[adishy]

I think a few of the other comments poke at this general idea, but I've come to be very intrigued by the meta-game of training these media recommendation engines (YouTube, Spotify, Netflix, TikTok?) by picking the right things to get a good feed. This sort of training is becoming more important than ever for a good experience in these platforms.

In my experience, YouTube seems to be one of the more reactive recommenders - especially if you go down rabbit-holes / dig deeper into certain types of topics. I have turned off both web history and Youtube watch history in my Google account and I find that the Youtube homepage very quickly devolves to lists of videos by channels I've subscribed to - a randomized subscription feed. I would really be interested in seeing recommendation behavior be quantified and documented in some systematic/objective way (if this is even possible).