The effect of AI generative tools on Stack Overflow [closed]

Question 1

Closed. This question is opinion-based. It is not currently accepting answers.

Want to improve this question? Because this question may lead to opinionated discussion, debate, and answers, it has been closed. You mayedit the question if you feel you can improve it so that it requires answers that include facts and citations or a detailed explanation of the proposed solution. If edited, the question will be reviewed and might be reopened.

Closed5 days ago.

Improve this question

This question was migrated from Stack Overflow because it can be answered on Meta Stack Overflow.Migrated11 days ago.

I would like to know how much the advent of AI tools like chatGpt and many other affected the usage of Stack Overflow. Before it was a holy place for developers. How much did AI affect the traffic in Stack Overflow?

In certain future, do you see Stack Overflow getting completely switch-off as AI models keep upgrading?

Question 2

AI is trained on SO, for that, many people say the quality of AI is comparable. If SO goes out, the AI becomes outdated and there is no new source.

Question 3

Although I use AI nearly every single day now as a software engineer with 28+ years of experience that doesn't really change my usage of StackOverflow much.

Question 4

AI generated content is not allowed on Stack Overflow. So the impact of AI tools like ChatGPT on the community is fairly small, other than the impact on moderation, in order to delete the content that is not allowed.

Question 5

The rate at which new questions are asked has dropped by more than a factor of 10 and keeps decreasing:meta.stackoverflow.com/a/434504/7483211 Traffic has likely decreased similarly, though it's harder to know without access to private data. Traffic will keep decreasing as older content becomes outdated. I'm pretty sure the bulk of the decrease is due to ChatGPT & co.

Question 6

@CorneliusRoemer then why did the decrease start before the release of widely available GenAI tools? You seem to be misinterpreting the data.

Question 7

related on MSE:meta.stackexchange.com/q/384355/997587 meta.stackexchange.com/q/386983/997587. also, ... why is this closed as not seeking discussion?

Question 8

@starball When I saw this thread a short while ago I assumed there were a bunch of deleted comments with an "unproductive" conversation. Apparently not :p Voting to reopen.

Question 9

@cafce25 it started earlier because Stack Overflow had a PR problem long before pseudo-AI came along. But when pseudo-AI did come along, people had the ammunition they needed to actually turn dislike into hate and avoid the platform altogether. It's cause and effect.

Question 10

I think this question is too general and cannot be answered reasonably within the bounds here. LLM services surely have affected everything and the future is unclear. You will likely only hear opinions (positive or negative) about them. Maybe you want to know something more specific? For example only about the traffic or only about specific usage patterns?

Question 11

@drescherjm which is asking 1 question in 15 years so that won't help SO.

Question 12

Obviously the quality dropped a lot - e.g.data.stackexchange.com/stackoverflow/query/1923221/… I also think SO is undermoderated given how huge it is and how some moderators reject any report that isn't blatantly obvious. Or, well, even answers that are of terrible quality and don't meet a lot of standards.

Question 13

The rate at which new questions are asked has dropped by more than a factor of 10 and keeps decreasing:https://meta.stackoverflow.com/a/434504/7483211 Traffic has likely decreased similarly, though it's harder to know without access to private data. Traffic will keep decreasing as older content becomes outdated. I'm pretty sure the bulk of the decrease is due to ChatGPT & co.

–Cornelius Roemer (ref)

While some people disagree that AI tools were the main cause of this decline, they certainly had a part in this.

What I imagine happened:

Users who wanted to ask easy questions started asking AI tools instead of Stack Overflow. AI can answer easy questions easily and often correctly. So there is no need to ask these questions on Stack Overflow anymore.
Users who were too lazy to pose their question started using AI instead. It's much easier to clarify your question when using an AI tool than Stack Overflow — you just make a chat instead of posting comments.
Users who were too lazy to search or unable to judge the relevance of search results started using AI tools. Search was always bad/broken on Stack Overflow, so AI tools surpassed its efficiency easily.

So we have much fewer "bad" questions now. A "healthy" process of replacing human labour by robots. If a robot can answer a question, humans shouldn't waste time even reading it.

What about answers? I don't have the statistics. I hope it has improved, but I have no reason to believe that.

