It's hard to say how exactly Google will identify AI content.
Most AI content just reads like human-written content, just a tad bit incoherent or rambling, no different than many outsourcers tbh. Probably because GPT3 itself has been trained on a ton of internet data (in addition to libgen books). If you look at some of the 'human-generated' content out there, some of it looks like markov text - - not everyone on the internet is Shakespeare or Hemingway.
Google will probably go after the most obvious AI content, like websites that don't have much authority but are spitting out 10k pages a day, or websites that have no pages other than PAA.
But perhaps not even that...who really knows, so far Google has been spectacularly bad at preventing spam, so it's not like they have them most amazing track record.
This isn't to say that google is terrible at everything - - they choose to focus on a few things and are really good at those things (extracting snippets from websites, for example), but to say Google is amazingly good at detecting spam is...optimistic, at the very least.