A false positive happens when an AI detector flags human-written text as AI-generated. It matters because a result can look more conclusive than the evidence behind it. A flag should lead to a closer review, not an automatic accusation.
If your own writing has been flagged, save the original and follow this checklist. The guidance below explains the result and the accuracy claims you may encounter.
What is the difference between a false positive and a false negative?
A false positive flags human writing as AI-generated. A false negative misses generated writing. A detector can make both kinds of mistake, and a high score cannot eliminate the need to check context.
Look for those two measurements separately when a service publishes results. One overall accuracy percentage can hide the difference between how often human writing is flagged and how often generated writing is detected.
Does a high AI score prove that someone used AI?
No. A score assesses the passage submitted. It does not verify how the passage was written, identify its author, establish intent or determine whether assistance was allowed.
Consider a student report containing a quoted definition, a bibliography and the student's own analysis. If one passage is highlighted, inspect that passage and its source. A highlight does not establish how every section of the report was written.
This is an illustrative review example, not a measured AIScan result.
What does research say about fairness?
Liang and colleagues found that the detectors they studied often misclassified non-native English writing. Their findings are a reason to examine fairness and the writing population in any evaluation. They do not establish an error rate for AIScan or every current detector.
Formal wording, a familiar transition or a polished structure does not establish AI use. Review the argument and the writer's process rather than relying on a style stereotype.
Why should I keep the submitted passage?
A result belongs to the text that was checked. If an excerpt was shortened, a quotation removed or copied text reformatted, the next check concerns a different input.
Keep the original document and the exact submitted passage. If the text came from a PDF or a photograph, check for transcription errors before scanning. This helps everyone discuss the same evidence.
Will another detector settle the question?
Another tool can provide another assessment, but it cannot certify authorship. Conflicting results leave a question to investigate; matching results do not turn a score into proof.
A low result is also not a guarantee that writing contains no AI assistance. Sadasivan and colleagues found reduced detection rates after paraphrasing in systems they tested. That research concerns its tested systems and conditions, not an AIScan-specific benchmark.
What should a fair review include?
Read the highlighted section with the surrounding argument, quotations and references. Ask the writer how they developed the work, then consider any available drafts, notes and version history. Apply the relevant course or workplace policy.
Give the writer time to respond and follow the organisation's review process. A missing draft does not establish AI use, just as a saved draft does not answer every possible policy question.
What accuracy evidence should I look for?
Look for the detector version, test date, sample counts, document origins, languages and lengths. The report should identify the decision rule and separate results for human, generated and edited writing. Small samples and missing results need to remain visible.
AIScan does not currently publish a validated overall text-detection error rate. Read our accuracy guidance before relying on a score. Teachers can use the student-writing workflow, and anyone choosing to run a check can open the text detector.