GPTZero

Accessible without institutional purchase and returns sentence-level rather than document-level output. The company has published more about its own limitations than most detectors.

Not recommended for determining misconduct. Its one defensible classroom use is as a demonstration object — having students run their own human-written work through it and watch it get flagged.

Liang et al. (Patterns, 2023) tested seven detectors and found an average 61% false-positive rate on TOEFL essays by non-native English speakers versus about 5% on native-speaker essays — a twelvefold disparity, with one detector flagging 97.8% of TOEFL essays as AI. The mechanism is low text perplexity, which is a feature of developing English proficiency and of some neurodivergent writing, so it systematically misidentifies exactly the students a community college serves. There is no appeal path: a flagged student cannot prove a negative.