I did know that the venue existed, but I didn't know that the volume was so low these days.

Yes, I think merging up to WP:AN is reasonable.

I wonder whether theWikipedia:Administrators' newsletter could benefit from an occasional reminder about what WP:U actually says, such as:

• Usernames like "User:Name at Business" are recommended. Disclosing a COI in your username is not promotional. Do not ask them to change their usernames, and do not block them unless you would block a user with a generic name for the same reason.
• Usernames like "User:BusinessName" and "User:Marketing department at Business" are not permitted underWP:NOSHARE rules. Do not block these editors (unless they are breaking other rules), but instead encourage them to follow the directions atWikipedia:Changing username#Requesting a username change.
• Accounts withdisruptive, inappropriate, profane, or offensive usernames should be blocked on sight, even if they have not made any edits.

Think of this suggestion as a type of filler ad: It's not time sensitive, but if there were one or maybe two short bullet points in most issues, then admins and their talk-page stalkers might be more aware of the rules.WhatamIdoing (talk)21:13, 6 September 2025 (UTC)

If it ain't broke, don't fix it. Have there been any miscarriages of justice in those 9 requests in 2024–25? Is there any way the dynamic at AN or ANI would be improved by very slightly increasing their caseload? If it's a no to both, then "too many noticeboards" isn't really a good reason to shut down something that works fine. If anything, I'd rather see movement in the other direction. Disputes about signature appearance could easily be transferred to RFC/N. That's a closely related issue and currently falls into a gap of "should be addressed but too small-ball for AN(I)"... which I suspect is the exact gap that moderate username issues would wind up in if we closed RFC/N. --Tamzin^{[cetacean needed]}(they|xe|🤷)09:03, 9 September 2025 (UTC)

I'm surprised that you see ourproliferation of processes as not broken. If a specific username is too small potatoes for AN(I), perhaps it isn't actually a problem: Only two of those nine actually resulted in consensus that the username was inappropriate (and one of those was also engaging in vandalism), so we clearly have editors being a little trigger-happy. Best,HouseBlaster (talk • he/they)12:59, 9 September 2025 (UTC)

I'll be honest, House, I could probably list 100 problems I have with Wikipedia off the top of my head without pausing, and I don't think "too many noticeboards" would even be in the back of my mind throughout that. I don't think I've ever heard anyone else cite it as a problem till today, either. I've seen discussions that led to noticeboards being shut down, but those were always based on some complaint that the noticeboard was not succeeding at its objective. Based on the link you piped, I think perhaps you're conflating proliferation of policies with proliferation of practices. The fact thatCategory:Wikipedia policies and guidelines has over 200 members is one of our most fundamental problems. The existence of specialized venues to deal with a few of them is neither inherently bad nor inherently good. --Tamzin^{[cetacean needed]}(they|xe|🤷)14:37, 9 September 2025 (UTC)

I see proliferation of practices as a problem, for a similar reason that I see proliferation of policies is a problem: they make it harder to learn how Wikipedia works and less likely that any given one will be followed. If this was, say,WP:BLPN orWP:CCI, I would agree with you that we should have those processes, because the problems they address are mission-critical. Addressing borderline usernames is not, in my view, important enough to have its own noticeboard for the sake of having a noticeboard. (If we regularly had usernames under discussion, that would be a different question.) I am far from the first person to talk about theproliferation of noticeboards. Best,HouseBlaster (talk • he/they)14:54, 9 September 2025 (UTC)

I guess all I can say is, if there's a problem, surely you can point to some bad outcome that has happened because of this being a separate noticeboard. Has anyone been blocked due to an RfC/N who you think wouldn't have been blocked at AN(I)? Has any new user complained that they didn't know what to do in a situation because of the existence of this noticeboard? I can respect your view that there are too many noticeboards—although I stand by itnot being a common view—but speculative harms shouldn't be enough to justify shutting down something that works fine. --Tamzin^{[cetacean needed]}(they|xe|🤷)15:04, 9 September 2025 (UTC)

I think additional venues have some inherit harm in the editor time it takes to monitor and maintain. We have shut down venues for underuse despite being functional (e.g.WP:EAR). I respect your viewpoint, but I suspect neither of us will convince one another of our views. Best,HouseBlaster (talk • he/they)15:28, 9 September 2025 (UTC)

Well sure, like I said, you can point to speculative harms. I can also point to speculative harms, like borderline cases now falling through the cracks because no one wants to take them to AN (which has a mediocre reputation at best) or AN/I (which has a terrible reputation among all but its regulars), or being handled worse because they're no longer before a group of editors who self-select for expertise in this topic. But this is the idea lab, and we're supposed to be workshopping your idea, which is why I'm asking if you have concrete harms that you can point to, because that's the best way to see if there's a workable proposal to be had here. That's part of why I suggested an alternate solution to the board's low volume, namely broadening its scope. The fact that signature issues can only be raised directly with a user or at AN(I) does, in my experience, dissuade people from reporting all but the most blatant issues. Meanwhile the lack of a good place to report these issues means that sometimes admins spring made-up signature rules on users, without a clear venue to appeal to. Or if you're not interested in workshopping your idea, then I'd suggest taking this to VPR rather than debating here. --Tamzin^{[cetacean needed]}(they|xe|🤷)15:49, 9 September 2025 (UTC)

Maybe it's good todissuade people from reporting all but the most blatant issues with aWP:CUSTOMSIG. Maybe I actually don't want someone wasting the community's time at any noticeboard just because they dislike your signature.WhatamIdoing (talk)15:56, 9 September 2025 (UTC)

Wikipedia:Notability/Noticeboard was also shut down primarily for lack of activity. The argument about "too many noticeboards" is made basically much every time someone proposes another noticeboard.

Splintering discussions into lots of little noticeboards risks having nobody reply, or only its enthusiastic creator, or having so few people involved that the Official™ Noticeboard doesn't give advice consistent with the community's current views. Imagine what that recent ArbCom case would have looked like if there were an underwatchedWikipedia:Capitalization/Noticeboard.

Think, too, about the problem from the perspective of a less-experienced editor: Do I take my dispute to this noticeboard or that one or the other one?Wikipedia:External links/Noticeboard sometimes gets questions that belong atWikipedia:Reliable sources/Noticeboard. I don't love telling editors that they're on the wrong page, especially if they're at this noticeboard because someone told them to post there.WhatamIdoing (talk)15:53, 9 September 2025 (UTC)

Merge RFC/Usernames with AN/ANI, merge NOORN, NPOVN, and FRINGEN into an article content policy noticeboard, merge COIN with AN/ANI, and change Administrators' noticeboard to Administration noticeboard.ScottishFinnishRadish (talk)15:48, 9 September 2025 (UTC)

I would agree with all of these changes, mostly because of the issues of smaller noticeboards leading to more insular communities, and because of the fact that they steepen the learning curve for new users wishing to report a specific issues. It might feel like these noticeboards are easy to navigate/understand for us as we're all pretty experienced users here, but this is not necessarily the case for a newcomer who might not even know where to find the correct noticeboard to begin with.ChaoticEnby (talk ·contribs)16:32, 9 September 2025 (UTC)

The proposed merge isn't especially logical. For example, NORN is more closely related to RSN than to NPOVN.

For high-traffic noticeboards like these, there is a problem of scale. When a noticeboard gets too big (e.g., ANI), then people will stop watching it because it's too much volume. When a noticeboard covers too wide a variety of complaints, then specialists will stop watching it because the signal:noise ratio has gone bad.

Also, if I had a dollar for every time someone claimedoriginal research when the actual problem was "no little blue clicky number", I'd be a wealthier woman. Putting everything in one place risks eliding the differences between the separate concepts. It will all become unoriginalverifiareliafringeBLPPOVCOIfringespam (hereinafter "doubleplusbad" or DPB).WhatamIdoing (talk)17:23, 9 September 2025 (UTC)

 You are invited to join the discussion atWikipedia:Village pump (proposals) § RfC: merging Wikipedia:Requests for comment/User names with AN(I).HouseBlaster (talk • he/they)16:16, 14 September 2025 (UTC)

I think we have a vested interest in making sure our deletion nominations have been sufficiently reviewed by humans and proposed by humans. I don't know if this is the only example, but currently there isWikipedia:Articles for deletion/World War 3 (video game) where a large language model was clearly used to propose the article's deletion. I personally am of the opinion that AI generated deletion nominations should be automatically speedy closed, and that we require all noms to be human written. Regardless, we currently do not address the issue at AFD and this a problem only likely to continue to grow. Others might have better policy ideas or a more nuanced approach. Regardless, I do think we need to consider how to handle AI generated deletion proposals atWP:Deletion/WP:AFD as a policy point. The issue isn't going to just go away.4meter4 (talk)13:38, 26 August 2025 (UTC)

You canWP:HATGPT the LLM text and then the AFD isWP:CSK speedy keep due to lack of an intelligible deletion rationale. --LWG^talk13:52, 26 August 2025 (UTC)

Also,Wikipedia talk:WikiProject AI Cleanup is a good place to bring concerns like this. --LWG^talk13:59, 26 August 2025 (UTC)

LWG Thanks. These are good to do suggestions. If this is how we are to handle this type of thing at AFD it should probably be articulated on theWP:AFD policy page itself so that the community is informed on how to respond in future similar contexts. That's the point of this thread, to formulate some sort of guide at AFD on a standard way to respond to this type of scenario. Best.4meter4 (talk)14:41, 26 August 2025 (UTC)

• It depends… if the fact that the AFD nom was generated by an LLM is discovered quickly (before other editors have started to respond) I agree that we should HAT the nom and speedy close.However, if other editors have started to comment (either supporting or rejecting the nom) I think it best to let the process play out. We can note that the original nom was AI generated, and focus the discussion on the subsequent (non-AI generated) comments.Blueboar (talk)15:22, 26 August 2025 (UTC)

  I can agree with that. Again this would be a good thing to articulate on a policy page.4meter4 (talk)16:15, 26 August 2025 (UTC)

  I like this suggestionCzarking0 (talk)12:38, 30 August 2025 (UTC)

• Noting that there is already an ongoing discussion atWikipedia:Village pump (policy)#LLM/AI generated proposals?.ChaoticEnby (talk ·contribs)15:26, 26 August 2025 (UTC)

  Thanks. I wasn’t aware this was already being discussed.4meter4 (talk)16:15, 26 August 2025 (UTC)

• If no one else participates in the debate, then we can speedy delete with G15 criterion.Graeme Bartlett (talk)23:22, 26 August 2025 (UTC)

  Not always possible, G15 is only for specific tells ofunreviewed LLM outputs, not a catch-all for all blatant LLM use.ChaoticEnby (talk ·contribs)23:58, 26 August 2025 (UTC)

I looked at the AFD page, and it takes ~550 words to say that there's a problem, but – wasn't there actually a real problem there, identified in the nom statement as "The body of the article relies almost entirely on two articles from TheGamer and on primary sources (the game’s official website and Steam store pages), rather than on significant, independent, reliable secondary coverage"?

I'm looking through the list of sources at the time of the nom, and it's an internet forum,TheGamer x2, official website, sales page x2, official website x3, and finallyPC Gamer. That's 70% non-notability-suggesting sources. When someone points out a problem like this, do we really want to be playingMother, May I? with them abouthow they described the problem, or do we want to say thanks for letting us know about the problem?WhatamIdoing (talk)00:27, 27 August 2025 (UTC)

Sure there were problems, but not ones that demonstrated a fundamental lack of notability as pointed out by multiple editors. I spent quite a lot of time fixing those problems through editing. The nominator could have easily done that themselves. I still found the LLM nomination disruptive, and I wouldn't want to see a flood of nominations made this way. It should be discouraged.4meter4 (talk)00:44, 27 August 2025 (UTC)

Wouldn't you have found a hand-written nom that made the same points, and put you on a seven-dayWP:HEY timer just as disruptive? I have a hard time looking at that AFD, and the work you put in as a result, and thinking that the biggest problem was the style of the nom's statement.WhatamIdoing (talk)22:58, 27 August 2025 (UTC)

I think over focusing on this nomination is ared herring. That nomination isn't the topic raised here.4meter4 (talk)23:35, 27 August 2025 (UTC)

That nom, though, gives us an opportunity to ask: Is the real problem bad noms (e.g., treating AFD like a form of on-demand cleanup), or is the problem bad explanations for the nom?WhatamIdoing (talk)00:36, 28 August 2025 (UTC)

Sure, if you want to distract attention away from the original more pressing question. I find the line of discussion unproductive within this thread.4meter4 (talk)00:45, 28 August 2025 (UTC)

I find the idea that an AFD nom should be considered invalid based on how the OP wrote the nom statement to be too close totone policing for my comfort.

It reminds me of other communication problems that we see in the real world. For example, after a natural disaster, the victims spend a couple of days saying how grateful they are for the help and support of the government agencies and non-profit organizations. Around day four, they start saying how terrible these same people are: the sleeping arrangements are bad, the food is lousy, nobody knows what's going on, and why isn't there a government agency sending a drone over my house this directly minute, so I can see whether it's still standing?! This is considered a good sign by the professionals (because it means that the complainers are no longer terrified that they'll die any minute now).

