What is Perplexity?
Perplexity is a measure of how predictable a piece of text is to a language model. Low perplexity means the model finds each next word unsurprising; AI detectors treat low perplexity as a hint that text was machine-generated.
Language models assign probabilities to every possible next word. Perplexity summarizes those probabilities across a passage: text the model predicts easily scores low, text full of unexpected word choices scores high. Because generators produce exactly the words a model finds most probable, generated text tends toward low perplexity.
The problem with using perplexity for detection is that plenty of human writing is also predictable. Technical documentation, legal prose, formulaic news copy, and writing by non-native speakers who learned standard constructions all score low — which is a major source of detector false positives.
Perplexity describes the output. It says nothing about the process that produced it, which is why two texts with identical perplexity can have completely different origins.
Burstiness
Burstiness is the variation in sentence length and structure across a text. Human writing tends to alternate short and long sentences unpredictably, while AI-generated text is often more uniform — so detectors use low burstiness as a machine signal.
AI detector false positive
An AI detector false positive is human-written text wrongly flagged as machine-generated. Documented rates are highest for non-native English speakers — one Stanford study found detectors flagged 61% of TOEFL essays as AI-written.