If your work has been flagged by an AI detector, the single most useful thing to know is that the detector's own vendors and your institution's own guidance agree a detection score is not sufficient evidence for an allegation. That is the ground you are standing on, and it is firmer than it feels right now.
What a detection score actually is
It is a statistical guess about a body of text, produced by a model that was trained on examples. It is not a measurement of your document, it has no access to how the document was written, and it cannot be checked by the person quoting it at you.
Two things follow from that, and both are documented rather than argued:
- Detectors flag human writing. Turnitin's own published false-positive rate — around one document in a hundred — describes tens of thousands of students across its user base.
- They do not flag it evenly. Writers whose first language is not English are flagged at several times the rate of native speakers, which means the tool is hardest on the people least able to argue with it.
- Several universities have turned AI detection off for exactly these reasons.
What to gather, in order
- The file itself, untouched. Do not clean it up, re-save it, or copy the text into a new document — re-saving overwrites the very record you need. Work on a copy.
- Your writing process. Drafts, notes, outlines, messages where you discussed the work, library or reading history, the version history of wherever you wrote it.
- The record inside the file. That is what this site reads: which editing sessions the document records, and which parts of the text belong to which one.
- Your disclosure. If you used a spell checker, a grammar tool, translation help, dictation or an accessibility tool, say so plainly and describe what it did. Permitted assistance described up front is far stronger than permitted assistance discovered later.
What to ask for
Ask what the allegation rests on besides the score. Ask whether the institution's own policy permits a detection report to stand alone. Ask to explain your process — most published guidance expects that conversation to happen, and it is where a real writing history does its work.
What we will not do
We do not sell a way to make writing "pass" a detector, and we will not build one. A tool for disguising authorship is a tool for deceiving somebody, and it would also be worthless to you here: the argument that works is that the accusation is unfounded, not that it became harder to make.