how r you going 2 know ai writing?
AI throws words together that don't make any sense, even if from afar it will look like regular writing. It also never gets character's personalities correct at all, and anything it does get "correct" would have been ripped from other writers.
To expand on this, a lot of AI language models are trained to string wordds together by analyzing a corpus and comparing how often groups of words (called n-grams, where n is some number) appear. These frequencies are computed, and then turned into a statistics-based model that tries to put words together based on these frequencies. For example, the bigram "Once upon" will probably appear more frequently then the bigram "Once goes" - the former, of course, is part of the classic phrase "once upon a time" while the latter is not really proper English. Let's say that our AI model has typed out the word "once". It's much more likely to select "upon" as the next word rather than "goes".
That being said, Emmet is right that AI will put together words that do not make sense, and has a hard time with consistent character personalities. The vast majority of language models out there do not retain a record of what they've previously typed out for a prompt. They're simply concerned with the current word and the handful of words surrounding it. This causes the lack of cohesion frequently seen in ai writing. Chatbots do a better job with this since they are only trying to emulate one person, rather than multiple characters and/or an overaching narrator. Just like with AI generated images, once you know what you're looking for it's not that difficult to spot stuff made with AI