An AI detector is a tool that estimates the probability that a piece of text was generated by artificial intelligence rather than written by a human. Schools use them to screen homework; students and writers use “humanizers” to try to defeat them. The critical thing to understand is in the word estimates: a detector produces a likelihood, never a verdict.
How it works
Most detectors don’t recognize AI so much as measure predictability. They score two properties: perplexity (how surprising the word choices are) and burstiness (how much sentence rhythm varies). AI text tends to be smooth and statistically ordinary, so low-perplexity, low-burstiness writing gets flagged. The catch is that plenty of human writing is also smooth and ordinary — which is exactly where the errors come from.
Why it matters for homework
Detectors matter because schools sometimes treat their scores as evidence, and the scores are unreliable in ways that hurt real students. In a landmark Stanford study, seven detectors wrongly flagged human-written essays by non-native English speakers 61% of the time on average, while flagging native-born students’ essays at near zero. Even a “99% accurate” detector produces roughly 15 false accusations for a teacher grading 1,500 submissions a term.
So the practical stance: a detector flag is a reason to have a conversation, never proof of anything. If you’re a student facing one, the defense is process — draft history, version history, and the ability to explain your own work. The full evidence is in our guide to whether AI detectors actually work (short answer: not reliably).
In one sentence
An AI detector tells you how predictable a text is, not who wrote it — so its output is a prompt for scrutiny, not a substitute for it.
Related: hallucination · academic integrity