GPTZero Score Meaning: What 30%, 70%, 100% Actually Signal
A plain-English guide to reading GPTZero's percentage score, the sentence-level highlights, and what threshold professors typically care about.
GPTZero returns a percentage and a color-coded sentence view. Students see "78% AI-generated" and panic. Professors see the same number and mostly shrug. The gap between those two reactions is where most academic-integrity misunderstandings live.
What the percentage measures
GPTZero's headline percentage is the estimated probability that any part of the document was AI-generated. It is not the fraction of the document that was written by AI. A 78% score means "we are 78% confident this document contains AI-written text somewhere" — not "78% of the words are AI".
The thresholds professors use
- Below 30%: Effectively noise. Not actionable.
- 30–60%: Suspicious. Usually not enough for an integrity report on its own.
- 60–85%: Trigger threshold at most universities. Usually paired with a second detector.
- 85%+: Very likely to be reported. GPTZero's own recommended action tier.
The sentence-level highlights matter more
The document-level number is a summary. GPTZero also highlights specific sentences it flagged, color-coded by confidence. Instructors who take detection seriously read those highlights, not the headline. A 45% overall with three deep-red sentences reads worse than 65% with everything mid-yellow.
Cross-reference: false positives
Any GPTZero score should be sanity-checked against another detector and against your own drafting history — see why human writing gets flagged. And if you did use a model, tools like Bypassify exist to bring the sentence-level highlights down alongside the headline number.