AI Detectors and Non-Native English Writers
Updated
AI detectors and non-native English writers are a poor match, according to published research: seven popular GPT detectors flagged more than half of a set of TOEFL essays as AI-generated, though they were written by people. If you write in English as a second language, or review work by people who do, here is what the evidence shows and how to act on it.
Limitations: Every result is a probability that can be wrong, not proof of how something was written. How results can be wrong →
3 free scans in total, shared across all five tools (not per day or per month). No signup needed to start.
What the Research on AI Detectors and Non-Native English Writers Found
Liang and colleagues tested seven GPT detectors on 91 TOEFL essays written by non-native English speakers. The detectors misclassified over half of them as AI-generated, with an average false positive rate of 61.22%. All seven detectors agreed on 18 of the 91 essays (19.78%), wrongly labelling them as AI-written.
The same detectors showed near-perfect accuracy on essays by US eighth-grade students, which points to a bias in how detectors read simpler, more predictable wording.
Why It Happens
People writing in a second language often rely on common words, familiar phrases and regular sentence structures. Those are exactly the features that make text look predictable to a language model, which many detectors treat as a sign of AI.
Grammar and translation tools can push writing further in that direction by smoothing it out.
If You Write in English as a Second Language
- Keep your process: drafts, notes, outlines and version history are your best evidence.
- Know the research: you can point to the published findings on detector bias if your work is questioned.
- Ask for a conversation: explaining your argument in person shows understanding no detector can measure.
- Check your institution's policy: many say detector scores alone are not proof of misconduct.
If You Review Others' Work
Never act on a detector score alone, and be especially careful with work by non-native English writers. Compare with earlier work, talk to the writer and look at their process before drawing any conclusion.
This site supports English only, and every result carries the same warning: it is a probability, not proof.
Frequently asked questions
Published research found seven GPT detectors misclassified over half of TOEFL essays by non-native writers as AI-generated, which indicates a strong bias.
The study didn't test it, so we can't say how it performs on this. Treat any result on second-language writing with extra caution.
Share your drafts and notes, point to the known limits of detectors, and ask to discuss your work. A score is not proof.
They can make writing more uniform and predictable, which may raise AI scores. Keep a record of the edits you made.
Sources
- GPT detectors are biased against non-native English writers (Liang, Yuksekgonul, Mao, Wu, Zou) (arXiv 2304.02819, read on )