I agree, but the point is AI, and e.g. chatGPT, will always have one way to generate content. Whereas humans may have diverse ways of writing and suppose if we consider an essay or an article, the way of writing by a human would vary with every single sentence but it would remain the same for AI throughout. That's how AI-generated content can be detected. If we do para-wise analysis, we would get better results and a clearer picture but it won't be the same for sentence-wise analysis. And there should not be any possible way that for a particular human, all the generated paragraphs come out to be detected as AI-generated.