I have been testing something recently and the results were honestly a bit confusing. I took a piece of AI generated content and then paraphrased it properly not just basic word swapping but restructuring sentences, adjusting tone, and adding a more natural human feel. After that, I ran both versions through a few AI detectors. The original version was flagged quite clearly as AI, which was expected. But the paraphrased version got a much lower AI score, and in some cases it even passed as mostly human written.
That made me question how reliable these tools really are. If paraphrased AI content can slip through so easily, are clients relying too much on detection scores? And on the other side, are writers sometimes being judged unfairly because of these tools?
I am not trying to prove a point here just trying to understand how people are looking at this right now.
Have you tested something similar and do you think paraphrasing is enough to get past AI detection or is it more complicated than that?