Can AI Detectors Produce False Positives?

An AI detector compares writing against patterns associated with generated text. That is a statistical judgement about style, not a record of how the document was created. Any method that infers authorship from style alone will sometimes be wrong.

The reverse is also true: text produced with AI assistance and then edited by a person can lose the patterns a detector looks for.

Results are more fragile when there is less text to analyse or when the writing is unusually regular.

• Very short passages, where a few sentences carry the whole judgement.

• Formulaic academic or technical prose that repeats predictable phrasing.

• Heavily revised text where the original patterns have been removed.

• Writing that uses less common phrasing, including some second-language writing.

• Documents that mix human and AI-assisted sections, which can produce a single misleading score.

Treat the result as a reason to look closer. Ask for drafts or version history, compare the flagged passages with the cited sources, and give the writer a chance to explain their process. A conversation usually resolves what a percentage cannot.

Institutions should also document how AI signals are used, so decisions are consistent and the writer knows what evidence is being considered.