To give another example, when there's bad medical news to share, there's an initial shock, and after that, whoever told you that the baby's going to die/the cancer is untreatable/fill in your nightmare here was the worst possible person to do this and did the worst possible job at it. It should have been the first provider to suspect it, instead of making you wait for the specialist. Or if the first person told you, they should have made you wait for the expert. They should have told you the situation straight out instead of beating around the bush, or if they were direct, they should have slowly and gently led you to the information, instead of dumping it directly on you. And so forth. These complaints, too, are considered good by professionals, because if you don't hear complaints about how you were told, it often means you didn't understand what you were told. The best way for an oncologist to get high scores as a "good communicator" is to never tell the patients any bad news and never encourage the patients to think they might die.

So when I see a complaint about AFD nomination statements being written in the Wrong™ way, I wonder: is the problem with the nom's statement, or is the problem with the nom's decision to nominate it in the first place?

There is no LLM-based bot that creates the AFD nominations. The decision to send an article to AFD is being made by a human, and the 'problem' of an AFD nomination is the fact that the article is listed at AFDat all, not that the nom's statement does/doesn't match a certain style. Noms are encouraged to write a nom statement, but it's not, strictly speaking, a necessity. The{{afd2}} template doesn't break if you provide no rationale at all.

I could imagine a "rule" that says that most AFD nom statements are less than 150 words, in the hope that any LLM users would then add "in 150 words or less" to their AI instructions. In practice, I would expect this to work just about as well as the existing rule that says not to merely say that it's non-notable, which a glance through AFDs suggests is "violated" in about 10% of the noms. But I don't think that we need a rule that says "BTW, after you used your human brain to decide to send this article to AFD, don't use LLM to polish up your reasons for AFDing it".WhatamIdoing (talk)03:41, 28 August 2025 (UTC)

Oh, and just for context: I looked through about a hundred AFDs without seeing a single one that looked like the nom used an LLM to write their explanation. So even if we agreed that it's inherently bad to use an LLM to organize the nom's statement, it probably doesn't make a practical difference in >99% of the cases.WhatamIdoing (talk)03:43, 28 August 2025 (UTC)

This diatribe is hardly helpful in this context, and I personally am not interested in engaging with any editor who uses an adversarial conversation style. This is a place to incubate solutions and ideas productively and collegially. We still need guidelines at AFD on how to respond to AI written text, and guidance for editors on what to do when AI written text creates problems. There is a reason why there isan entire WikiProject dedicated to this problem, and already policies being crafted at RFCs; such asWP:HATGPT. If you aren't here to make useful contributions in policy writing, and can't communicate in a calm and respectful manner, I respectfully suggest you move on so that others can work.4meter4 (talk)04:10, 28 August 2025 (UTC)

We still need guidelines at AFD on how to respond to AI written text – do we? If almost none of the noms, and very few of the responses, use LLM-generated text, do we reallyneed rules about this, or might that beWP:CREEPY?

We still need...guidance for editors on what to do when AI written text creates problems – Maybe, but do we need anything actually specific to AFD for this? Do you expect a rule that says "If it's a talk page, do X, if it's an RFC, do Y, if it's AFD, do Z"? I don't.WhatamIdoing (talk)04:38, 28 August 2025 (UTC)

I believe we do, yes. There's already confusion over how or when to implementWP:HATGPT for example. If you aren't in agreement that there is a problem, that's fine. Don't expect me to engage with you though. I'm interested in developing guidelines that restrict AI use and/or limit its damage at AFD because I do perceive it as harmful. If you don't agree with the premise and aren't interested in helping with that I respectfully request you keep your opposition to yourself and wait to express it during a formal policy proposal. Please don't impede other editors wishing to incubate an idea that they believe in, even if you do not.4meter4 (talk)04:52, 28 August 2025 (UTC)

Please don'tWP:INTERLEAVE comments.

The idea lab isn't a "supporters only" page. If there are problems with an idea, those should be pointed out clearly and early, so that editors can think of ways to adjust the idea, or at least to prepare an explanation. For example, you think this is a big deal, and I've determined that less than 1% of AFDs are affected, so you should either adjust your perception (maybe it's a pet peeve instead of a big deal) or you should prepare an argument ("Sure, it's only 1%, but 1% drives me nuts, so we should have special rules for AFD that are different from the normal TPG rules anyway just in case it gets worse in a few years"). It's better to hear about the holes in your idea now than to have your proposal fail later.WhatamIdoing (talk)05:00, 28 August 2025 (UTC)

My simple response at present to you is that, AI use can be a form ofWP:DISRUPTIVE editing at AFD both intentionally and unintentionally. It could also be a tool used for good when used responsibly. This is where guidelines on using AI at AFD could be helpful in both determining appropriate use and what is disruptive use.WP:VANDALISM only happens in a small % of edits, but that doesn't mean the consequences aren't serious and that a standard response isn't needed with set community guidelines/protocols in place. Additionally, current AI use at AFD is low because AI use overall is in its infancy. The problem is likely to grow rapidly in the near future and we need to have responses and guidelines in place before that exponential growth happens not after. Being prepared for what's coming is responsible and smart. Additionally, the assertion that humans are automatically behind the decision to enact AFDs is a supposition that we can't necessarily hold fast to when LLMs are used. As AI becomes more sophisticated and more independent from human oversight it's possible in the near future that we may have nominations made with no human engagement, input, or direction at all in the process. These are issues we should be discussing and planning for now as they are foreseeable problems. Lastly, there isn't yet a formed idea here to vet, and for that reason I don't find the adversarial questioning particularly helpful. Constructive input on what a potential guideline could be would be more useful at this stage as no fully formed idea has been developed yet. Best.4meter4 (talk)05:49, 28 August 2025 (UTC)

How exactly was that response not "calm and respectful"?Gnomingstuff (talk)05:36, 28 August 2025 (UTC)

To me the tone came off as overly combative, but perhaps that's just my impression.4meter4 (talk)05:59, 28 August 2025 (UTC)

The decision to send an article to AFD is being made by a human Hardly relevant – said human was so lazy they then turned their brain over to an LLM to write a shoddy nomination for them, and we should dismiss it with the same lack of effort. Unlike you, I don't believe the ends justify the means: we should inspect the process, not just the result – hence our heavy restrictions on paid editing, on COI, and mainspace bot use.Cremastra (talk ·contribs)06:55, 3 September 2025 (UTC)

I thinkWikipedia:Product, process, policy has the order right.WhatamIdoing (talk)17:00, 3 September 2025 (UTC)

Related to the above: I have justWP:PGBOLDly added a new point "E" toWikipedia:Articles for deletion#Before nominating: checks and alternatives, to say that nom statements are usually less than 150 words. Maybe, eventually, it will have some small effect on bloated statements from newbies.WhatamIdoing (talk)15:29, 4 September 2025 (UTC)

Which policy or guideline defines a percentage limit on how many sources in an article have to be "notability-suggesting"?Anomie⚔01:49, 27 August 2025 (UTC)

None. Regardless, this is off topic to the main point of discussion in this thread which isn't really about any one specific nomination.4meter4 (talk)02:02, 27 August 2025 (UTC)

NPOV says "In principle, all articles should bebased onreliable,independent,secondary published sources with a reputation for fact-checking and accuracy", which could be interpreted as meaning (using very approximate numbers) that half or more of an article should come from/bepossible to cite to a "notability-suggesting" source. This isn't about the percentage of sources (e.g., three out of the ten sources cited in the article) but about the percentage of the article content (e.g., everything in two out of three sections).

Having said that, I don't find that this is a very common interpretation (hardly surprising, as the sentence itself is only a couple of years old, and it usually takes a couple of years for a critical mass of editors to notice the existence of such a sentence in a policy).WhatamIdoing (talk)00:39, 28 August 2025 (UTC)

Sigh, more primary source paranoia getting snuck into policies by people who focus only on the contentious parts of contentious articles.Anomie⚔00:55, 28 August 2025 (UTC)

The "secondary" language was boldly addedabout 2.5 months ago; "independent" isfrom 2.5 years ago, and was discussed on the talk page. If you don't think it's helpful in that location, you could always revert it.WhatamIdoing (talk)02:22, 28 August 2025 (UTC)

I don't have the energy to fight a battle against all the people who loveWP:PSTS because it lets them say "this is primary so it's bad" instead of having to address the actual reliability or POV of the source directly. Same goes for "independent" there.

Primary, first-party sources are perfectly fine, and possibly even preferable, for statements of plain fact about which there's no challenge or controversy. If an article consists of mainly those kinds of facts, it would be fine for the article to be mainly based on such sources onceWP:N is satisfied. Maybe the article is just that uncontroversial, or maybe it's still stub- or start-class and no one has added the controversial parts that need independent sources to avoid POV issues or the kind of analysis that isn't present in primary sources.Anomie⚔14:33, 28 August 2025 (UTC)

In particular, as pointed out in the response to the deletion request, an article being based almost exclusively on primary sources is not a reason for deletion. Rather, it would be a reason to mark the article as low-quality and in need of improvement per the guidelines documented atWP:SURMOUNTABLE.

But beyond that, this thread is about the use of AI tools for creating deletion proposals rather than the specifics of a particular proposal. However, the debates around the legitimacy of this deletion request point to a good reason to disallow AI agents to create the requests. The rules for whether an article should be retained are not clear cut and there are instances where a discussion is needed before we can decide that an article should be deleted. If a human did not submit the request, who is going to defend its existence when somebody objects to the deletion? Normally, I'd expect that the requestor can respond with their rationale for submitting the request.

I would also be concerned about the fact that if an AI-submitted deletion request is accepted by one individual, that would mean that only one person was involved with the deletion. Under the expected process, we would at least have two people making the decision: the person that initially reviewed the article and submitted the request for deletion, and the other who approved it.

However, I can see the benefit of using machine learning models to identify articles that should be deleted, but I don't think that those deletion requests should be listed among the normal deletion queue. At the very least, they could be in a separate queue that humans can review and decide whether they want to sponsor the deletion request through the normal process. And if they do prove to generally be of high quality, then we could make a better determination about how to integrate LLMs into the deletion process.

Ovenel (talk)21:11, 28 August 2025 (UTC)

This is almost entirely irrelevant:

• AI agents are not submitting requests, humans are.
• Only one person is involved with very nearly all XfDs: the person who nominated the page for deletion (the exceptions are when someone is unable to create the nomination page for some reason).
• The only time a single person is involved in a deletion is when an administrator speedily deletes a page themselves. All administrators are human.
• The person who nominated the page for deletion can respond to/defend the nomination. Whether the deletion rationale was or was not written using an LLM is completely irrelevant to this.Thryduulf (talk)23:04, 28 August 2025 (UTC)

I agree that the idea of a bot submitting articles to AFD feels like a bit of a tangent to the original question, but I also agree with Ovenel, in that I think such a bot would be a bad idea.

If it were possible to have a bot that could do a properWP:BEFORE search (and someday it might be, though probably not any time soon), I wouldn't trust its (non-existent) judgement about notability, and I'd worry about flooding AFD. But I might also wonder about whether it could be useful in reverse, e.g., to suggest sources that the human noms had missed.WhatamIdoing (talk)23:15, 28 August 2025 (UTC)

I agree with you. Any such bot would obviously be limited to machine-readable sources (although it's probable that a greater proportion of future sources will be machine-readable and possible that some sources which are not currently machine-readable will be in future) that are indexed online in some fashion (I don't have an opinion on whether it is likely that the proportion of sources this represents will change), which is obviously insufficient for a proper BEFORE search in at least some topic areas. These all mean that a bot would be much more reliable for articles related to e.g. 21st century American pop culture than it would be for articles related to e.g. 17th century African poetry.WP:SYSTEMATICBIAS would therefore obviously be very relevant. This is getting very off-topic for this discussion though.Thryduulf (talk)23:34, 28 August 2025 (UTC)

I guess I was misunderstanding the policy proposal that is being discussed here. I thought the issue was in regards to setting up a bot to submit AFD requests, using an LLM to generate the language of the request after some automated process has determined that an article should be deleted. I.e., supplementing the sorts of bots that already exist in the Wikipedia ecosystem with an LLM. If the concern is that people would be using an LLM to write up their request after they've already determined an article should be deleted, I'm not opposed to that.

I can appreciate the concern about allowing minimal effort submissions, and I know that people who have experience working with open-source software have likely seen the issues that LLMs can bring to the table (namely, many open-source maintainers have found that reviewing low-quality, LLM-generated pull requests have ultimately made it harder to work on their projects rather than easier). But I'm reticent to reject a new tool before we really see it being used unless there is a very clear cause for concern.

