It's by far the strongest sign of AI writing. I counted 14 instances of the word "not" in this piece, and now I'm wondering if ratio-of-nots-to-length might be a really cheap way to spot text that came out of an LLM.

I think the reason is grates so much is that it's such a cliché. Given their nature, it's not surprising that LLMs would turn to clichés so much.