That's actually a good point, your markup can reveal some info about you. Like, today this kind of info is mostly unnoticed, because there are too much details in the world and too few eyeballs to notice them. But in the (not so distant) future AI agents might be able to do a lot of inference by drawing from info like "this person hand-authored their website since the 2000s and cared a lot about obscure details etc etc"

Specially if the context size grows to the point that wasting tokens on such trivia is not seen as too wasteful

I also think it's pretty likely that if a website has fully valid HTML, that website makes little to no money. Because if it actually was a revenue-generating venture it's extremely unlikely the person or people behind it would be worrying about what tags have which attributes and which tags are self-closing and which aren't when it has zero impact on revenue.

Being valid XHTML 1.0 strict was a badge of honor

IIRC, that "w3c validator" was overly strict, and would error on attributes it didn't recognize, even though they were technically 'valid' (and supported in ~every browser). So there were always a couple things like, ehhh, doesn't matter.