Stack Overflow's question volume has been falling since ChatGPT went public in November 2022 (OpenAI). The site that trained a generation of developers, and most of the AI tools those developers now use, is slowly emptying out.
In October 2023, Stack Overflow laid off 28% of its staff (Stack Overflow Blog). CEO Prashanth Chandrasekar framed it as a restructuring toward profitability. Everyone in the industry understood the real cause. Traffic was down. The thing causing it was sitting in every developer's browser tab.
This is not another "AI killed Stack Overflow" piece. That take is everywhere and it misses the actual problem. The interesting part is the feedback loop, and it points somewhere uncomfortable for the AI industry itself.
The conventional story, and what it misses
The popular version goes like this. Developers used to paste error messages into Google and land on a Stack Overflow thread. Now they paste the same error into ChatGPT, Claude, or Copilot and get a direct answer. Why click through to a forum, risk a condescending comment, and wait for a human when a model answers in two seconds?
That part is true. It explains the traffic drop. It does not explain why the people building the AI should be worried.
The seed corn problem
Here is the part most coverage skips. Every large language model trained on internet text consumed a huge amount of Stack Overflow. The site's archive of voted, edited, human-reviewed answers is one of the highest-quality programming datasets in existence. It is the reason an AI can answer your Python error at all.
Now run the loop forward. AI tools answer questions directly. Developers stop posting on Stack Overflow. The archive stops growing. The next round of models trains on a corpus that is increasingly old, increasingly stale, and missing everything that happened after 2022.
When you train an AI on data generated by another AI, quality degrades. Researchers proved this formally. Shumailov and colleagues showed that models trained on model-generated output collapse over generations, losing the long tail of real human knowledge and drifting toward a narrow average (Nature, 2024). They call it model collapse. The mechanism is simple. Synthetic data amplifies the most common patterns and erases the rare ones. After enough rounds, the model forgets how humans actually write and solve problems.
Stack Overflow's slow death is the input side of that collapse.
What Stack Overflow tried
To its credit, Stack Overflow did not just watch. In late 2022 the community banned AI-generated answers, because ChatGPT responses looked plausible but were frequently wrong and were flooding the review queues (Meta Stack Overflow). The site then flipped the strategy. In 2024 it signed a data-licensing partnership with OpenAI, getting paid for the archive AI companies had been scraping for free (Stack Overflow Blog). It opened an API for programmatic access to its data and started charging commercial users for it (Stack Exchange API).
The logic was reasonable. If AI was going to eat the archive, Stack Overflow might as well get paid for the meal. But it does not solve the core issue. Licensing old data does not create new data. And the flow of new human answers is exactly what is drying up.
Why this matters for developers
If you build software, you have already noticed something. The AI tools are good at the common cases. They nail the top questions because those were answered a thousand times on Stack Overflow before 2022. Ask about something newer, a library released last month, or an obscure flag in a recent framework version, and the model gets vague. It hallucinates. It confuses versions. It gives you an answer that was correct two years ago and is wrong now.
That gap is the archive gap. It will widen as the human pipeline shrinks. The models that impressed everyone in 2023 and 2024 were trained on a living, growing Stack Overflow. The models shipping in 2026 and 2027 will be trained on a shrinking one.
You can verify the trend yourself. The Stack Exchange Data Explorer lets anyone run SQL queries against the live database, and the numbers on question volume and accepted answers are public (Stack Exchange Data Explorer).
So what replaces it?
The honest answer is that nobody knows yet. Some signal is moving to Discord and Slack communities, which are closed and unsearchable, so they cannot train models at all. Some goes to GitHub issues and discussions, which are messier and more project-specific. A small fraction still goes to Stack Overflow, but the velocity is a fraction of what it was.
The result is a slow-motion version of a problem the web has faced before. Encyclopedic human knowledge gets centralized in one place, that place gets valuable enough to scrape, the scraping removes the incentive to contribute, and the place decays. Wikipedia survived because editing is its own culture. Stack Overflow's value was always the answers, and the answers are exactly what the AI took.
If you want the AI tools to keep getting better, the health of the human knowledge sources they train on is your problem too. Answer a question publicly once in a while. The model that copies your answer in three years will thank you, in its own average, increasingly synthetic way.
Sources:
Top comments (0)