The irony here is that ChatGPT can operate as a fairly good copyrighter. But since it is detectable, there are now human copyrighter services like this one to humanize the output.
The main giveaway when AI is used is that people generally have a predictable pattern in their writing. It's not perfect. But when it is completely absent, it's obviously not human.
I'm sure there is more to it than that, since I'm basing that solely on how some software sniffs out student plagiarism - look for pattern mismatches, then search for them to see if they appear in similar documents elsewhere.
AI content detectors usually operate under the assumption that AI writes even more predictably than humans, who have a level of randomness in factors such as sentence length and word relatability. This program attempts to humanise AI writing by ensuring its patterns are slightly random but not too random, right in that human sweet spot.