AI detector false positives
By Sourabh Singh18 September 20266 min read
If you wrote every word and still scored 90%, here is what actually happened.
Being flagged for work you wrote yourself is the most common complaint about this technology, and the most poorly handled. It is not rare, it is not your fault, and it is not evidence of anything.
Why it happens
Detectors measure two things: how predictable your word choices are and how much your sentence lengths vary. Writing that is plain, careful and consistent scores as machine-written because it genuinely has those properties. The detector is not malfunctioning, it is measuring a style and the institution is reading the result as an origin.
Who gets hit hardest
- Non-native English speakers, whose writing tends toward narrower vocabulary and steadier construction.
- Technical and scientific writers, where precise repeated terminology is required.
- Anyone taught to write in short, clear, uniform sentences.
- Writers who use AI-assisted grammar tools, which smooth rhythm in exactly the flagged direction.
If it happens to you
- Do not rewrite in a panic, that destroys the version history that is your best defence.
- Gather your process evidence: drafts, notes, search history, document version history.
- Run the text through a second and third detector and screenshot the disagreement.
- Get a sentence-level reading so you can point at what specifically triggered it.
The institutional picture
Several universities have stepped back from detector-based accusations after cases collapsed, and Turnitin itself advises that its indicator should not be the sole basis for a finding. If you are facing an accusation, that guidance is worth quoting.
Get a sentence-level account of what fired and why.
Check your draftCommon questions
Detectors measure how predictable your word choices are and how much your sentence lengths vary. Plain, careful, consistent writing genuinely has those properties, so it scores as machine-written even when you wrote every word.
Non-native English speakers are hit hardest, because their writing tends toward narrower vocabulary and steadier construction. Technical and scientific writers, people taught to write short uniform sentences, and users of AI-assisted grammar tools are also at higher risk.
Do not rewrite in a panic, because that destroys the version history that is your best defence. Gather drafts, notes, search history and document version history, then run the text through a second and third detector to show they disagree.
No, they are a design consequence. The detector is measuring a style, and the problem comes when an institution reads that result as proof of who wrote the text.
