What AI did to Stack Overflow, in one graph

A short query published on Stack Exchange's Data Explorer has become one of the clearest public records of how artificial intelligence changed programming: it counts the questions posted to Stack Overflow every month since the site opened, and the curve falls off a cliff. The link, titled “What AI did to stackoverflow in a graph,” was created on July 18 and was among the most upvoted posts on Hacker News that day.

What the query counts

The SQL is deliberately plain. It groups every post of type 1 — a question — by the month it was created, counts them, and orders the result by month. That returns 217 rows, one for each month from July 2008, when Stack Overflow opened, through July 2026. Anyone can fork the query and rerun it against the live database; the yearly totals below come from the same public dataset.

The curve after ChatGPT

Question volume peaked in 2016 at 2,183,175 for the year and stayed above 1.5 million as late as 2021, with a bump to 1,855,038 in 2020. Then the line bends: 1,335,306 questions in 2022, 787,777 in 2023, 398,434 in 2024 and 109,761 in 2025. Between January 1 and July 18 this year the same data records 17,014. The monthly view is starker still: November 2022, the month ChatGPT launched, saw 109,150 questions; June 2026 saw 1,729, about 98% fewer.

Why the graph matters

Stack Overflow spent two decades as the place programmers went when they were stuck, and its public archive became a standard ingredient in the datasets used to train coding assistants. The graph is now the most cited picture of that trade: questions that once trained the tools are increasingly answered by them, inside the editor, before anyone opens a browser tab. The open question for the site is whether the remaining flow of questions — a fraction of its peak — can still support the moderation and curation that made the archive worth training on in the first place.