Disrupted
Stack Overflow and the questions that stopped coming
Stack Overflow · Developer platform · United States
For fifteen years, developers asked Stack Overflow. Then they started asking ChatGPT instead. New questions have fallen more than 75% — and the site trained the models now replacing it.
−76%
new questions since ChatGPT
~50%
monthly visits lost, 2022→24
28%
staff cut, 2023 (reported)
A Q&A network hollowed out by a faster answer.
The record
- New questions on Stack Overflow have fallen roughly 76% since ChatGPT’s November 2022 launch; by April 2025 monthly posts were down about 64% year-on-year and more than 90% from the 2020 peak. (Stack Overflow Data Explorer, via reporting)
- Monthly visits fell from around 110 million (2022) to about 55 million (2024) — roughly a 50% drop. (Similarweb; reporting)
- Stack Overflow cut about 28% of its staff in October 2023. (Reported)
The answer got faster
For fifteen years the reflex was the same: hit a wall, ask Stack Overflow, wait for the crowd. ChatGPT and Copilot gave the same answer instantly, in context, without the friction of posing the question to strangers who might close it as a duplicate. The reflex moved, and question volume fell off a cliff.
None of that required Stack Overflow to do anything wrong. The thing it did was simply done faster somewhere else.
It trained its own replacement
Here is the bitter symmetry: the models draining Stack Overflow’s traffic were trained substantially on Stack Overflow’s own fifteen years of answers. The value the community created flowed into the tool that displaced the community.
A network effect is a powerful moat against another network. It is no moat at all against a substitute that removes the need to ask the network in the first place — least of all one your own data taught to answer.
The freeze, at internet scale
As with Chegg, there is no bad automation to point at, no reckless cut. A network effect that took fifteen years to build unwound in about two, because the job it did was suddenly done better elsewhere. The platform still runs, and has added AI features of its own — but the reflex it owned is gone.
That is the shape of the freeze: not a dramatic failure, but a slow subtraction of the reason anyone needed you.
The lesson
A network effect is not a moat against a substitute that removes the need for the network — especially one your own data trained.
How we’re reading this
The traffic and question figures come from Stack Overflow’s own Data Explorer as reported by third parties; the ~28% October 2023 layoff is well-reported. Part of the decline also reflects broader shifts in developer forums and search traffic, not AI alone. Stack Overflow continues to operate and has added AI features.
Sources
- 01Stack Overflow is a ChatGPT casualty: traffic down sharply
Similarweb · Apr 16, 2024
- 02Stack Overflow Usage Plummets as AI Chatbots Rise
Slashdot · Jan 10, 2025
- 03Stack Overflow’s decline
Eric Holscher · Jan 21, 2025
Compiled from public filings, court records, company statements and reputable press. Figures are attributed to their sources; allegations are labeled as such. Not legal or investment advice.