Flagged, but you wrote it yourself: preparing a response

The most important thing to know before you write anything is that you are not required to prove a negative. The UK’s Office of the Independent Adjudicator - the ombudsman for student complaints in England and Wales - addressed this directly in a July 2025 casework note on AI and academic misconduct: the burden of proof sits with the provider, not the student.

That is also the vendor’s position. Turnitin states that it does not determine misconduct and provides data for an educator to make a decision under institutional policy. The score is an input. The finding has to be made by people, on evidence.

You are not being asked to demonstrate that you wrote your own essay. The institution is being asked to demonstrate that you did not.

Knowing that changes what your response should contain. It is not a plea. It is a request that the process meet its own standard.

The OIA has upheld student complaints in this area, and the reasons it gave are the most useful map available of where these cases fail.

• The panel could not say what specific evidence led to its conclusion that AI had been used. A score alone did not survive review.
• The university did not properly consider the evidence the student presented, including their essay planning and preparation materials. Provenance was offered and ignored, and that was a procedural failure.
• The university did not consider whether Turnitin’s AI detection might be less reliable for non-native English speakers. Failing to weigh that was itself a ground for upholding the complaint.

The third one is worth restating plainly, because it converts a research finding into a procedural obligation. If you write English as a second language, the published evidence that detectors misclassify non-native writing at high rates is not merely context - an ombudsman has treated ignoring it as a defect in the decision. Times Higher Education reported students winning appeals on these grounds.

Every piece of guidance converges here. Gather the record of how the document came to exist.

• Version history. Google Docs keeps it under File → Version history → See version history, and it can be exported. Word keeps it for files stored in OneDrive or SharePoint, under File → Info → Version History. This is the strongest single artefact because it shows the shape of composition over time.
• Drafts and outlines as separate files, with their original timestamps. Do not tidy the folder.
• Reading notes, annotated PDFs, library loan records, search history in a database. These establish that the argument was assembled rather than requested.
• Anything with a timestamp that predates submission - supervisor emails, feedback on a partial draft, a message asking a friend to look at section three.

All of this only exists if it existed before you needed it. If you are reading this and have not been accused of anything, the useful action is to stop overwriting your drafts.

You are entitled to understand the case. Three requests are reasonable and specific.

• The full report, not the headline percentage. Section-level highlights show where the flag falls. Concentrated in methods, definitions or background - the passages that are supposed to read formulaically - is a different picture from spread through your argument.
• Which detector, which version, and when it was run. Vendors change models; Turnitin’s reporting rules changed in July 2024, and scores below 20 percent are now withheld entirely because of false positives in that range. A report generated under an older rule is not the same artefact as one generated today.
• What the institutional policy says the score means. Many policies treat a flag as a reason to open a conversation rather than as a finding. Ask which yours is.

Keep it factual and short. Three things, in this order: what you are providing as provenance; what you are asking the panel to consider about the detector’s reliability; and a request that the decision state the specific evidence it rests on beyond the score.

Do not rewrite the essay to lower a number, and do not run it through a second detector hoping for a better verdict. Detectors disagree with each other constantly - in the Stanford corpus, 89 of 91 essays were flagged by at least one detector and only 18 by all seven - so a clean second opinion proves nothing and a second flag hands the panel another number.

This page is not legal advice, and processes differ by country and institution. What it describes is what review bodies have said and what evidence has actually mattered.