Is the reduction in posts an effect of AI?

We can't know for sure, but this wouldn't be a shocking conclusion. In my experience, the paid version of ChatGPT is very competent at answering basic questions in all branches of mathematics. I have been able to learn about other subjects that I am not very familiar with very quickly, far faster than old-fashioned methods like directly reading textbooks or even google searching. This was a niche that MathOverflow and more generally Stack Exchange sought to fill, but this market share seems to have been taken over by AI.

Of course, even the best models do not come close to human experts when genuinely cutting-edge level discussions are involved; for those MO is still relevant. But this is a much more niche market than general, introductory questions about research-level mathematics.

Here's one sign it may be due to AI.

At the end of 2024, I tried asking ChatGPT about MO questions, as documented in this meta question. Since then, I've periodically done similar informal experiments (although never actually posting answers). As far as I can tell, AI systems don't have a higher success rate now than they did back in 2024.

On the face of it, this apparent stagnation is a little mysterious since AI capabilities have certainly improved in the last 18 months. But it makes sense if one assumes people are asking AI first, and only posting the questions that stump the current technology.