The graph that says it all
StackOverflow, the giant of programming questions and answers, is on the decline. A graph shared on Hacker News this week shows it without filter: the number of new questions on StackOverflow has plummeted since the arrival of ChatGPT and AI assistants for programming.
StackOverflow was for almost two decades the absolute reference for programmers around the world. Millions of questions, millions of answers, a relentless search engine and a reputation system that gamified knowledge. But something has changed.
The graph that says it all
The query on the public StackExchange dataset shows a clear trend: since 2017, the number of new questions had already been declining slowly. But the real collapse began in 2023, right when ChatGPT burst onto the scene. The curve, which previously fell at a gentle rate of 2% per year, plunged vertically. The COVID pandemic also left its mark with a peak of activity in 2020, but the subsequent trend leaves no room for doubt.
Why it is happening
The reasons are multiple and feed into each other:
- AI chatbots are faster: while on StackOverflow a question could be answered in minutes or days, an AI answers in seconds.
- The community became hostile: StackOverflow had a toxicity problem. Too-basic questions were closed abruptly, novices received sarcastic answers. AI never judges you for asking a silly question.
- Content stagnates: the system of marking questions as duplicates froze answers in time. A 2013 answer about a library that has completely changed is still the canonical one.
- The rise of alternatives: GitHub Issues, Reddit and official documentation have been gaining ground.
Is it AI’s fault?
Not entirely. AI has been the accelerant of a trend that already existed. The graph shows that the decline began around 2017. AI has simply delivered the final blow: if you don’t need to wait for a human to answer you, why would you ask?
What future awaits it?
StackOverflow has tried to react. In 2024 it launched its own AI assistant (Stack Overflow AI). But the existential question remains: does a human Q&A site make sense when a machine can answer you instantly?
StackOverflow remains a very valuable resource as a knowledge archive. Its millions of historical answers have fed the training datasets of most current language models. That is the paradox: AI was trained on StackOverflow, and now it is making it irrelevant.
For the everyday programmer, the conclusion is bittersweet. We have gained immediacy and access to information, but we have lost one of the most unique spaces on the internet: a place where programmers helped each other, with its lights and its shadows.
Sources: Hacker News (What AI did to stackoverflow in a graph), StackExchange Data Explorer.
