Yeah, SO's downfall started looong ago. The community was frankly horribly managed, and its strict "no remotely duplicate-esque questions ever" policy meant that answers to common questions just got more and more outdated as time went on. It's still common to search for something, find an SO thread, and find that the only answers are from 2013. The world has changed since 2013, answers in 2026 would be different, but because the question would be the same, any contemporary attempt at asking the same question would get marked as a duplicate, so the 2013 answer remains SO's only guidance forever.
They also had the problem that easy questions would get downvoted for being too easy, and hard questions would just not get answered because they weren't seen in time by the narrow group of people who could answer so they get buried by the algorithm. Working in something of a less common niche myself (embedded Linux), I never had questions get answered. I believe the question ranking systems and moderation policies really only worked for questions about new, popular web frameworks.
It was ChatGPT which did it in, but it could've been anything. It could've been a new group of people with some clout starting a fresh new knowledge site. People were ready to abandon SO.
> and its strict "no remotely duplicate-esque questions ever" policy meant that answers to common questions just got more and more outdated as time went on.
No; answers got outdated because of people not doing what they were supposed to do, which is: post a new answer on the old question. (Assuming that the old answers actually are outdated. It happens all the time that decade-old answers are still perfectly relevant, or even still represent the Right Way to Do It(TM).)
Somehow, people never wanted to do that when it actually mattered, but would constantly post new, repetitive, low-quality answers on popular old questions (often re-combining multiple old answers) even though the solutions hadn't actually become outdated, simply because they were popular questions and they hoped to gain visibility from it (whether from reputation, or because user profile pages had a "number of users reached" metric which literally just sums the view counts of the pages where you have a question or answer, including views that occurred before your post).
That said, there are plenty of cases where an old question got closed as a duplicate of a newer question: either because it went unnoticed before, or because of deliberate action to reverse the direction of closure, with the goal of pointing future viewers to the best possible Q&A. This was explicitly in accordance with policy: see e.g. https://meta.stackoverflow.com/questions/404535 , https://meta.stackoverflow.com/questions/251938 , https://meta.stackoverflow.com/questions/423085 . As a general rule, policy is "timeless": the timestamp on a post is not relevant to determining the right thing to do with it.
It was also the rise of Github and the importance of the software hosted there. More consistent documentation and transparent issue trackers/PRs helped a lot dealing with evolving software.
The data showed Stack Overflow declined from 2017. This followed GitHub's rise distantly.
GitHub made documentation more consistent how? GitHub issues were more transparent how? A PR answered a question I would have asked Stack Overflow never I think.