Ovenel (talk)14:44, 2 September 2025 (UTC)

@Ovenel Check outCategory:Articles containing suspected AI-generated texts from August 2025. And these are the article-space issues obviously, but it’s definitely turning into a time sink there already. —Locke Cole •t •c15:12, 2 September 2025 (UTC)

Also, if you want an idea of how many LLM/AI comments are being made and documented so far,here's a good place to start. —Locke Cole •t •c15:31, 2 September 2025 (UTC)

@Locke Cole Thank you for the links. From a quick review, I found this chain inTalk:History_of_Christianity#Errors_according_to_AI that exemplifies the problem well. The AI-generated text can be very long and take a lot of effort to review. And the issues that the AI identifies do not appear to generally be legitimate, or, at the very least, miss the context that justifies the choices made by the contributors to the article. In this case, it looks like somebody spent a minute generating a big wall of text to post in the talk page, and it had to have taken at least an hour for somebody else to review it and ultimately find that no changes to the article were warranted.Ovenel (talk)04:27, 3 September 2025 (UTC)

it looks like somebody spent a minute generating a big wall of text to post in the talk page, and it had to have taken at least an hour for somebody else to review it and ultimately find that no changes to the article were warranted. And that is kind of the crux of the problem. We need moreeditors to replace folks who retire, quit, burn out, or simply die, so there's this natural inclination to not want toWP:BITE the newbies (or in this case, someone who isn't even that new as far as account age goes). But if this goes unchecked, our volunteers (you, me, everyone else) who have to deal with this will find ourselves bogged down in many times pointless discussions when we could have been researching, writing, or maintaining many of the other important elements of a project like this.

I don't know what the answer is. Or if there even is one that would be compatible with how Wikipedia works. —Locke Cole •t •c04:40, 3 September 2025 (UTC)

When discussions feel low-value, people will ignore them.WhatamIdoing (talk)19:22, 3 September 2025 (UTC)

I'm not sure if this is really a continuation of thesorts of bots that already exist – as far as I know, we don't have bots determining whether articles should be deleted, or submitting AfD requests by themselves.ChaoticEnby (talk ·contribs)13:26, 7 September 2025 (UTC)

We don't, nor am I aware of any proposals to do so.Thryduulf (talk)13:52, 7 September 2025 (UTC)

When I say "sorts of bots that already exist", I mean it as, "automated processes that perform editor duties on Wikipedia." I.e., the types of bots that are listed atWP:Bots. I was merely saying that we already have bots performing editor tasks, and I thought that this discussion was about whether we should allow bots to be created that use an LLM to expand the scope of automated activities, but I understand that that is not what was being discussed here.Ovenel (talk)15:53, 15 September 2025 (UTC)

• Didn't read the full discussion but my 2 cents - if an AI started the deletion discussion, no matter how many participants have commented, the discussion should be closed, negated, and filed away. AI does not run Wikipedia and should certainly have no "voice" in determining which articles, categories, etc., to keep or not to keep.Randy Kryn (talk)14:55, 29 August 2025 (UTC)

  No AIs have started (or can start) deletion discussions.Thryduulf (talk)15:14, 29 August 2025 (UTC)

  People often use AI as a proxy for their own thinking, so I can see where Randy is coming from.Bremps...16:30, 29 August 2025 (UTC)

  @Bremps, could you explain this a little more? I'm not seeing the connection between what I think @Randy Kryn is talking about (an autonomous AI bot starts an AFD discussion with no human input whatsoever – because that's "an AI started the deletion discussion", as opposed to "a human started the deletion discussion, and happened to use AI to write out some of the sentences") and what you seem to be talking about (people using AI to produce sentences).WhatamIdoing (talk)17:04, 29 August 2025 (UTC)

  I'm not 100% sure if this is what Randy meant, but I interpreted the comment as referring to someone lazily using AI to decide to nominate for deletion and to come up with the rationale.Bremps...17:07, 29 August 2025 (UTC)

  I'm not sure how AI could decide to nominate an article for deletion? Even if it somehow did, a human would have to be the one to decide to nominate, place the tags, create the nomination page, and submit a rationale. They should also be the one to transclude the nomination page but a bot will do that for them if they don't.

  The only relevance any sort of AI has to deletion is that it is possible for someone to use an LLM to write or assist with writing some or all of the deletion rationale. I don't think that we need anything more than the existing policies and guidelines to deal with this - particularly HATGPT and speedy keep.Thryduulf (talk)17:19, 29 August 2025 (UTC)

  Thanks for the speculation, there's a good Wikipedia short story in there somewhere. As I mentioned, I didn't read the discussion and jumped in after reading the original post above, which strongly implied that AI can open a deletion discussion without human input. Having no idea what AI can or cannot do (I've never used any of the things, including Chat, and have no intention of trying them out) I followed up on that. Glad to hear that it, and I, were mistaken.Randy Kryn (talk)22:55, 29 August 2025 (UTC)

  I just wanted to clarify that building a script that uses AI to fully autonomously start deletion discussions is within the technical of many editors. The design would look like this.

  1. Find an article that you want to delete
  2. Press F12 and go to your network tab
  3. Inspect the network traffic that occurs when the existing XfD tool runs the AfD script.
  4. Port these curl requests to your favorite scripting platform
  5. Modify the requests to take in parameters
  6. Create a list of articles to check similar to how AWB works. If I was to do this I would get the list of low quality status for WikiProject Energy
  7. Use existing libraries to get the article text for each article in the list
  8. Implement a chunking strategy that provides some of the article text, a prompt, and makes the LLM return JSON indicating a deletion likelihood and reasoning
  9. Create a separate but similar script that runs a BEFORE on the articles that score high. Existing google search packages can be used to run the BEFORE without AI
  10. Then the results of the two scripts are returned back to the LLM which is requested to create another JSON indicating if after the BEFORE AfD is still the right path.
  11. This json is used to algorithmically call the curl endpoints that the existing XfD tool runs.

  One could even use the tokens from their browser to make the AfD appear as their user.Czarking0 (talk)15:32, 2 September 2025 (UTC)

  The part that worries people isn't 2 through 11. The part that worries people is "Find an article that you want to delete". That's the part that must be done by a human.WhatamIdoing (talk)17:56, 2 September 2025 (UTC)

  It does not have to be done by a human. As I stated you could pick a category and recurse arbitrarily. I could run this on every article in category BLP. That does not meet the bar of a human picking an article to delete.Czarking0 (talk)19:36, 2 September 2025 (UTC)

  Oh nevermind my first comment, you misunderstood the first point. That first article is just to get the network traffic that you need to replicate with the script. If you already know the network traffic or are familiar with how the existing XfD tool works you can skip the first step.Czarking0 (talk)19:38, 2 September 2025 (UTC)

  I suspect that automating step 1, in the sense of sifting through categories to find unwanted articles (but not so much in the sense of getting the network traffic needed to automate the rest), is exactly the thing that scares people and would result in flooding AFD.

  I suspect that the very minimum amount of human thought people would normally accept is:

  • a human finding the unwanted article (I personally sift through the category; I personally scroll throughSpecial:NewPages to spot articles about subjects I believe are non-notable; I personally stumbled across the article while doing something else;maybe I personally check the result of some complex search or scan [e.g., BLPs inCategory:Professors with anORES category ofSTEM*, whose inline citations all have URLs whose domain names end in .edu]), and
  • the human glancing over the bot's output to see whether it's reasonable (e.g., to notice that the BEFORE search relied entirely on checking recent news reports, when the subject is not the kind you expect to find in the news).

  Even that would upset people if too many of these happen. (And speaking for myself, I want the human to decide what the deletion reason is, even if they're feeding their reason to an LLM, and even if their reason is wrong/on theWP:ATA list.)WhatamIdoing (talk)20:11, 2 September 2025 (UTC)

I agree with this, though I also like @Tryptofish's idea at the RFC of making LLM/AI something peoplemust disclose. That sub-discussion is ongoing atWikipedia:Village_pump_(policy)#Alternative_approach:_make_transparency_policy. —Locke Cole •t •c06:37, 30 August 2025 (UTC)

I have been thinking about the "must" and wondering whether we should treat this the same way as we do other rules about obviously bad behavior. For example, we have{{uw-vandalism1}},{{uw-vandalism2}},{{uw-vandalism3}}, and{{uw-vandalism4}}. Vandalism is obviously/inherently bad in a way that LLM use isn't (i.e., someone with dyslexia or limited English might think that using an LLM is helpful, whereas nobody thinks that poop vandalism is helpful), but we give four warnings for it, and we normally require one warning before you can make a report toWikipedia:Administrator intervention against vandalism. Only vandalism happening after the first warning risks a block (unless it's egregious).

So maybe we should follow our usual pattern. That would mean re-writing{{uw-llmtalk}} to assume that the person is doing their best and posting their own ideas (just not in their own words), and not strike or remove the comments until multiple warnings have been delivered and the person persists in using LLMs post-warning.WhatamIdoing (talk)19:41, 30 August 2025 (UTC)

That would certainly be less bitey.Donald Albury20:46, 30 August 2025 (UTC)

I keep thinking about how I respond to rambling, long winded talk page comments that are NOT written by LLMs… my reaction is almost always: “TLDR - please summarize”.

So perhaps the first warning when LLMsare used should be something similar… a friendly request to “summarize using your own words”.Blueboar (talk)20:46, 30 August 2025 (UTC)

There are already escalating templates for llms:Template:Uw-ai1,Template:Uw-ai2,Template:Uw-ai3,Template:Uw-ai4, although they are not included in the Twinkle defaults so I don't think they are very well known.CMD (talk)04:57, 3 September 2025 (UTC)

Thanks for those links. It is in Twinkle, under "Behavior in articles".WhatamIdoing (talk)17:55, 3 September 2025 (UTC)

Sigh, looks like another trip back to the personal script mines.CMD (talk)14:19, 4 September 2025 (UTC)

•Opinion : I generally support the use of LLMs, specifically in Edits, 3-O, RfC, RSN, NPoVN, DRN, ArbCom etc, as I haveconveyed in the other discussion ongoing in the village pump.

But constructing an entire article simply takes too much time and effort to be nominated for deletion by a single promt. Thus, using LLMs to delete an entire article should be considered disruptive and bad faith.Cdr. Erwin Smith (talk)11:50, 4 September 2025 (UTC)

I don't think the second half of this makes sense. We have no way to know if the LLM's output is from a single prompt. You can do much more precise things with multiple prompts linked by scripts. See my design above.Czarking0 (talk)14:12, 4 September 2025 (UTC)

If you are talking about using an LLM to write the deletion nomination, then a single prompt is very likely going to be sufficient to produce a coherent and relevant nomination if the rationale is straightforward and the application of the relevant policy/guideline to the article is clear. I do fully agree though that an LLM should not be deciding whether and/or why to nominate an article for deletion.Thryduulf (talk)14:26, 4 September 2025 (UTC)

I agree that there is a significant distinction between having an AI search for articles to delete, and an editor using an LLM to generate thetext to explainwhy the article should be deleted.

For the latter… the primary issue seems to be that LLM’s tend to generate walls-of-text. However, so do editors writing their own rationals.

We deal with human written walls-of-text by saying “TLDR” and asking the human tosummarize their argument (and respond to the summary. We can do the same with nomination text generated by LLMs. This forces the humanbehind the LLM to engage.Blueboar (talk)14:58, 4 September 2025 (UTC)

The same LLM can summarise its generated text as well.Cdr. Erwin Smith (talk)15:48, 4 September 2025 (UTC)

So? If someone is asked to summarise their arguments there are three possibilities:

1. They don't.
2. They do, and it results in a concise, coherent and relevant rationale
3. They do, but the result is still lacking in one or more of those attributes.

In none of those cases does it make a difference whether any of the text was written by a human or by an LLM - we already have all the policies, guidelines and practices to deal with the matter, and we already apply them as appropriate on a daily basis so we don't need anything more than we currently have.Thryduulf (talk)16:00, 4 September 2025 (UTC)

But I do like the idea of asking people who post long LLM-generated messages on the talk page tosummarize their comments.

The existingTemplate:Uw-ai1 template series is all about suspected LLM-generated content in articles. I think we might need a separate one for discussions.WhatamIdoing (talk)17:13, 4 September 2025 (UTC)

As I've stated, I have no problem with people using AI for all the different purposes, but using AI for deleting an ENTIRE article is on a whole other league.

I've witnessed an article in which I've contributed significantly getting deleted firsthand, and believe me, it hurts.

People need to feel the gravity of this type of a request. The only barrier which can prevent any random person from waking up and posting a request for AFD is the manual mental labour that he'd have to put in.

So I believe atleast this sector needs to be kept clear of any AI.Cdr. Erwin Smith (talk)17:50, 4 September 2025 (UTC)

Random people don’t have to put in much mental labor to file an AfD, even if they are human. We getlots of AfD nominations where the only rational given is “Not notable”.

Hell, such minimalist nominations probably takeless mental effort than using an LLM (with an LLM you at least have to type out a prompt)Blueboar (talk)19:31, 4 September 2025 (UTC)

One wrong doesn't justify another.

With LLMs in the mix, you'll start getting a flood of increasingly sophisticated proposals which you would't be able to dismiss instantly.

