Last month I tested 4 AI content detectors on the same 600 word blog post and got 4 different scores
I run a small marketing shop in Boise and last month a client asked me to prove their blog posts were human written. I grabbed one 600 word post, fed it through 4 different AI detectors, and got results of 98% human, 12% human, 87% human, and 41% human. Same post, same words, same author. That is a 86 point swing on nothing. I paid $29 each for two of those tools. OpenAI shut down their own classifier last year because it was wrong so often, but these companies still sell the same idea to scared business owners. Now I am seeing agencies charge clients $500 audits where they run copy through a detector and call it proof. If a tool can not agree with itself on one document, it should not decide if someone gets paid. Has anyone here found a detector that stays consistent across tries, or is the whole category just noise?