What to do when a detector flags your writing

A detector score is not proof, and the detector vendors say so. Your job is to show how the text was made. Gather the version history, drafts, notes and sources, in date order. Ask which tool produced the score and read that tool's own documentation, which will describe the score as an indicator. Cite the published false positive research, especially if English is not your first language. Ask for the written policy that governs how scores may be used. Stay calm and specific: the person across the table usually has a number and nothing else, and process evidence beats a number.

Step one: collect process evidence today

Everything that shows the text being made, in order, is evidence. Google Docs keeps a version history under File, then Version history; Word keeps one for files saved in OneDrive or SharePoint. Export or screenshot it before anything is overwritten. Find the drafts, the outline, the notes, the reading list and the browser history for your research. If you used Grammarly's authorship feature or wrote in a document with GPTZero's writing replay, those records show the typing itself. If you used any AI tool for brainstorming or grammar, say so and show what you asked; honesty about limited use is far stronger than denial that later unravels.

Step two: find out what flagged you

Ask which detector was used, what the score was, and on what length of text. Then read the vendor's own guide. Turnitin's AI writing detection guide says the percentage is an indicator that should prompt a conversation with the student, warns that short passages are unreliable, and limits its bypasser detection to English. GPTZero and the others carry similar language. An institution that treats a score as proof is going beyond what the vendor itself claims, and that is worth saying out loud, politely.

Step three: cite the research

Three findings are enough. Liang and colleagues at Stanford, 2023, published in the journal Patterns: seven detectors flagged essays by non-native English speakers at an average false positive rate of 61.3 percent, and more than 91 percent of the essays were flagged by at least one tool. Sadasivan and colleagues, 2023: as model text approaches human text, any detector must trade false positives against misses, so a low miss rate implies false alarms. And the practical point from every vendor's documentation: detectors disagree with each other on the same text, which is why a second detector often clears a passage the first one flagged. Links to all three are in the research section.

Step four: ask for the policy

Most institutions have a written academic integrity procedure that says what evidence is required and what the student's rights are. Ask for it. Many have added language since 2023 saying detector output alone is insufficient; several universities have switched the AI indicator off entirely. If the policy requires more than a score, the conversation changes. If it does not, you have identified the thing to appeal.

How to write the response

Keep it short and chronological. State what the assignment was and when you worked on it. Attach the version history and drafts. State which tools you used and for what, if any. Quote the vendor's own language about the score. Cite the Stanford study if it applies. Ask what further evidence would resolve the question. Do not argue about the detector's algorithm, and do not run your essay through a rewriter to lower the score; that destroys the process evidence that is your strongest asset and looks like exactly what you are accused of.

If you are the one holding the score

The same steps run in reverse. Ask for process evidence before reaching a conclusion. Check the length and language of the submission against the tool's limits. Run a second detector and expect disagreement. Read your own institution's policy. And weigh the cost of a false accusation, which the Stanford authors put plainly: even if the accusation is later withdrawn, the student's reputation has already been damaged.

Questions people ask

Can a Turnitin score alone prove I used AI?

Turnitin's own guide says the score is an indicator that should start a conversation, not a verdict. Ask for the policy that allows a score to decide an outcome.

What evidence shows I wrote something myself?

Version history in Google Docs or Word, dated drafts, notes and sources, browser history of research, and any writing replay. Produce it in order and it is hard to argue with.

I am a non-native English speaker. Does that matter?

Yes. The 2023 Stanford study found detectors flagged non-native essays at a 61.3 percent average false positive rate. Cite it.