And guess what will be their targets? Articles crafted by you and me through our hands.Cdr. Erwin Smith (talk)19:56, 4 September 2025 (UTC)

WP:EFFORT is on the list of arguments to avoid in deletion discussions.WhatamIdoing (talk)20:32, 4 September 2025 (UTC)

That was written with humans in mind, not machines.Cremastra (talk ·contribs)21:18, 4 September 2025 (UTC)

Yes: It was written to say that even if we're talking about "Articles crafted by you and me through our hands", the fact that it took human effort to create an article does not mean that the article is exempt from our notability standards.WhatamIdoing (talk)21:39, 4 September 2025 (UTC)

As @Cremastra stated, this rule should be abided if a human writes the AfD. But the discussion is aboutcreating a new rule which discards all types of AI written requests for AfD, and that was the rationale behind my comment.Cdr. Erwin Smith (talk)03:54, 5 September 2025 (UTC)

But we don'tneed such a rule, because the rules that apply to humans already apply to AI equally: if it's disruptive in some way we deal with it in the manner appropriate to the nature of the disruption. If it isn't, then we treat the good faith contribution as a good faith contribution.Thryduulf (talk)04:39, 5 September 2025 (UTC)

I agree.

My comment is narrower: If it's bad for me to write the article ___, for a human to decide it's non-notable, and for a human to use an LLM to write a fancy-sounding AFD nomination statement because deleting "my" article will lose all myWP:EFFORT, then it's equally bad for the same thing to happen, only using a hand-written AFD nomination statement.

The nom statement isn't what causes myWP:EFFORT to be kept or lost. The other editors' analysis of whether "my" article is appropriate for Wikipedia is what determines whether my EFFORT is lost.

To put it another way: No matter who/what writes the AFD nom statement, and no matter what they write in in, editors are not going to agree to delete articles I've written, such asBreast cancer awareness. And conversely, if someone writes an article about a totally non-notable subject (e.g., a brand-newgarage band, a very minor character in a video game), there is nothing anyone can write in a nomination statement that will make editors agree to keep it.

AFD runs on sources and facts about articles. It does not run on eloquent statements.WhatamIdoing (talk)04:57, 5 September 2025 (UTC)

What's slightly more annoying is when you have articles that fall in between "garage band" and "something the vast majority of English speaking are familiar with".Bird Conservation Nepal, is my go to example, especially as ofthis revision, notability tagged since 2018 and sourced mostly to the organization's own website. It was tagged for deletion[18] by an UPE sock gaming autopatrol, and in part using AI-generated !votes and deletion nominations to make themselves seem like a real editor[19] much more quickly and lazily and, from the looks of the SPI page, effectively than they would otherwise. This nomination in particular irked me at the time, because part of it wasn't true and searching the organization's namegave me a lot of sources, but I knew that, as a short stub on a business on the Indian subcontinent sourced to the organization's own website, written not in slightly SEA-influenced English that was very deferential to the org's founder, was quite likely going to get killed at AFD no matter the merits of the article. But yes, the use of AI was disruptive in this case, because it allowed the spamfarm to spamfarm more effectively, but then again, I and four other editors didn't catch on at the time & the discussion had to go for two weeks.

What I'm really disagreeing with, is, I suppose, is that idea that AfD runs on sources and facts, not eloquent statements, because people themselves tend to give a lot of attention to presentation than they should. I had a great conversation with a linguistics professor once, about the fact that when he switched to his natural (American) Southern accent, people looked at him like he was stupid, but when I spoke in my (then quite strong) English accent, people saw either "sexy" or "genius", and it had very little if anything to do with what we actually said. Similarly, I once had a mind-boggling conversation with somebody who kept insisting that somebody using the phrase "do the needful" was obviously up to no good, because 'only scammers spoke like that'. And maybe I'm really cynical, but I don't think AfD participants are immune to those biases. I definitely don't think a non-obvious AI telling them in perfect, American English that this slightly clumsily written article on an Indian actress or business isn't compliant with Wikipedia policies is going to prompt much push back. Not that this is entirely a problem with AI... I mean, it was an open secret, for years, that a large number of U5s were being done on just normal, attempted drafts or Wikipedia profile pages from SEA editors. And, back to the 'LLM generated deletions are used by bad faith actors for disruption' - if anybody discovers a way to get intentional UPE sockfarms to stop being the little shits that they are, lmk and I'll try and get you to, idk, negotiate a peace deal in Gaza while you're out there doing miracles.GreenLipstickLesbian💌🦋07:20, 5 September 2025 (UTC)

I don't think that LLM nomination statements (i.e., the ones we've noticed) cause Wikipedia editors to agree with the nom. If anything, the reaction seems to be the opposite: If you're so unfamiliar with AFD and its conventions that you post 450 words with bullet lists and bold-faced headings, your nomination is automatically suspicious.

A nom statement of "lacks significant independent coverage, relies on non-reliable sources, or serves as promotional content rather than a neutral, verifiable encyclopedic entry" (fromyour example) doesn't make me suspect the use of an LLM. (It might make me wonder whether someone copy/pasted this from an unofficial summary of our notability standards.)WhatamIdoing (talk)20:34, 5 September 2025 (UTC)

Yeah, the copy-pasting thing wouldn't surprise me either - but that phrase has never been used before on the Google-indexable internet[20], so, combined with the other evidence at the SPI, I'm pretty sure they copy-pasted a sentence fragment from a chatbot's explanation, to obscure what they were doing. And I wouldn't have thought LLM either (and didn't - tbqh I thought "oh, joy, somebody gunning for NPR who has no idea what they're doing"), until I saw the fact that several other socks had made the same style of !votes and nominations, after the SPI had concluded. But that brings us back to the same issue - the obvious LLM generated statements are (likely) going to be ignored by human editors some of who will !vote to spite the machine, and the non-obvious ones brought by editors acting in bad faith won't be impacted by any policy, because, well, editors acting in bd faith don't listenGreenLipstickLesbian💌🦋22:00, 6 September 2025 (UTC)

Hmm, I was thinking more of an inept person copying from a UPE business' internal documentation:

"Dear New Hires, welcome to Scammers R Us. Your first assignment, after setting up your account, is to build up your credentials by finding and nominating some articles for deletion. Wikipedia will delete any article that lacks significant independent coverage, relies on non-reliable sources, or serves as promotional content rather than a neutral, verifiable encyclopedic entry. Look for articles about smaller businesses or non-profit organizations that have very few citations and no active editors. Before nominating an article, you must get written approval from your manager, so your manager can check the list of past clients in our confidential client database. After you have approval, useWikipedia:Twinkle "XfD" to nominate the article for deletion. Put a short description of the reason for deletion into the Twinkle form."WhatamIdoing (talk)23:00, 6 September 2025 (UTC)

But sterner regulations can and should cover AI use.

Your approach of ignoring the use of AI as if its not problematic and steadfastly denying that AI is fundamentally incompatible with a human encyclopedia is troubling to me, as I think it misses the gravity of the situation we're facing.Cremastra (talk ·contribs)05:33, 5 September 2025 (UTC)

But sterner regulations can and should cover AI use. Why?

Your approach of ignoring the use of AI as if its not problematic I'm absolutely not doing this. AI can be disruptive, and when it is disruptive it needs to be dealt with. However given that we can deal with this disruption using existing policies and guidelines we don't need new policies or guidelines. As an advantage this means that disruption can be dealt with as disruption even if it isn't clear whether the editor causing the disruption is using an LLM or not.

AI is fundamentally incompatible with a human encyclopedia this is statement of your personal world view presented as if it were incontrovertible fact. You are entitled to hold that opinion, but it is not factual or based on fact and you really need to stop claiming it as fact.Thryduulf (talk)09:11, 5 September 2025 (UTC)

The capability of AI in making a good faith AfDs isn't in question here. The significant uptick in AfDs if it's kept legal is what worries me. Tomorrow, anyone with an agenda in mind could get up and file multiple sophisticated AfDs by using AI generated text.

This can cause a major disruption in the entirety of Wikipedia, and with how much popular the AI apps are getting day by day, I don't think we should take that chance.Cdr. Erwin Smith (talk)07:14, 5 September 2025 (UTC)

If someone does this then they will be quickly stopped as disruptive, regardless of whether we have a specific policy against it because we already have policies about disruptive deletion nominations and policies about unauthorised bot and bot-like editing. The existence or otherwise of a policy like this specifically targetted at a very low likelihood event will not make any difference to whether it happens or not - relatively few people are capable of scripting in the manner that would be required, only some of them will be aware we have a specific policy against it and even if they are they would be violating several other, better known policies anyway so one more won't stop them. At absolute best it's pointless theatre, at worst it will make it harder for the disruption to be dealt with given that there will be utterly pointless but time-consuming debate about whether it is LLM disruption, human disruption or some combination of human and LLM disruption that will take time and effort that could be better spent dealing with the disruption (or you know, improving the encyclopaedia).Thryduulf (talk)09:21, 5 September 2025 (UTC)

@Cdr. Erwin Smith says thatTomorrow, anyone with an agenda in mind could get up and file multiple sophisticated AfDs by using AI generated text.

I say:Today,anyone with an agenda in mind could get up and file multiple sophisticated AfDs by usinghand-written text.

If flooding AFD is a concern, we need a rule against flooding AFD. If flooding AFD is a concern, we do not need a rule against one of the many ways that AFD could be flooded.

• YGood: Flooding AFD would be bad, so we create and enforce a rule against flooding AFD.
• NBad: Flooding AFD would be bad, so we create and enforce a rule against using LLMs to write the nomination statement, but all the other ways people might flood AFD are okay with us!

Previous discussions on the risk of flooding AFD have suggested limits around 20 articles on a similar subject at a time (e.g., 20 Brazilian Olympic athletes on Monday, 20 Alaskan cosmetic surgeons on Tuesday, 20 European military units on Wednesday, 20 African businesses on Thursday...). A simpler approach might be "maximum of X nominations every Y days".WhatamIdoing (talk)22:13, 5 September 2025 (UTC)

•Support a blanket rule against flooding, since that's the primary issue, and this will automatically solve the potential threat posed by LLMs.

But the limit needs to be much lower. Honestly speaking, 20 at a time per user sounds ludicrously high regardless of the subject barrier. I would like to hear :

1. What others think this limit should be?

2. Whether filing an AfD should be ECU restricted?Cdr. Erwin Smith (talk)07:22, 6 September 2025 (UTC)

@Cdr. Erwin Smith, the community won't agree to restrict AFD nominations only to 0.25% of registered accounts. For one thing, many hands make light work: we actually need newer editors (like you) to let us know when they find a problem. For another, if you don't allow AFD nominations, the restriction will encourage other editors to misuse other process, such asWikipedia:Proposed deletion,Wikipedia:Drafts, andWikipedia:Merging to get rid of articles they believe are inappropriate. I wouldn't bother having that discussion at all.

For your first question, why don't you start a new ==Topic== at the bottom of this page? You might want to look at first are how many AFDs we see in a given week, or how many are usually from the same person. For example,Wikipedia:Articles for deletion/Log/2025 August 20 has 79 discussions and 752 comments. About 18 of them are relisted discussions from previous dates. So if you said there was typically 60 new nominations each day, there would be 420 new nominations each week. Maybe 20 separate nominations per week (less than 5%) is just fine. Maybe even 10 per day would be fine.WhatamIdoing (talk)20:12, 6 September 2025 (UTC)

I would also support a rule against flooding. To refine the rule I think we should keep two kinds of flooding in mind.

1. Flooding the number of AFD. To determine this limit I suggest we use a data driven approach where we run a query to determine the number of AfD it is appropriate to have at once. Specifically, I think for each week of the past five or ten years we should count the number of AFD opened by each user who opened an AFD. Then for each year we plot a histogram of the number of AFD opened in a week by each user. These should be compared to observe any significant time series effects. If a time series effect is seen it may need to be investigated on a month by month basis. Then combine all the data together into a sample histogram and determine the 99%-tile of the histogram. Non admins should be limited to opening less than the 99%-tile number of discussions in a week. To enforce this a bot could speedy keep AFD noms that exceed the limit.
2. Flooding in an AFD. In this scenario an editor uses an LLM to make extremely long winded points causing them to dominate the conversation. To prevent this, I think non-admin users commenting on an AFD should be character limited once the length of the discussion has exceeded 10,000 characters. The limit should be also data driven by checking the distribution of number of characters per user in the past year's AFDs. Also set around the 99%.

