Intent matters. If you use the fonts list to identify persons, it's PII.

Perhaps a reasonable line to draw would be that if the content of the information is enough to fingerprint, the it is PII (regardless if broken into many messages) and the individual data doesn't matter.

If we talk purely about "reasonable", I don't think it's even possible to draw any line, besides something obvious like unique user identifiers. It's hard to predict what data can be used for fingerprinting until someone actually does it, just like it's hard to predict what bugs are actually possible to exploit. I could never predict some of those things being used for that and yet here we are. So it can be a cat-and-mouse type of situation. And some of that information is probably legit useful for purposes other than tracking.