What Is an AI Detector?
An AI detector is a tool that estimates the probability that a piece of text was generated by an AI model. It works by scoring statistical signals — mainly perplexity (predictability of word choice) and burstiness (variation in sentence length) — that machine writing tends to share. It produces an estimate, not proof.
How AI detectors work
Detectors are trained on large samples of human and AI writing and learn the patterns that separate them. The strongest signals are perplexity — how predictable each word is given the previous ones — and burstiness — how much sentence length varies. AI text is typically low on both: smooth, even, and predictable. Detectors also weigh stock vocabulary, uniform structure, and a lack of concrete detail.
What they output
Most detectors return an overall AI-probability score (often 0–100 or a percentage), sometimes a verdict label, and sometimes sentence-level highlights of the passages they suspect. Humanit’s detector adds a ten-signal breakdown so you can see *why* a passage scored the way it did, not just the number.
How accurate are they — really?
Accuracy varies a lot. Detectors produce false positives (flagging genuine human writing, especially from non-native English writers) and false negatives (missing edited AI text). Short passages are especially unreliable. The responsible takeaway: a score is a signal to look closer, never a final verdict on its own.
FAQ
Are AI detectors accurate?
They are estimates. Detectors can be wrong in both directions — flagging human writing or missing AI text — so a single score should never be treated as proof.
Can AI detectors be bypassed?
Text can be rewritten to change the signals detectors score on (perplexity, burstiness, phrasing). No approach is reliable on every detector version, which is why verifying each time matters.
Try Humanit free
Rewrite AI text to read human, then verify with the built-in detector.
Open the humanizer