False Positive

A false positive is when a checker wrongly signals that something is wrong even though there's actually no problem at all. It's like a smoke detector going off with no smoke around, or a legitimate email getting filtered into the spam folder. The same thing can happen with services that try to spot AI-written text. If a tool scores a sentence a person wrote themselves as "78% AI-generated," that score could simply be wrong — that's a false positive.

The issue is that a false positive doesn't just stay a number. It can cause real harm: a publisher might cancel a contract based on that score, or an author might be unfairly suspected of lying. This matters especially because AI-detection services themselves often rely on AI models internally to make their judgments, so it's important to always remember that these tools aren't perfect.

The article describes a case where the self-published novel Shy Girl was flagged by Pangram as 78% AI-generated, which the author strongly denied. If the author's denial is true, this could be a false positive — a case where human-written text was wrongly judged to be AI-written. The possibility of false positives is exactly why an AI detection score alone should never be treated as "proof" that AI wrote something.

Prepare one paragraph you wrote yourself and one paragraph written with the help of AI. Run both through an AI detection service and compare the scores. If your own writing also gets flagged with a high AI score, you'll get a real sense of how often false positives actually occur.