For some reason I can't really put my finger on, this is true - we humans can detect AI-content better than detection softwares - but the HOW is again I can't really put my finger on...
The only challenge here is "proving" it was AI-generated - but once such a proof framework is made public - the same AI-generator writers will adapt to be less AI-detect-provable...
It's a Catch-22 of sorts.