Gemini speculated that it might have to do with the AI not mimicking the audios phase properly.
Classic over-processing issue. It's very pronounced with voice "enhancement" or "clean up" tools that are not properly applied. Usually it's someone applying it in the 40-70% range of wet vs. 60-30% dry, thinking they're being conservative with the tool (you really need it to be very light or very heavily applied if you can't get granular with the specifics of how it's applied).
It creates an effect where different parts of the EQ surge up and down gradually over a few seconds but the volume stays relatively the same. The tone is constantly changing which is aggravating. It's a very uncomfortable sensation for a lot of folks.
I speculated that they seem to be adopting some kind of AI based audio cleanup tool that has the same effect, since I first noticed it when one of my favorite Youtubers mentioned they were trying out a new AI audio tool. I think your guess is probably the explanation I was looking for.
Thanks!
A web app for blind testing this would be an interesting project
I found out the hard way though that some modern audio denoisers/etc use something similar. One of my favorite youtubers audio has the exact same audio signature, even though it's clearly not AI generated. Eventually he mentioned that he was using a new AI cleanup tool and it clicked. Since then I've noticed it creeping into more and more videos. Linus Tech Tips uses it on some videos, for example.
So even for people who are sensitive to it, the false positive rate is now probably very high. It's making me a little crazy that other people don't seem to be bothered by it.
Not sure why you got downvoted, but I too think it would be interesting. I vaguely remember a story about it posted here a while ago.