Analysis
AI detector false positives: why human writing gets flagged
Why AI detectors flag human writing, who gets flagged most often, how to contest a wrong result, and how institutions can build a fair process.
Strategy14 min read
In this piece
- Detectors score predictability, and predictability is a writing style, not a confession. Careful, plain English is the style most likely to be mislabelled.
- The published evidence is one-directional. Stanford's 2023 study measured a 61.3% average false-positive rate on non-native essays, and OpenAI withdrew its own classifier citing low accuracy.
- Every major vendor now tells users not to act on a score alone. Turnitin, GPTZero, Originality.ai and Copyleaks all publish some version of that caveat.