That seems like a reasonable idea to me but I think part of the answer is that a mind boggling amount of videos are uploaded to Youtube every second. Tracking categorization abuse seems straightforward in theory but good luck automating that reliably over millions of videos especially when you know that some very dedicated people will try their best to game the algorithm in their favor.

People will start using the "blood and violence" category exclusively if it means they're less likely to be demonetized, rendering it useless. Or the other way around: if this category means they get fewer views they'll try their best to be always nearly avoid it, except of course it can get very subjective and some people are going to be flagged "blood and violence" even though they argue it's not that violent and channel XYZ did worse and didn't get tagged etc...

When it come to moderating Youtube, and given the ridiculous amount of content hosted there, you shouldn't think "how would I do it" but rather "how would I design an algorithm that would do that". And suddenly it becomes a lot less obvious. You can't teach algorithms common sense (yet).