If flooding is still seen as a problem after these limits then the percentile can be lowered.Czarking0 (talk)23:48, 6 September 2025 (UTC)

If you're able to do the analysis for #1, then I believe you can get the data viaWikipedia:Request a query. It sounds like you'll just need the username for the editor who created the AFD subpage + the creation date.WhatamIdoing (talk)00:28, 7 September 2025 (UTC)

It's worth noting that absolute number of AfDs isn't the only concern regarding flooding, but also the number of concurrently-open AfDs about similar subjects, e.g. If someone has opened 5 AfDs each day this week about African sportspeople active in the 1950s, all of which are still open, then another 5 AfDs on the same subject is potentially problematic as the number of people interested in and able to look for sources etc is limited and they only have a finite amount of time in which to do so. In contrast that same person opening 10 AfDs about North American business people active in the 1990s is unlikely to be problematic as the editors involved with those articles are unlikely to significantly overlap with those dealing with the sportspeople. Whether this is something that can be even investigated algorithmically I don't know.Thryduulf (talk)08:32, 7 September 2025 (UTC)

I don't think a 99% limit is necessarily ideal. In a case where AfD nominations and character counts per AfDs are reasonable (no flooding), then that cutoff would still penalize 1% of (non-admin) editors, while the same amount would be penalized after a year with massive AfD flooding issues. Also not sure about limiting this to non-admins specifically – besides the person closing the discussion, admins shouldn't have more power than non-admins as participants in the discussion.ChaoticEnby (talk ·contribs)13:43, 7 September 2025 (UTC)

I think to address your first concern the community could, after seeing the data, determine a consensus cutoff (maybe it is 99.9%) that automatically expires after one year. The cutoff could be reanalyzed the next year and if the same value is chosen the community could opt to implement the cutoff for two years.

I think it is wise to not limit admins because there are niche cases where a lot of articles need to be deleted and admins are trusted users who if they engage in flooding have additional mechanisms to be held accountable like ARECALL.Czarking0 (talk)17:06, 7 September 2025 (UTC)

I think that it's reasonable to calculate a 99% limit, but if we think the number is unreasonably low, then we should probably pick a higher number. We can start with some basic descriptive statistics and use our judgment and experience to figure it out from there.

@Czarking0, can you do this kind of analysis?WhatamIdoing (talk)17:41, 7 September 2025 (UTC)

Can I? Yes. However, I am not planning to unless there is somewhat wide agreement that it will be used.Czarking0 (talk)17:46, 7 September 2025 (UTC)

If RAQ will give us the data, and you will do the analysis, I'll use your numbers to write an RFC to adopt an anti-AFD-flooding rule. Does that sound like a fair plan to you?WhatamIdoing (talk)17:48, 7 September 2025 (UTC)

In principle yes, but for me think an RFC is not a waste of time I would at least need to see that at least half the people in this thread agree. Things are not going to get implemented if just you and I agree on something.Czarking0 (talk)23:42, 7 September 2025 (UTC)

Trying to form a data driven rule is a good plan.Cdr. Erwin Smith (talk)11:44, 10 September 2025 (UTC)

Hi, the first sentence of articles should contain definition, i.e. only "highly critical" data about a concept. But in many Wikipedia articles likeNeural network,Saadi Shirazi orMedina there exist many synonyms in the first sentence which some of them are not considered "highly sensitive". So to improve readability of articles, I propose to

• Write the first sentence with the most common synonym and remove any non-sensitive synonyms
• Add a new sentence at the end of lead section for other synonyms

In my opinion, this style improves readability of Wikipedia articles significantly. Please discuss. Thanks,Hooman Mallahzadeh (talk)07:42, 13 September 2025 (UTC)

Personally, I struggle to imagine a case where a synonym is common enough that it should be mentioned in the lead butnot common enough that it shouldn't be at the very start for readers looking for that term. Any reasonably common synonyms should be included in the first sentence so readers know they've arrived at the correct article (this seems like highly critical info for readers). If a synonym is not common, it shouldn't be in the lead.RunningTiger123 (talk)18:12, 13 September 2025 (UTC)

There are examples where synonyms are mentioned later on in the lead, examples include synonyms of/alternative terms for concepts/things that are introduced later on in the lead or terms that were historically more important than they are today. For example atBoombox the synonym "ghetto blaster" is bolded in the middle of the second paragraph of the lead where it is presented in context. I do fully agree with your main point though that most complete synonyms should be towards the beginning of the lead to maximally help readers.Thryduulf (talk)19:24, 13 September 2025 (UTC)

Good point, but yeah, it's still moot – those examples shouldn't be moved to the end.RunningTiger123 (talk)01:33, 14 September 2025 (UTC)

I have some sympathy for this. If there are several synonyms, it would often be better to follow this advice.

But "Last sentence" is too specific, and it doesn't work for mobile. Remember that the layout on mobile is:

• first paragraph
• infobox
• rest of lead (if any)

so if someone is reading an article with a two-paragraph lead, they have to scroll all the way past the entire infobox before they will be able to see the synonyms – and that means sometimes having to scroll and scroll and scroll before they will be able to figure out whether they are in the right article.

Looking at your three examples, I think that there are multiple ways to improve them. DoesNeural network (machine learning) really need the "neural net" abbreviation? I doubt it. InSaadi Shirazi, some of his names could be moved to the infobox (e.g.,|native_name=). InMedina, one of the older names is mentioned in the relevant historical context in the second paragraph, so that could be removed from the first, and perhaps a similar arrangement could be made for Taybah, which is also a historical name.WhatamIdoing (talk)20:25, 13 September 2025 (UTC)

What about an explanatory footnoteMOS:LEADCLUTTER.Moxy🍁01:37, 14 September 2025 (UTC)

Readers should not need to look in a footnote to be able to determine whether they have arrived at the right page.Thryduulf (talk)02:10, 14 September 2025 (UTC)

The title of the page should have the most common name so it shouldn't be too much confusion. I think readability should be a concern here.Medina in my view would be a good example of first sentence clutter..... simply a screen reader nightmare with all those footnotes instead of one foot note that explains everything.Moxy🍁02:33, 14 September 2025 (UTC)

The title of the page should have the most common name that's fine when there is a single common name that applies globally to the whole topic. As your example of Medina shows that's not always the case. Combining those footnotes, each of which just give the Arabic transliteration and sometimes literal translation, into one would be significantly more confusing for a non-English speaker.Thryduulf (talk)09:47, 14 September 2025 (UTC)

If there is some sort of exceptional cases....I am thinking that the first sentence should describe things quickly.....as in...Medina is the capital and administrative center of Medina Province in the Hejaz region of western Saudi Arabia. And the second sentence deals with specifics....likeOfficially al-Madinah al-Munawwarah, the city is also known as Taybahnand and in pre-Islamic times as Yathrib, it is one of the oldest and most important places in Islamic historyMoxy🍁15:12, 14 September 2025 (UTC)

That simplification seems more fitting for the Simple English Wikipedia than it does for the standard English Wikipedia. For instance, insimple:Medina, the first sentence involves fewer variations of the name and excludes the footnotes, as you are requesting.

Medina (/məˈdiːnə/;Arabic:المدينة المنورة,al-Madīnah al-Munawwarah , "the radiant city"; orالمدينة,al-Madīnah (Hejazi Arabic pronunciation:[almaˈdiːna]), "the city"), also transliterated asMadīnah, is a city in theHejaz, and the capital of the Al Madinah Region ofSaudi Arabia.

In general, the policy is to have a separate section to discuss the varying terminology if it is sufficiently complex or contentious, but we generally want to keep the common names at the forefront so that people know what the article about. When it comes to the names of cities, countries, or regions, this can be particularly contentious as there may be political reasons for preferring one name over another, and disputes around that are a common reason for edit wars. So specifying the different terminology can also be important for quelling discontent amongst the editors.Ovenel (talk)16:20, 15 September 2025 (UTC)

This sounds like it is sometimes worth doing, while it sometimes should not be done. Best not to mandate what to do too precisely. For an example with famously many synonyms, seesoft drink, where all of[21] is relegated to a Terminology section. On the other hand,trunk (car) should certainly have "boot" in bold in the lead section, to make sure Brits notice they are at the right page. —Kusma (talk)13:11, 15 September 2025 (UTC)

There are situations whereWP:MOS specifies a format or where a format is consistently used in an article, but data from an external source are in a different format. It would be convenient to have tools in the wikisource and visual editors to transform those data to a canonical form. Examples are:

• Transform selected date to article default or to explicitly specified format.
• Translate upper case title to sentence case or title case

The transforms need not be perfect; even if manual tweaking is needed, they would still save time. --Shmuel (Seymour J.) Metz Username:Chatul (talk)12:23, 3 September 2025 (UTC)

The{{Date}} template performs transformations of dates.Template:Case templates table lists some case-transformation templates. Sentence case and title case are not in this list, though. It's not hard to make a simple version work most of the time, but there are exceptions (like dealing with upper case abbreviations) that would be hard to handle without maintaining an exhaustive list of them.isaacl (talk)16:03, 3 September 2025 (UTC)

I was thinking of editing tools rather than templates. A tool for title case shouldn't require manual fixup of more than a few letter except where there is an initialism. A tool for sentence case would only need manual fix-up for initialisms and proper nouns. --Shmuel (Seymour J.) Metz Username:Chatul (talk)20:41, 3 September 2025 (UTC)

Sorry, when you said data from an external source, I was thinking it would be transcluded into the article through some mechanism, and so passing it to a template for formatting would fix it up automatically.isaacl (talk)21:57, 3 September 2025 (UTC)

Maybe take a look atAWB andAutoEd —GhostInTheMachine^{talk to me}06:47, 4 September 2025 (UTC)

The two tasks I'm most concerned about are

• Convert a selected date to the appropriate format for the article, including
  • Month Day, Year
  • Month Year
  • mm/dd/yyyy
  • ISO
• Change the case of a selected string to
  • Upper
  • Lower
  • Sentence
  • Title

The two tool mentioned look usefull for bulk edits, but not for those two tasks. --Shmuel (Seymour J.) Metz Username:Chatul (talk)12:11, 4 September 2025 (UTC)

In all the cases you mention, changes must be made with care and human attention because context matters massively. There are many examples where articlesshould contain mixed usage of dates/cases, including articles discussing different formats, quotes, and instances where case changes the meaning of a term.Thryduulf (talk)16:12, 4 September 2025 (UTC)

For date conversion,substituting the{{Date}} template is a workaround you can use. Unless someone implements a case-changing feature, the only workaround I can think of is to use your favourite external text editor's case-changing feature. (Google Docs, for example, has a capitalization sub menu under Format → Text.)isaacl (talk)16:56, 4 September 2025 (UTC)

This, frankly, is the use case for developing some form of AI-assisted editing, to find and prompt edits for likely instances of overcapitalization and incorrect date presentation.BD2412T17:03, 4 September 2025 (UTC)

The dates I'm concerned with are in correct formats but not consistent with other dates in the same article. --Shmuel (Seymour J.) Metz Username:Chatul (talk)18:34, 4 September 2025 (UTC)

The documentation forTemplate:Date saysThis template should only be used internally in other templates.

Yes, I sometimes use an external editor for long uppercase titles; I copy them into Tritus SPF (TSPF), use the lower case (LC) line command, over-type individual letters that should be capitalized, insert hyphens to indicate explicit line breaks, use the text flow (TF) line command to remove extraneous line breaks, remove extraneous hyphenation and copy the text back. --Shmuel (Seymour J.) Metz Username:Chatul (talk)18:34, 4 September 2025 (UTC)

By substituting the template contents, the transformed results are saved to the page, so you won't actually be using the template within the wikitext source.isaacl (talk)23:47, 4 September 2025 (UTC)

@Chatul, I'm not understand what you're trying to accomplish. I see your suggested solution, but what's the specific problem itself? For example, here are two different stories:

1. As aWikiGnome, I'd like to tweak the existing article content so that the dates and capitalization are correct. It would save me time and effort if I could selectAN ALL CAPS TITLE in an existing citation template, click a button in the toolbar, and have it transformed into sentence case or title case (whichever is appropriate for the individual article's style).
2. As an article creator, I frequently add citations to articles. It would save me time and effort if automated citation tools such asCitation bot ormw:Citoid would correct the capitalization when automatically generating the citation.

Your story might be different, of course, but it'd help if you gave me something concrete to think about.WhatamIdoing (talk)17:26, 4 September 2025 (UTC)

My specific problem is the second: creating a citation from an online PDF document, doing cut-and paste for, e.g., date, title. I normally use the source editor rather than VE. --Shmuel (Seymour J.) Metz Username:Chatul (talk)18:34, 4 September 2025 (UTC)

(PDFs are the worst.) If the date is a common form, and you're using a CS1 citation template, then they'll display the date in the correct format. That is, if the article has{{use dmy dates}}, but you paste "2025-09-01", it will display "1 September 2025" to the readers. Eventually, someone running AWB will "correct" the wikitext, but you don't usually need to worry about the dates.

