How the AI detector works
Instead of one black-box percentage, the detector runs several independent layers and combines them. A statistical layer measures burstiness and sentence-length variation, the signals that separate human rhythm from model-smooth prose. A pattern layer looks for the giveaway phrasing that large language models overuse. A language-model judge reads the passage the way a careful editor would and estimates how likely each sentence is to be machine-written. Because the layers are independent, a text has to fool all of them at once, which is far harder than beating a single score. The result is a stable verdict you can trust, not a number that swings every time you paste the same paragraph.