AI Killed Stack Overflow's Question Volume — Now What?

The signal: A viral HN graph shows Stack Overflow’s question volume in freefall since ChatGPT and Copilot went mainstream, and developers aren’t shocked — they’re just now looking at the receipts.

Why it matters: Stack Overflow was the default place to get unstuck for two decades, and its slow death changes where junior developers learn debugging patterns and where LLMs get their next generation of training data. If the well dries up, models trained on future code Q&A lose a key feedback loop, and new devs lose the messy, human back-and-forth that taught pattern recognition, not just syntax.

Does this mean Stack Overflow is actually dying?

Yes, as a place people go to ask new questions — but no, as an archive, because its 20+ years of answers are baked into every model’s training data and will outlive the site’s traffic. The graph isn’t showing a slow decline, it’s showing a cliff that lines up almost exactly with ChatGPT’s public release and the rise of Copilot-style inline assistance. Developers stopped posting because the marginal cost of asking a model is near zero and the answer comes back in seconds, not hours. What’s dying isn’t the knowledge, it’s the incentive to publicly document new problems — which means whatever isn’t already indexed on SO risks never getting written down anywhere at all.

The pattern I’m watching: This is the same shift playing out across every “ask a human community” surface — Reddit threads, dev Discords, even GitHub issue Q&A are all seeing traffic move toward private LLM sessions instead of public, searchable posts. The irony is that models need exactly this kind of fresh, human-verified content to keep improving on genuinely new problems, and that supply is shrinking right as demand for it — via RAG and fine-tuning pipelines — goes up.

What I’d do with this: If you’re building dev tools, stop assuming Stack Overflow-style data will keep flowing — start capturing your own users’ Q&A exhaust (support tickets, Slack threads, PR comments) as a private knowledge base before it all migrates into ephemeral chat logs. If you’re a working developer, keep writing up genuinely novel bugs somewhere public; you’re not just helping the next person, you’re one of the last sources of fresh training signal for the tools you rely on.

Key takeaways

• Stack Overflow’s question volume didn’t gradually decline, it dropped off a cliff that tracks almost exactly with ChatGPT and Copilot adoption.
• The site’s real value now is as a frozen archive baked into model weights, not as a living Q&A community.
• New, genuinely novel technical problems are increasingly solved in private chat sessions instead of public forums, which starves future models of fresh training data.
• Builders who capture their own users’ technical Q&A now will own a data moat that public forums used to provide for free.