False positive

In plain English

The detector says machine. The truth is human. That gap is a false positive.

It is not a rare edge case in this category. It is the central reliability problem, and it falls unevenly on particular kinds of writer.

A worked example

What a false positive looks like in practice is ordinary, careful, slightly plain writing — not anything a reader would find odd.

The experiment was conducted over a period of six weeks. Data was collected
at the end of each week. The results were then analysed using standard
statistical methods. The findings are presented in the following section.

A person wrote this. It is a methods paragraph doing exactly what a methods paragraph is supposed to do: state what happened, in order, without decoration.

Every property that makes it good methods writing also makes it score as machine-written. Uniform sentence length, no specifics beyond the ones required, no authorial voice, entirely predictable word choice.

The writer cannot fix this without making the methods section worse. That is the shape of the problem — the flagged features are often the features the genre demands.

Why it matters for AI detection

The measured rates are not small. In peer-reviewed testing published in Patterns, seven detectors misclassified 61.22% of essays by non-native English speakers as AI-generated, while classifying essays by native speakers accurately. The bias is systematic, not random noise.

OpenAI withdrew its own classifier in July 2023 after it identified only 26% of AI-written text correctly while flagging human writing as AI 9% of the time.

This is why a detection score is evidence to investigate rather than a finding of fact, and why several universities have disabled detection tools outright rather than defend the resulting accusations.

Commonly confused with

False positive rate

A false positive is one document wrongly flagged. The false positive rate is the proportion across a set — and a rate that sounds tiny still produces a large number of accusations at institutional scale.

Plagiarism match

A similarity match says your text resembles a source in a database. A false positive on AI detection says your text resembles machine writing in general. Different systems, different evidence, frequently confused in the same report.

Read next

• Why AI detectors produce false positives
• If you have been falsely accused
• How to prove you wrote it
• Hedging
• Detection outside English