Capitalization can't be handled that way. You may remember participating in theRFC about preserving mismatched "found state" capitalization (which concluded that mismatched capitalization does not deserve the protection ofWP:CITEVAR). But you may have forgotten that someone in that discussion said there is auser script that converts to title case. I understand that it's fairly easy to turn a user script into a button inside the 2010 wikitext editor. Maybe you could ask at VPT for someone to do that?WhatamIdoing (talk)20:28, 4 September 2025 (UTC)

User:ZKang123/TitleCaseConverter puts a menu item in the "Tools" menu that I believe will make changes to the edit box for the wikitext editor (not sure if it will work with Visual Editor). (I see in the source that it has a list of manually maintained exceptions, as I mentioned would be necessary.)isaacl (talk)23:58, 4 September 2025 (UTC)

The code for inserting a button in the 2010WTE toolbar vs in the VE toolbar is different, so I always assume that a script will not work in both. So far, my assumption has never been proven wrong.WhatamIdoing (talk)02:17, 5 September 2025 (UTC)

Sure; I wasn't commenting on how to add a button to the toolbar. Since the original poster typically uses the wikitext editor, they might find the current user interface sufficient.isaacl (talk)02:22, 5 September 2025 (UTC)

Until VE is ready for prime time, I'm only concerned with a tool for the source editor. --Shmuel (Seymour J.) Metz Username:Chatul (talk)12:12, 16 September 2025 (UTC)

I'm looking for something that will operate only on the selected text or only on the clipboard. There are too many edge cases to trust a global case fix-up script as opposed to a limited scope script with human sanity checks. --Shmuel (Seymour J.) Metz Username:Chatul (talk)12:12, 16 September 2025 (UTC)

If you're not comfortable writing Javascript, then maybe someone can create a variant of ZKang123's script for you.isaacl (talk)16:41, 16 September 2025 (UTC)

I've seen a few times over the last 6 months that accounts with far less than 25 edits (not sure whether more or less than 10) have moved other users base user and user talk pages. Should that need more rights/ more edits?Naraht (talk)23:48, 7 September 2025 (UTC)

Could you link to an example? Generally a user should NEVER move another user’s base page or talk pages… at least not without an explicit request to do so. However, there could be extenuating circumstances that we are unaware of.Blueboar (talk)23:54, 7 September 2025 (UTC)

Well,here's the last thousand third-party base user/user-talk page moves, which goes back to February. A bit more than a hundred of them were by users who currently have 100 or fewer edits, and are mostly problematic; the remainder were overwhelmingly moves into draft, with a scattering of renames. (quarry:query/96990 has the last thousand by users currently with 200 or fewer edits; those thousand go back almost a decade.) —Cryptic01:59, 8 September 2025 (UTC)

Interesting… the vast majority of these seem to be moves from USERspace to DRAFTspace… ie editor “A” has created a potential article in their userspace, and editor “B” has moved the potential article to draftspace. I am of mixed opinion on this.Blueboar (talk)12:19, 8 September 2025 (UTC)

The most common reason for that is "theDraft namespace is the preferred location forArticles for Creation submissions", or some variant. I don't mind this part of the AFC workflow, and I presume that the vast majority of these were tagged by their creator with an AFC submission template, indicating consent to move forward with the AFC process. This specific issue should not weigh too heavily on Naraht's question about brand new editors, since they will not likely be making these AFC-oriented moves.Firefangledfeathers (talk /contribs)14:02, 8 September 2025 (UTC)

I am actually surprised that this isn't already a thing. We already have the similarSpecial:AbuseFilter/803, which prevents non-autoconfirmed users from editing (most) user pages, as well asSpecial:AbuseFilter/733 which prevents them from creating such pages. Moves out of userspace can also be more difficult to clean depending on what happens to the leftover redirect, so it makes sense to have a similar edit filter for these. Not sure whether the restriction should be AC, XC, or something in-between like 733 does.ChaoticEnby (talk ·contribs)18:18, 8 September 2025 (UTC)

The idea of duplicating 733, except for moves rather than creation, is appealing. If you can't create that page, you probably don't need to move it (or to be able to move a page into someone's User: namespace).

The only time I would expect a newer editor to move a User: page is if they have renamed their account, but perhaps that's part of an automated system/already being done for them? There are a number of "Automatically moved page while renaming the user" edit summaries in Cryptic's query (none by editors with fewer than 100 edits).WhatamIdoing (talk)00:17, 9 September 2025 (UTC)

On-rename pagemoves are logged to the global renamer, not to the renamed user. It's possible that a global renamer might have <100 edits here, but I don't think a filter would stop them, given thateven blocks don't stop global renamers. As I understand it, renames are really executed by a server-side script, which spoofs the username of whoever pressed the button on Meta. --Tamzin^{[cetacean needed]}(they|xe|🤷)09:13, 9 September 2025 (UTC)

That's not entirely correct. Abuse filters have a history of interfering even with server-side maintenance scripts (eg.phab:T368275). It's likely it disrupts renames as well –phab:T69936 suggests this may be the case. –SD0001 (talk)17:54, 11 September 2025 (UTC)

Isn't 803 already taking care of this? The difference is that the OP above has picked a new threshold. 803 prevents unconfirmed, not "<25 edit" users. —xaosflux^Talk10:12, 9 September 2025 (UTC)

Might be, I wasn't sure if AbuseFilter logged page moves as edits to the base page name.ChaoticEnby (talk ·contribs)13:27, 9 September 2025 (UTC)

In general, anything stopping an "edit" should stop a "move". If there is an example of it not working, we can certainly check on that. —xaosflux^Talk13:46, 9 September 2025 (UTC)

XaosfluxAs OP, I don't have a particular idea in mind for threshhold. (other than >0) And I believe that the user did it before reaching confirmed status, which is sort of the bottom rung.— Precedingunsigned comment added byNaraht (talk •contribs)18:13, 9 September 2025 (UTC)

On others userspaces, I don't think new users should be moving things so I would support auto-confirmed restriction. I think this would require change of what the non-autoconfirmed user right can do which might require strong supporting evidence. If implemented, all the new user will need to do is to talk with the owner of the userspace, which is good because it trains them for conflict resolution. I do not support this within own userspace 🇪🇭🇵🇸🇸🇩Easternsahara 🇪🇭🇵🇸🇸🇩23:45, 16 September 2025 (UTC)

• https://en.wikipedia.org/w/index.php?title=User:Curbon7&diff=prev&oldid=1309987836
• https://en.wikipedia.org/w/index.php?title=User_talk:Curbon7&diff=prev&oldid=1309987839

CheckingDicenMooX6 has since been banned. And I'm fine with this being done by an experienced user, I don't have a good guess right now as to what userrights should correspond to it, so it seems ideal to bring it up here.Naraht (talk)13:57, 8 September 2025 (UTC)

I’ve been trying to find a wiki-friendly way to make a user group (Not the kind inWikipedia:User groups that manages permissions) between me and some of my friends/fellow editors. We all edit different pages, and I’d like to be able to help them on their drafts, and vice versa. Would this just be some semiofficial thing done in someone’s subpages?PhilDaBirdMan (Talk |WikiProject Socialism |Current Incubator Initiative)11:49, 16 September 2025 (UTC)

If there's a specific topic you could find a relevant WikiProject, if it's just casual collaboration you can do that in any participant's userspace.CMD (talk)11:54, 16 September 2025 (UTC)

Thanks for the help.PhilDaBirdMan (Talk |WikiProject Socialism |Current Incubator Initiative)12:00, 16 September 2025 (UTC)

Just about anything you do is OK, as long as you are all transparent and above board about it. You should avoid any hint ofmeatpuppetry.Phil Bridger (talk)12:05, 16 September 2025 (UTC)

Okay, sounds good.PhilDaBirdMan (Talk |WikiProject Socialism |Current Incubator Initiative)12:50, 16 September 2025 (UTC)

If it's a small group, then just posting on each others' User_talk: pages is usually enough. You are allowed to make a subpage in your User: space, if you all prefer that.WhatamIdoing (talk)04:17, 17 September 2025 (UTC)

Thanks. I’ll see if that helps.PhilDaBirdMan (Talk |WikiProject Socialism |Current Incubator Initiative)11:25, 17 September 2025 (UTC)

Anecdotally speaking I feel like I am seeing a nontrivial amount of "newcomer tasks" -- particularly the copyediting and expand tasks -- that result in large quantities of dubious AI-generated edits being pumped out without time for review. Seethis thread for a recent example. May be a case where the gamified interface kind of encourages such things to happen.

That being said I have no proof whatsoever besides "I've seen this before and it feels intuitive to me." I don't know whether it's possible to view such tagged edits en masse, if so that would help look into the issue more. (Might be a permissions issue?)Gnomingstuff (talk)15:38, 17 September 2025 (UTC)

I'm sure that there is a more elegant way with a Quarry query, buthere isSpecial:RecentChanges filtered bynewcomer andlearner users, and edits that have been tagged as#newcomer task of various sorts. Orhere with only the copyediting and update tasks that you mentioned. Cheers,SunloungerFrog (talk)16:07, 17 September 2025 (UTC)

PingingUser:KStoller-WMF.

Maybe the instructions need a "Don't use AI" warning?WhatamIdoing (talk)17:17, 17 September 2025 (UTC)

Thanks for the ping! I’m the Product Manager for the Wikimedia Foundation’s Growth team, which develops and maintains newcomer tasks, so I can help link to some more stats:

• The WMF Growth team maintains a dashboard with stats on newcomer tasks:here's a filtered view for English Wikipedia here.
• You can also see how many suggestions are available for each newcomer task type:Special:NewcomerTasksInfo
• Administrators can make adjustments to how Newcomer Tasks (Suggested Edits) work through Community Configuration:Special:CommunityConfiguration/GrowthSuggestedEdits
• If you are interested in seeing all of this from a newcomer perspective, you can see Suggested Edits here:Special:Homepage

At present, the Copyedit task and the Link Recommendations (Add a Link) task are the default “easy” tasks available to newcomers immediately after account creation. Because these tasks are often a first step, many of the people completing them are brand new to Wikipedia and therefore more likely to make mistakes.

I recognize that there is an inherent tension between lowering barriers to entry for new editors and the additional review work this can create for patrollers. My hope is that the increase in newcomer edits will ultimately support the long-term health of the community by helping more people learn Wikipedia’s norms and develop into active editors and patrollers over time.Recent A/B testing on English Wikipedia supports this, showing that the "Add a Link" newcomer tasks encourage more new account holders to try editing and remain engaged.

I’m also open to suggestions on how we can reduce the impact of these edits while continuing to create a welcoming and supportive environment for newcomers. We can certainly consider adjustments to onboarding, though I do worry that warnings are often overlooked and may not meaningfully influence newcomer behavior. Thoughts? -KStoller-WMF (talk)22:30, 17 September 2025 (UTC)

The likely problem is not Copyedit or add links, but "Expand short articles". We have multiple cases of editors sprinting through the relevant tags by just running everything through an AI, and all the associated issues this causes. "Expand lead" in general is probably a poor task for newcomers, as they are unlikely to be familiar with our expectations for Leads and it is also the most visible part of the page.CMD (talk)00:11, 18 September 2025 (UTC)

I wonder if that one can be replaced with update data, for sentences tagged with update needed. Editors have an example of a good source (hopefully) that's dated, and are asked to replace it with something that's more up-to-date. Feels like a logical next step after link adding and copyediting.—Femke 🐦 (talk)07:06, 18 September 2025 (UTC)

It seems like there's a rate-limiting feature? One solution might be to limit to 1-2 expansion edits per day.

Better yet, I don't know how technically feasible this is, but maybe a rate-limiting feature by time; the expand task says it should take about 20-30 minutes or so (I forget the exact numbers), but some users are pumping these out within a handful of minutes.

Might be good for copyediting as well as people do pump out AI copyedits quickly too (AI "copyedits" have a tendency to introduce errors, insert editorializing, etc.;one example), but then you'd need to distinguish between changing one typo or something else that legitimately can be done fast, and a full quasi-line-edit.Gnomingstuff (talk)17:06, 18 September 2025 (UTC)

@Gnomingstuff, @CMD, @Femke: on the topic of potential mitigations, I wonder: to what extent (if any) do you think the work the Editing Team is doing onPaste Check could be helpful here?

In essence, Paste Check will introduce a prompt that asks people to confirm whether they wrote the text they are pasting on-wiki. To start, the Check isconfigured to activate when newcomers (≤100 cumulative edits) paste ≥50 characters of text into a VE edit session.

The Check wasoriginally intended to help avoid and fight copyright violations. Although, this discussion [i] (and others that have emerged off-wiki) is causing me to wonder whether the Check could be helpful in the context of combating AI-generated edits as well...

---

i. Thank you for drawing my attention here, @KStoller-WMFPPelberg (WMF) (talk)22:22, 18 September 2025 (UTC)

Hard to say, would have to see it in action. My initial thought is many people will just answer yes no matter what, either to just click through or because of "well I wrote it I just used AI to rewrite it."Gnomingstuff (talk)13:44, 19 September 2025 (UTC)

Maybe add a specific option for "I only used AI to rewrite it", as that seems to be a pretty common explanation people give for using AI? Also not sure but that could definitely deserve to be tried in practice. Beyond that, I absolutely agree with @PPelberg (WMF) that PasteCheck can definitely help – the only case that might be missed is some AI-powered browsers like Perplexity's directly writing the AI output in the context window.ChaoticEnby (talk ·contribs)15:28, 19 September 2025 (UTC)

The paste check definitely seems worth testing! Would be curious what its effects are, even though we might need to some manual checking if it stops clear AI slop.—Femke 🐦 (talk)19:25, 19 September 2025 (UTC)

It will probably help a little, although I don't know if you will be able to measure that (can you measure who on Commons doesn't click own work and later aborts an upload?)? Asking people not to paste is good for many reasons, so I like the idea, however small the impact.CMD (talk)15:10, 19 September 2025 (UTC)

@SunloungerFrog Awesome, this is exactly what I was looking for. To be clear, I don't know how much of a problem there is yet, being able to see the full feed will help.