Regarding the future: we still have the "problem" of quality of AI's answers. If your question is moderately hard, if it contains something new which AI has never trained on, if it's interesting — it should be asked on Stack Overflow. Or some other site which will replace Stack Overflow. I see only two ways for Stack Overflow to become useless:

Software development becomes an insignificant part of human culture. Do we have a Stack-Overflow-like site for hide-and-seek strategies? No. If software development becomes as insignificant as that, there you go.
AI will solve all humanity's problems. This is an apocalyptic scenario which can happen, and all bets are off.

Question 14

So existing contributions were to difficult to find and categorize, leading to questions being asked more than once at various levels of quality, and only in rare instances could the existing answers be linked to the new questions. Add the low quality garbage that cannot be answered without a crystal ball, and AI generated nonsense, and you have the perfect storm for reducing participation of actual developers who can answer questions (like myself)

Question 15

If AI ends up solving all of humanity's problems, that's a good thing, isn't it? It would only be apocalyptic if it made the problems worse, or introduced new problems.

Question 16

@cigien One might solve all the problems in code by erasing it (sure, loosing some features). Or it's just easier to solve all the problems by rewriting the code from scratch. Don't replace 'code' with 'humanity'.

Question 17

3. SO itself decides to invite low-quality questions with low-quality answers and no means of community moderation. In turn, this site will eventually become a landfill of software trivia. At this point, it will be impossible for AI to tell gems from garbage and the site becomes useless as a training resource - but this couldn't possibly happen, right?

Question 18

The number of questions decreasing in itself is not a problem imo because it was always supposed to go down right? I think a far bigger issue is experts leaving the site. If experts are not here, it doesn’t matter how many questions are posted, quality answers won’t be here. I think SO Inc’s homework is to retain the experts, not increase questions or comments or whatever that artificially increase traffic.

Question 19

How much did AI affect the traffic in Stack Overflow?

There has been a 25-fold (!) decrease in the number of questions asked on SO since the peak use a few years ago. We debated the reasons onThe number of questions has decreased 10-fold: where did users go and what can be done to prevent that? My stance is that nowadays AI canproperly answer most SO questions even if it doesn't see the SO page containing the answer.

As Imentioned on MSE, most of my questions fall into 2 categories:

Easy questions, which AI typically answers well (Gemini, Grok, GPT, Claude, deepseek, kimi, qwen, etc.).
Hard questions, which SE typicallyremoves with Roomba (no upvotes and no fast answers since too niche, or sometimes closed by some users who don't understand that niche).

So there's little incentive left for people to keep using SO.

do you see Stack Overflow getting completely switch-off as AI models keep upgrading?

Depends on how successful SE's enterprise offering is. They will need that money to pay for SO's bill. Since the number of questions has sunk drastically I don't think they can still make much moneyselling our data.

Question 20

"AI can properly answer most SO questions" No, only basic stuff."which SE typically removes with Roomba" You're letting your perception bias speak for you, there. Most questions fallwell between those 2 extremes. And even then, "Abandoned" questions don't get "Abandoned" because they're "hard", they're just way too specific, or poorly written and don't get any traffic from people searching for similar problems.

Question 21

@Cerbrus"Most questions fall well between those 2 extremes" Sure, and honestly, I read Frank’s post as humor. But the reality is that time isn’t on SO’s side. The gap between those two extremes keeps narrowing. We can’t slow down what’s happening on the AI side (it's getting better and better and that's real), so the only part we can really influence is our own. And that, I think, is the core of all the recent tension between SE and the community.

Question 22

@saastn you're massively overestimating the capabilities of AI... Also, Frank isn't the type of user that'd make jokes like that, in my experience.

Question 23

@Cerbrus I don’t think I’m overestimating anything. AI is a tool, just like every other tool we creted to extend our capabilities. No one can predict the future, but yes, I do believe something has happened: it has finally passed the Turing test, or at least reached an acceptable score. Is it fully mature yet, especially in coding? Of course not. But IMHO it’s only a matter of time. I see many people exagerating about complexity of coding. I think in a few decades, we won’t need as much human-written code anymore, just as we no longer need as many handmade shoes as we once did.

Question 24

Seeing is believing, @saastn. AI has been consistently overvalued, up to now. AI constantly makes mistakes on more complicated codebases. You're talking about "decades"... That's aninsanely long time, in the IT world.

anatolyg 28.5k1 gold badge33 silver badges43 bronze badges · Answer 1 · 2025-11-18 12:07:51Z