Flagged, Not Caught: Challenging an AI Detector’s Verdict on Your University Essay

Getting an email that says your essay has been “flagged” can feel like being accused of cheating in a language you didn’t know existed. The message may be vague, the evidence may be thin, and the whole thing may arrive just before a deadline or dissertation hand-in. Breathe. A flag is a signal, not a verdict. In UK higher education, misconduct still has to be investigated by people, against a policy, with a chance for you to respond.

That does not make the experience harmless. AI detectors are probabilistic tools. They produce scores, not proof. Some are better than others; almost none publish robust, independent false-positive rates for the exact kind of writing they are asked to judge. And when a detector gets it wrong, the consequences can be serious: a delayed grade, a misconduct panel, a cloud over your academic record.

First, what a flag actually means

Most detectors look for statistical patterns — often described as perplexity and burstiness — that supposedly distinguish human writing from machine output. In practice, they are comparing your work with models of what “typical” text looks like. That makes them sensitive to style, structure and subject. A lab report, a legal problem question and a reflective essay do not read the same. Neither do students.

A flag usually means one of three things: the detector gave your text a score above a threshold; a marker noticed something and ran it through a tool; or a third-party service was used as part of a wider check. It does not mean the university has already decided you cheated. But it does mean you are entitled to ask how the decision was reached.

Why some students get hit harder

False positives are not spread evenly. Neurodivergent students may write in highly structured, consistent prose, or use assistive technology that shapes phrasing. Mature students may have developed a formal, polished academic style after years away from study. Students who speak English as an additional language often learn formulaic academic conventions — signposting, repeated sentence frames, cautious hedging — because that is what the style guide asks for. Detectors can mistake consistency for automation.

There is also the editing problem. Using grammar tools, translation aids or predictive text is not the same as generating an essay. But a detector cannot reliably tell the difference between “I used a tool to tidy my sentences” and “a tool wrote this”. That is why a human review matters, and why you should not accept a score alone as evidence.

What your university should be able to tell you

Universities in the UK are not obliged to publish every internal detector threshold, but they should be able to explain the process. Ask which tool was used, which version, what score was produced, and what the provider says about its error rate. If the university cannot answer basic questions about the evidence, that is a problem with the case against you, not with you.

Under UK data protection law, you can also make a subject access request for personal data held about you, including assessment records. That is not a magic wand, and it will not force the university to hand over someone else’s notes, but it can surface the evidence bundle. Many institutions also have a formal request route for assessment feedback and appeals. Use it.

The challenge playbook

1. Freeze the scene

Do not delete drafts, version history, notes, reading lists, supervisor emails or earlier feedback. If you used spellcheck, a reference manager or a translation tool, note that down. Keep the original file and any backups. The strongest defence is often a paper trail showing how the work developed.

2. Ask for the tool, version and error rate

Send a short, calm email to your module leader or course administrator. Ask which detector was used, who ran it, what threshold was applied, and what the provider states about false positives. You are not asking for a favour; you are asking for the basis of an accusation.

3. Request the evidence bundle

Ask for the flagged passages, the scores, any human reviewer notes and the policy that was applied. If the university says the tool is confidential, ask for the relevant extracts or a summary. You need enough detail to respond meaningfully.

4. Read the policy, then ask for a human

Find your institution’s academic misconduct policy. Look for wording about AI tools, evidence standards and the right to respond. Ask whether the detector was used as the sole evidence. Many policies say it should not be. If a human academic has reviewed the flag, ask for their reasoning in writing.

5. Use the academic appeals process

If the flag turns into a formal misconduct finding, follow the internal appeal route. Stick to deadlines, keep copies of everything and focus on procedure and evidence: was the tool reliable, was the policy followed, were you given a chance to explain? Do not turn the appeal into a general complaint about the course.

6. Escalate if the process stalls

For higher education in England, the Office for Students sets conditions but does not usually hear individual appeals. The Office of the Independent Adjudicator is the usual external route once internal processes are exhausted. Ofqual regulates exam boards and qualifications in England, not university degrees, so it is relevant if your issue concerns a regulated qualification rather than a university module. In Wales, Scotland and Northern Ireland, different regulators and complaint routes apply.

Template email to a module leader

Subject: Request for information about AI-detection flag on [module/assessment]

Dear [Name],

I am writing about [assessment] submitted on [date]. I have been told it was flagged by an AI-detection tool. I would like to request:

• the name, provider and version of the tool used;
• the specific score or flag and what it means under the university’s policy;
• the tool’s stated false-positive rate and any validation data the university holds;
• the full evidence bundle, including flagged passages and reviewer notes;
• the name of the human academic who reviewed the flag and their reasoning;
• a copy of the relevant policy and the appeals route.

I am not asking to bypass academic judgement. I want to understand and respond to the evidence. Please confirm receipt within [X] working days and let me know if a meeting is possible.

Kind regards, [Name], [Student ID], [Course]

The bigger point

AI detectors can be useful as one signal among many. They are not a judge, a jury or a reliable lie detector. If you are flagged, the practical move is not to panic or to fire off an angry email. It is to ask for the evidence, insist on a human decision and use the process you are entitled to. At AIWF, that is the theme we keep coming back to: when software makes a claim about you, make it show its working.