The good news is that at least according to the above link there seem to be a manageable amount of edits per day to at least go through and flag if necessary. I may end up eating those words but who knows.

The bad news: It took me about 5 minutes to find two edits (this andthis) by the same user that I am pretty sure were AI generated, especially given that they were made only a few minutes apart despite being substantive rewrites. Of course it's theoretically possible that they did the rewrites and post them one after the other, but the UX of the dashboard seems to make that unlikely since it presents articles one after the other.Gnomingstuff (talk)23:25, 17 September 2025 (UTC)

Agree that this is a concerning issue, I've cross-posted it toWikipedia:Growth Team features to maybe get people more familiar with the interface.ChaoticEnby (talk ·contribs)21:53, 17 September 2025 (UTC)

And on the topic of data that couldhelp [i] us assess the prevalence of this issue, the Editing Team is planning an analysis to estimate how frequently newcomers (≤100 cumulative edits) paste ≥50 characters of text into a VE edit session. That work is scheduled to happen inT403861.

I'll report back what we learn. Of course, if this brings questions to mind in the meantime, I hope you'll share.

---

i. Emphasis on "help" seeing as how this metric does not equate precisely to what you're seeking to learn here: prevalence of people publishing AI-generated content onto the wiki.PPelberg (WMF) (talk)22:35, 18 September 2025 (UTC)

Would it be possible to evaluate the variation in prevalence between newcomer tasks and other edits?ChaoticEnby (talk ·contribs)15:30, 19 September 2025 (UTC)

@Chaotic Enby Do you mean assessing whether AI-generated edits occur more frequently through Newcomer Tasks compared to other edits made by newcomers? That might be difficult to determine with precision, since it is not always clear whether an edit is AI-generated. What I can do is pull data on revert rates to see if there are notable differences. Would that type of comparison be useful? Or do you have ideas for how we might evaluate the variation in prevalence of AI-style edits between the two groups? -KStoller-WMF (talk)16:27, 19 September 2025 (UTC)

I was thinking of PPelberg's methodology of seeing whether large pastes are more often done in newcomer tasks compared to other edits. The issue I can foresee is that some editing tasks (like adding wikilinks) will by nature be less likely to require large pastes, so we have some confounding variables that need to be considered. Picking tasks like copyediting or expansion could be more insightful in that regard.ChaoticEnby (talk ·contribs)16:32, 19 September 2025 (UTC)

Thanks for the clarification! I'm not sure if an evaluation like that is planned, but it seems feasible especially if Paste Check triggers an edit tag. @PPelberg (WMF) - Will a specific tag be added to completed edits that triggered a Paste Check?KStoller-WMF (talk)23:21, 21 September 2025 (UTC)

I've read some articles, discovering that when mentioning amass shooting, the identity of the perpetrator or a superlative is often unnecessarily mentioned. For instance, inMarion Hammer the name of the perpetrator ofStoneman Douglas High School shooting is given, and some social scientists think that giving the name of the perpetrator when reporting a mass shooting causescontagion. Nowadaysmass shootings in the United States are more lethal than in the 1970s, and at that timegun control there was more lenient, and civilian gun ownership was more ubiquitous, meaning news media, social media platforms, social media users and Wikipedians are much more responsible for the current mass shooting epidemic than theNational Rifle Association. So I suggest that such information usu. should be unmentioned. Even if it must be given, it should be as de-emphasized as much as possible - and not repeated.RekishiEJ (talk)08:23, 17 September 2025 (UTC)

I have no idea how muchWP:RS considers the NRA responsible for the "current mass shooting epidemic" (according toMass_shootings_in_the_United_States#Frequency, "epidemic" is not without some basis). Is the argument "There are many mass shootings because there are many guns, and NRA is part of the reason there are many guns."?

Anyway,Wikipedians are responsible for the current mass shooting epidemic.^{[citation needed]}Gråbergs Gråa Sång (talk)15:19, 17 September 2025 (UTC)

Please remember that we are not here toright great wrongs, but instead to present a summary of the knowledge of the world from aneutral point of view. There are good reasons for restricting when and how to nameliving or recently deceased persons in WP, including privacy, avoidance of libel and defamation, promotionalism, relevancy, proportion, etc., but I doubt that WP is inspiring new mass murders by mentioning in an article the name of a mass murderer that is widely publicized in all kinds of sources.Donald Albury15:57, 17 September 2025 (UTC)

There does seem to be a desire among certain people/sources to engage in adamnatio memoriae exercise: He is dead to me; do not speak his name in my presence or acknowledge that he ever existed.

We also know that many (but not all) mass shooters are mentally unstable; we know that many of the younger ones are obsessed with school shooters; we know that many of them hope to be (in)famous.

When you combine these two facts, it is easy to think that we can thwart the next shooting by accidentally-just-happening to not mention the perpetrator's name. But this won't work: obsessed people don't read just the Wikipedia article(s), so omitting the name here will not keep them ignorant of the perpetrator's name.WhatamIdoing (talk)17:14, 17 September 2025 (UTC)

I know, but by not mentioning the perpetrator's name so frequently can reduce the infamy the man gets, thus mitigating mass shooting contagion. And there is evidence that some people think (thought) that NRA is partially responsible for the current American public safety problem (cf.[22]).--RekishiEJ (talk)04:52, 18 September 2025 (UTC)

I don't think that Wikipedia is that influential. An obsessed person may read thousands of webpages. They might read whole books. It's not realistic to think "Wow, they're super obsessed – but maybe if one (1) website mentions the perp's name a bit less often, that subtle stylistic change will really change the way they think about the whole subject".WhatamIdoing (talk)22:52, 18 September 2025 (UTC)

This may seem off topic, but I am reminded of the time that the board of directors where I worked decided to fight inflation by keeping annual raises for staff below the inflation rate. That was also like trying to weaken a hurricane by spitting into the wind.Donald Albury23:41, 18 September 2025 (UTC)

They are also much more likely to watch a YouTube video and read reddit —TheDJ (talk •contribs)11:42, 19 September 2025 (UTC)

Wikipedia followsSecondaryReliable Sources unlike leading media practices like flashing 'Breaking News' to sensationalize. Neither does it cite 'Social Media'WP:RSPFB/WP:RSPX nor does it cite itselfWP:RSPWP in any of its articles.

We giveWP:DUE weight to every aspect in an article. So If the consensus of reliable sources shifts to de-emphasize perpetrators’ names, Wikipedia will likely follow suit.

Revoking civilian rights on using lethal firearms will be much more effective against mass shootings than censoring some information on Wikipedia.Cdr. Erwin Smith (talk)19:32, 18 September 2025 (UTC)

@Cdr. Erwin Smith:You're wrong, since a study shows that in America highly lethal mass shooting were due to the shooters' desire to be infamous, and social and news media were the only reasons that they fantasized about committing more lethal events (cf.[23]).--RekishiEJ (talk)04:06, 19 September 2025 (UTC) changed a link 04:14, 19 September 2025 (UTC)

What's more, a study shows that news media actually cause mass shooting contagion by making mass shooters very noted - even more noted than some professional athletes (cf.[24]).--RekishiEJ (talk)04:09, 19 September 2025 (UTC) altered a bit 04:11, 19 September 2025 (UTC)

When did I deny media houses influencing such events? I simply stated a fact that Wikipedia doesn't broadcast prime time 'breaking news' in flashy colors.

It has a boring black and white design. It doesn't spoon feed attractive audio-visual content like TV and forces you put effort and read it manually. Even in this process, lion share of the source articles aren't put in to reduce bloat and keep it concise. So Wikipedia doesn't come anywhere close to the media houses in influencing such events.

I think you're at the wrong place. You have 2 options that I can suggest :

• Go to the court and demand such censorship on the news media. Wikipedia will automatically follow suit because it cites these articles.

[This will address your temporary concern]

• Go to the court and demand revoking civilian rights to lethal firearms.

[This will permanently eliminate the evil]Cdr. Erwin Smith (talk)08:52, 19 September 2025 (UTC)

I think both RekishiEJ and Cdr. Erwin Smith are straying intoWP:NOTFORUM territory here. Any US citizens among you can vote appropriately to further your views on gun control in the US, but here we can only influence what happens on Wikipedia. It is perfectly possible (but not our concern here) that both the media and the NRA are partly responsible for mass shootings.Phil Bridger (talk)09:11, 19 September 2025 (UTC)

Agreed. Had to step a bit beyond the line to show him a proper path since he clearly couldn't understand that Wikipedia isn't the root cause of this problem.Cdr. Erwin Smith (talk)10:18, 19 September 2025 (UTC)

Yet Wikipedia is a very popular website, meaning its information about a particular mass shooting influences a lot of people, including would-be mass shooters (e.g.Nikolas Cruz).--RekishiEJ (talk)07:52, 20 September 2025 (UTC)

This is a singular case. A few cases like these doesn't mean Wikipedia is the root cause of this problem. A perpetrator doesn'tjust read Wikipedia and decides to go on a rampage. The confidence gained from the destructive potential of a firearm remains unmatched in influencing his ultimate descision. AKnife or aBaseball Bat can never inflict so much damage, thus can never make him that famous.

Even then, since the loss of lives is too many even within a few cases like these,I can suggest taking the criminal's name just once at the beginning of such articles, and using pronouns like 'He', 'Him' or adjectives like 'The perpetrator', 'The convicted', 'The guilty' for the rest.

In connected articles likeMarion Hammer, only adjectives and pronouns be used instead of the perpetrators' names.

I will like to hear what @WhatamIdoing, @Phil Bridger, @Donald Albury, @Gråbergs Gråa Sång and @TheDJ think about an RfC on this proposal in VP Proposals?Cdr. Erwin Smith (talk)17:47, 20 September 2025 (UTC)

I think that an RFC is unnecessary.WhatamIdoing (talk)17:57, 20 September 2025 (UTC)

Agree with WhatamIdoing. The appropriate use of a perpetrators name is going to depend on circumstances.Donald Albury19:03, 20 September 2025 (UTC)

Can you elaborate?Cdr. Erwin Smith (talk)09:13, 21 September 2025 (UTC)

Wikipedia is not, and will never beWP:CENSORED for any reason, no matter how noble the causemgjertson (talk) (contribs)18:05, 22 September 2025 (UTC)

I know that Wikipedia is in the real world, so, in principle, we can do something, but in this case I don't think that any change is necessary.Phil Bridger (talk)19:35, 20 September 2025 (UTC)

I have to agree with the general consensus above. First, although this is somewhat outside the remit of our role as editors and continuing the WP:FORUM discussion above, the theory that the best response here is a strategy of "contagion" containment, to adapt Rekishi's vernacular, does make a certain amount of impressionistic and intuitice sense. Most pop psychology speculation does, or it typically wouldn't get the social/memetic currency to propogate through serious communities. But "seems to add up" is a world away from "empirically validated by robust research and earnestly suggested by a consensus of relevant experts." And this idea is very much the forme rand not at all the latter, as yet. Because of the complex variety of psychological pathologies, profiles, and backgrounds of mass and spree killing perpetrators, and the pace of development in areas dependent on a less absolute form of evidence-based research, it may be some time before there is a consensus.

On the other hand, much less speculative is the fact that the open societies of the modern world, including the large majority of the anglophone societies that this version of Wikipedia primarily provides assistance to, there is a clear conclusion that there are immense benefits from making the details off mass crimes transparent to the larger public. These are of course subject to some restraints for the sake of protecting against potential harms to victims and the falsely accused. But I do not think that there is anything in the same universe of general public consensus for the belief that the need to starve a certain attention-seeking criminal type is so great that we should abrogate freedom of information principles around the details of these serious crimes. I'll give you one of just about a thousand reasons why I think the public might be hesitant to adopt that standard:Alex Jones. There was no especially lasting or ardent attempt to prevent the identity of the perpetrator (or many of the horrific details) of the Sandy Hook massacre from becoming express and easily verifiable facts of public record. And yet it was still possible for many millions of people to be brought under the umbrella of misinformation about this supposedly being a fabricated event. How much easier for the bullshit peddlers does that job become once we collectively start to suppress details as pertinent as the names of perpetrators?

And again, that's just one of any number of reasons why the "starve them of oxygen" theory might not get traction with specialists and policy makers, to say nothing of gaining robust support through our cultures. But the more important point here is that we aren't even meant to be basing our approach here on such speculation. It suffices that, for the present time, open societies are continuing to make such information generally a matter of public record, and Wikipeida is expressly not supposed to be about trying to move the needle one way or another on such questions. We are very much a "try to accurately summarize the whole subject as far as we have information to relay" kind of resource, and when doing that job leads us to perceived ethical questions, we tend to try to emulate the larger institutions and prevailing norms. There is an overwhelmingly strong presumption on this project that we follow trends and usually try to avoid efforts of setting them, as this presents serious complications to our neutrality and reliability. Of course, both the expert consensus and the general received wisdom on this topic could easily change. Where we are now, however, I see no valid policy argument for obfuscating this particular type of information for that particular reason.SnowRise ^{let's rap}03:10, 21 September 2025 (UTC)

I mostly try to find a middle ground.

