I haven't run into AI bots much with messaging, but I have had trouble with AI photos. I understand it's natural for providers to touch up their photos, but when folks use a filter from an app, it's hard to determine whether they've done light editing, heavily distorted their features, or sometimes generated a person from scratch. AI detectors flag all of these as "likely AI use".
A couple years ago I'd look for tattoos - it's hard to consistently apply heavy filters without messing that up, or generate a person with the same tattoo across multiple photos, and easy to verify in person. But even those are straightforward to replicate with enough effort now.