Too much / too little, both are problematic I guess. This ends up becoming a war of algorithmic generation vs understanding, and devolves to effectively answering the age old question "is this spam"?

I was going to suggest that perhaps adding info about whether the site had ads, etc. might help the usability. Then I tried marginalia ... :) Perhaps we need to invest in similar detection for other antipatterns (listicles, blogspam etc.) - this is mostly what I was talking about. Less the understanding for indexing, and more the classification for user safety and authenticity of results.

But again, this is very cool.