Somy proposal was aGolden Mean between complete clarity and complete obscurement. It took the name of the perpetrator once at the very beginning for the sake of clarity, and then obscured it through the use of negative adjectives which would portray the criminal in a bad light to demotivate any potential 'to be' criminal reading it.

Your thoughts on that?Cdr. Erwin Smith (talk)04:30, 21 September 2025 (UTC)

I don't think we need to go overboard with negative adjectives, beyond presenting facts as facts – adding them only to demotivate potential future criminals strikes me as close toWP:RGW, and likely againstWP:NPOV.ChaoticEnby (talk ·contribs)05:58, 21 September 2025 (UTC)

I don't think writing truthful negative adjectives about a person who did get convicted of committing negative deeds is 'righting' anything 'wrong'. Nor does it tilt the existing PoV in anyone's favour.

Adjectives and pronouns are used regularly in Wikipedia just like everywhere else in our day to day life. It's just that we would be using them 'more than normal'for the greater purpose of stoppingcontagion.Cdr. Erwin Smith (talk)06:49, 21 September 2025 (UTC)

I can appreciate the effort to try to thread the needle, but imo, any likely benefit from this is so speculative, ephemeral, and likely exceedingly thin even if actual (given our small slice of the picture in representing the overall image of a given perpetrator) that I can't view it as being established as useful to the extent that even justifies the potential lack of clarity that might come with referencing the individual pseudo-euphemistically. If we want to represent these people as psychotics, lowlifes, and undesirables, there is a policy-consistent way to do that: attributed statements from any of the very deep well of people willing to describe them as such every time one of these atrocities occur. I for one would very rarely find objection to lengthening content which accomplishes this.

But compromising the encyclopedic tone and factual clarity of our wikivoice prose does not strike me as the appropriate way to reflect the general censure and repugnance of the acts. That should come from the various corners of a society's collective voice, at which point, we faithfully report on it. And if there is any significant degree of validity to the idea that this kind of framing can influence potential future attackers, the idea that there is a general social judgment that the people who commit such acts are pathetic will carry more credence when represented in an encyclopedia article if we treat it like just another fact. Betraying an "angle" (however morally justified on the issue) through heavy-handed or emotive or even just structurally awkward prose probably will actually undermine that effort.SnowRise ^{let's rap}07:46, 21 September 2025 (UTC)

Personally, I believe human life is invaluable. It's the most important thing in the world.

So if Wikipedia is partially responsible for the loss of even 1 innocent life, we should take all possible measures within the policies to try and stop it from occurring again, to the best of our abilities.

Thus if your opinion gains a general consensus, I'll be all for it as well.Cdr. Erwin Smith (talk)09:11, 21 September 2025 (UTC)

That's a very strong claim, one that is almost certainly wrong and that requires aRS to have any credibility. However, I believe that we (TINW) should use terms such assuspect andaccused when referring to a specific person who has not been convicted, usingperpetrator only when referring to the potentially unknow person who committed the crime. --Shmuel (Seymour J.) Metz Username:Chatul (talk)10:44, 21 September 2025 (UTC)

I had a problem with the tool bar not appearing while using the source editor. When I checked my preferences I saw that it was enabled. This was happening with a really old Firefox and a more recent m$ edgy. On a hunch I tried Chrome and the task bar appeared as expected.

It would be helpful if wiki could display some sort of induction that some preferences require a different browser. This should appear when attempting to use the option, not just on the preferences pages. --Shmuel (Seymour J.) Metz Username:Chatul (talk)10:28, 21 September 2025 (UTC)

some preferences require a different browser - they should not though.Sapphaline (talk)12:38, 21 September 2025 (UTC)

Yeah, and under the hood Chrome and Edge are the exact same engine. Everything that works with Chrome should work the exact same under modern Edge. Regardless, if anything that would be a browser/web standard compatibility issue. There's no specific action the code is taking to block it from working in one browser or another.andritolion (talk)13:40, 21 September 2025 (UTC)

Browser support is described atmw:Compatibility § Browser support matrix. If a feature fails to work on one of the indicated browser versions, then you can report a bug to the corresponding support channel.isaacl (talk)14:27, 21 September 2025 (UTC)

I can't reasonably expect them to support, e.g, Firefox ESR 45.9.0, which is whatArcaOS currently ships with, but it would be nice to get an indication that a feature was disabled because of it. The edge in windows 10 is Chrome 140, the same as the Chromium where it works; do you have a wikilink for reporting the problem? Thanks. --Shmuel (Seymour J.) Metz Username:Chatul (talk)23:16, 21 September 2025 (UTC)

As Andritolion said, typically features aren't disabled. They just fail because, for example, they try to make some Javascript API function call that doesn't exist or behaves incorrectly in the older version. Accurately detecting why a script fails can be tricky, so while it would of course be nice to catch all failures and display an appropriate message, it's not always feasible.isaacl (talk)00:24, 22 September 2025 (UTC)

@Isaacl, would you elaborate on the "corresponding support channel"? When I look at that page, I don't see anything that looks like a support channel. For the last couple of weeks, I've been having a problem with my usual browser (Firefox, currently 143.0.1 for MacOs 15.6.1) where the visual editor stops loading 2/3s of the way through, and I have no idea why. The visual editor works fine if I switch to Safari, but I normally use Firefox, so that's where my bookmarks, etc., are. The source editor works on both browsers, but I prefer the visual editor and would love for someone to figure out how I can fix this problem.FactOrOpinion (talk)01:27, 22 September 2025 (UTC)

Bug reports should be filed with whoever supports the feature in question.Wikipedia:VisualEditor has a link for reporting bugs.isaacl (talk)04:34, 22 September 2025 (UTC)

@FactOrOpinion, have you done all the usual things, like seeing if it works correctly inmw:safe mode?WhatamIdoing (talk)02:43, 23 September 2025 (UTC)

I got some help from an editor at Phabricator, who had me try a couple of things (including checking it in safe mode, which I wasn't familiar with), determined that it was a problem on my end, and suggested a couple of things to try, neither of which worked. I thought I had done all of the usual things, but I hadn't tried removing all WP cookies, and that cleared up the problem, though I don't understand why. Thanks for the suggestion.FactOrOpinion (talk)03:43, 23 September 2025 (UTC)

I'm glad it cleared up.WhatamIdoing (talk)03:50, 23 September 2025 (UTC)

This is a continuation of a conversation onWP:DISCORD. This will operate like howGit .patch series's will work for copyright issues.

Co-Authored by @Andritolion

This allows for all current tenses to be immediately brought onto the non-sub page on aWP:BLP death.

Technical ideas:

• A bot will sync the main page to the sub page, respecting copyright
• When a source of death is provided somehow, the bot will automatically merge the 2 histories w/ the tenses.

Lillia (She/They)03:27, 20 September 2025 (UTC)

What?Johnuniq (talk)03:45, 20 September 2025 (UTC)

You know how Huggle can undo a diff that's in the middle of anything? That's because it undoes the specific wikitext change the vandal did. With that in mind, we could queue up a diff that changes every "is" into "was" and put the current date as the death date. Particularly useful when a public figure is imminently going to die but it wasn't announced yet. Helps prevent inconsistencies in the middle of a page being live updated.andritolion (talk)04:00, 20 September 2025 (UTC)

Maybe we could call the bot The Grim Reaper, lmao.andritolion (talk)03:54, 20 September 2025 (UTC)

MediaWiki currently doesn't support complex Git-like history merging, with histmerges usually only being doable when the two page histories don't overlap in time. Implementing a Git-like system would be a much bigger task and far out of the scope of this idea. A less pressing issue is that subpages are not enabled in the main namespace, meaning that this should instead be a subpage of the talk page, or a draft.ChaoticEnby (talk ·contribs)04:00, 20 September 2025 (UTC)

It doesn't necessarily need to be implemented within MediaWiki, it could be a diff maintained and performed by the bot.andritolion (talk)04:03, 20 September 2025 (UTC)

And how will the bot determine if the report of a death is premature?Donald Albury13:23, 20 September 2025 (UTC)

A source needs to be provided before the bot will take action. Once someone finds a source for death, you submit the proof and the bot will transform the page.andritolion (talk)14:12, 20 September 2025 (UTC)

Submit the proof to who?AndyTheGrump (talk)14:27, 20 September 2025 (UTC)

I'm trying to think of a response to this suggestion that doesn't include words to the effect of 'crass stupidity' or similar, and failing. Creating past-tense drafts in advance for people who Wikipedia decides are 'imminently going to die' isgrossly offensive (yes, I know newspapers sometimes write draft obituaries in advance, but they aren't going to be publicly visible, and they don't just puke them out unedited when the subject dies). And just what percentage of deaths can be predictedper reliable sources in advance, anyway? The potential for screw-ups is huge, not just with misreported/badly-sourced deaths, but in a multitude of other ways, and just changing a biography to past tense without adding text which details the cause of death etc (written by a human being who can actually read the source being cited, as policy requires) is grossly improper, and almost certainly WP:BLP violation. Any change to a biography of this significanceabsolutely requires human editing. And what exactly is this ridiculous proposal supposed to accomplish anyway? Updating a biography a few minutes earlier? Absurd.AndyTheGrump (talk)

A simple question:what documented and verifiable problem is this bot actually supposed to fix? In as much as Wikipedia has issues with updating biographies on the death of the subject, from my experience they almost always arise from people being in a mad rush to update the article. Very frequently without proper consideration of sourcing. Sure, in this mad rush, tenses don't always get updated, but that is a minor concern compared to the real issue, and automating it (even ignoring all the other problems noted above) is only going to encourage further rushing, on the basis that 'the bot will sort it out'. We don't need to speed up editing biographies in such circumstances. We need to encourage people toslow down and think about what they are doing.AndyTheGrump (talk)14:35, 20 September 2025 (UTC)

In the event where a public person's death is widely rumored or speculated, the page can be locked to extended confirmed users earlier to prevent vandalism or premature edits. Only an extended confirmed user can operate the bot and actually commit the change upon having proof submitted. Particularly useful when a person's page gets massive attention after the person dies. However, anyone can write in that draft, make tense and tonal changes, and the change will instantly be committed and merged with the current revision when the person actually dies. For example, in some languages a sentence needs to be restructured after a person dies. People can get to work on revising the article to fit this and once the person is actually confirmed dead, an article is published with as many accurate tenses as possible.andritolion (talk)13:32, 21 September 2025 (UTC)

You haven't answered the question I asked earlier. Who is the 'proof' submitted to? You aren't seriously suggesting that some random extended confirmed contributor gets to decide whether we have WP:RS for a death? I'd guess that about 1% of such users were competent to make such a judgement. And perWP:BEANS, I'm not going to point out just how easy it would be to abuse the bot.We don't need bots to update biographies faster when the subject dies, we need real human contributors, doing the job responsibly, and properly.AndyTheGrump (talk)11:41, 23 September 2025 (UTC)

In English (the only language that matters here at the English Wikipedia; did you mean to have this conversation atMeta-Wiki?), sentences don't need to be restructured after a person dies, so that's irrelevant.

The problem is largely practical:

• Alice edits the article.
• Bob makes a 'draft' to prepare for the person's death.
• Chris copyedits the article. Now Bob's draft is out of date.
• Bob re-syncs the draft with the new version of the article and re-does the preparation work.
• David expands the article. Now Bob's draft is out of date again.
• Bob re-syncs the draft with the new version of the article and re-does the preparation work.
• Eve reverts David's edits. Now Bob's draft is out of date again.
• Bob reverts to the prior version of the draft.
• Frank rearranges the article. Now Bob's draft is out of date again.
• Bob re-syncs the draft with the new version of the article and re-does the preparation work.
• Grace has a template deleted. Now Bob's draft is out of date again.
• Bob re-syncs the draft with the new version of the article.
• Heidi copyedits the article. Now Bob's draft is out of date again.
• Bob re-syncs the draft with the new version of the article and re-does the preparation workagain.

At what point do you think Bob will quit wasting his time on re-re-re-syncing the article and re-re-re-preparing for the person's death?WhatamIdoing (talk)02:41, 23 September 2025 (UTC)

I feel like having to going like a scavenger hunt of other articles that are subject to category/pages that are so scattered everywhere instead of trying to be neat and in one page like it literally feels like incomplete list of articles that are so far apart away. Unbeknownst that I have a solution to fix it like creating a chart with types/nationally like for example going through military related weapons articles that are in one page when listing of each type of a weapon on a chart and even the State owned enterprise article with the see list of and category at the same time. And could expand over time with new additions on the chart with subjects of related articles in order to reduced bloat.2600:1702:37E0:1710:9969:B5B3:AF1B:BD4D (talk)04:50, 23 September 2025 (UTC)

As long as it largely refers to articles within Wikipedia, it won't violateWP:NOTCATALOG.Cdr. Erwin Smith (talk)11:45, 23 September 2025 (UTC)

Are you looking for theLists of weapons?WhatamIdoing (talk)16:40, 23 September 2025 (UTC)