Loading...
Loading...
Your professor emailed you. Turnitin flagged your essay as 68% AI-generated. The next sentence said the academic integrity office will be in touch. You wrote the essay yourself. You have a draft, an outline, and a Google Doc with three weeks of version history. What do you actually do, in what order, and what should you say?
This guide is the practical playbook. It assumes you are a non-native English speaker, you have been flagged by an AI detector, and you did not submit AI-generated text. The steps are based on what has worked in cases that have gone through formal academic integrity review, including the precedent set in Newby v. Adelphi. It is the action companion to our complete guide to AI detector bias against non-native English writers.
The first 48 hours matter. The instinct most students have is to reply immediately, explain themselves, and offer whatever clarification seems reasonable. That instinct is dangerous.
Do not email your professor with a long emotional explanation. Do not admit to "using AI" if what you mean is that you used Grammarly, spell-check, or ChatGPT for brainstorming. Those are not the same thing as submitting AI-generated text, and conflating them creates a paper trail that is hard to undo.
Do not delete drafts, browser history, or research notes. Those are your evidence. The instinct to "clean up" before a meeting is the worst possible move in a false positive case.
Take a breath. Open a new document. Write down what you actually did, in order: when you started, what sources you read, what tools you used, when you drafted, when you revised, when you submitted. This private timeline is the basis for everything else.
The strongest evidence in a false positive case is not your word. It is the documented process by which you wrote the essay. Gather the following before your first meeting with anyone.
Version history. Open your Google Doc or Word file. Export the full version history. Each revision should show small, incremental changes, not large blocks pasted in at once. AI text that is copy-pasted looks different from text that was typed out sentence by sentence. The pattern of edits matters more than the existence of edits.
Research trail. Download the PDFs you cited. Screenshot the articles you read. Export your browser history for the period you were working on the essay. The point is to show you engaged with sources the way a real researcher does, not the way an LLM produces a summary.
Outlines and notes. If you scribbled an outline on paper, photograph it. If you took notes in a separate document, save it. Even rough notes that do not match the final essay are evidence of an organic writing process.
Earlier drafts. The version you submitted is the polished one. The version before that, messier and less complete, is more useful for an appeal because it shows the thinking, not the result.
Communication. If you discussed the essay with a tutor, peer, or TA, get a short written statement from them confirming what you talked about and when. Email threads count. Slack messages count. Texts count.
Tool usage disclosure. If you used Grammarly for grammar, say so. If you used ChatGPT to brainstorm topics before you wrote, say so. Disclose what you actually used, with specificity. The line you want to draw is between writing assistance and text generation, and the only way to draw it clearly is to be specific.
You are not asking the integrity office to take your word over the detector's. You are asking them to weigh the detector's score against documented scientific evidence that the detector systematically misfires on writers like you. Two pieces of research matter most.
The Stanford Liang et al. study (2023, *Patterns*). This is the most cited peer-reviewed work on AI detector bias against non-native English writers. It found a 53.5% average false positive rate across seven detectors on TOEFL essays written by real humans. One detector flagged 97.8% of the essays. This study is the scientific basis for the claim that AI detector scores on ESL writing are not reliable evidence of misconduct.
The 2026 ToHuman GPTZero replication. A more recent test using GPTZero's production API on informal ESL writing from Reddit language-learning communities found a 16.0% false positive rate, a 1.4x lift over native-English corpora tested on the same model. This shows the bias has not gone away in the three years since the Liang study.
The Newby v. Adelphi decision. A federal court ruled against Adelphi University for relying on AI detection scores as the primary basis for an academic misconduct finding. The case is now cited as precedent in similar disputes. You are not asking the integrity office to overturn detector use generally. You are asking them to follow the standard Newby established: detector scores alone are not sufficient evidence.
Bring printed copies to your meeting. The integrity office is more likely to engage with a paper trail than with URLs.
Your appeal letter should be short, factual, and structured. Three paragraphs.
Paragraph one: what happened. State the essay title, the course, the submission date, the detector score you received, and the specific allegation. Keep it factual.
Paragraph two: what you actually did. Describe your writing process with specifics. When you started, what sources you read, what tools you used and for what, how many drafts you wrote, who you talked to. Do not editorialize. Just describe the process.
Paragraph three: the evidence and the research. List the evidence you are providing, version history, research notes, earlier drafts, communications. Cite the Liang study and the Newby v. Adelphi decision. State that AI detector scores on ESL writing have a documented false positive rate too high to be used as sole evidence of misconduct. Request a human review of your full writing process, not just the detector score.
A template:
> Dear [Name], > > I am writing to appeal the academic integrity finding regarding my essay [title], submitted on [date] for [course]. The finding is based on an AI detector score of [X]%. I did not submit AI-generated text. I wrote the essay myself. > > My writing process: I began research on [date], reading [specific sources]. I drafted an outline on [date]. I wrote the first draft between [dates], revising it across [number] versions. I used [Grammarly for grammar / ChatGPT to brainstorm topics before writing], and I have disclosed this specifically. I did not use AI to generate the text I submitted. > > I am attaching: my full Google Docs version history; my research notes; an earlier draft dated [date]; a statement from [tutor/peer] confirming our discussion on [date]. > > AI detector scores on non-native English writing have a documented false positive rate of 53.5% (Liang et al., *Patterns*, 2023), with one detector flagging 97.8% of human-written TOEFL essays. The Newby v. Adelphi federal ruling established that detector scores alone are insufficient evidence for academic misconduct findings. I respectfully request a human review of my full writing process, not just the detector score. > > Sincerely, > [Your name]
Adjust the specifics. Send it through the official channel your institution specifies, and keep a copy.
If your case goes to a hearing, three things matter.
First, do not over-claim. You do not need to argue that detectors are useless. You need to argue that the specific score on your specific essay is not reliable evidence. The difference matters. Over-claiming weakens credibility. Narrow arguments win.
Second, bring a witness if you can. A tutor, a peer who reviewed your draft, or a faculty member who knows your writing. A witness who can speak to your writing process is more persuasive than a written statement.
Third, ask the integrity office for their false positive rate. They almost certainly do not know. The act of asking puts the burden back on them to justify the detector's use, which is the right framing.
If the internal appeal is denied, you have options outside the institution. The Office for Civil Rights at the US Department of Education accepts complaints regarding disparate impact on the basis of national origin, which is the framework the Doe v. Yale case is using. A complaint to OCR does not require an attorney and can trigger an investigation.
A second option is to consult an education attorney. Many offer free initial consultations. The Newby precedent has made several firms more willing to take academic integrity cases involving AI detection.
Neither path is fast. Both are real.
A short list of moves that make appeals harder:
A false positive is not a verdict. It is a starting claim that you can contest with evidence, research, and a structured response. The system is not set up to handle ESL writers fairly, but it is set up to respond to documented authorship processes. Your job is to make yours visible.
*This is the appeal companion to our complete guide to AI detector bias against non-native English writers. For the technical mechanism behind why TOEFL writing gets flagged, read Why AI Detectors Flag TOEFL Essays as AI. For writing techniques that reduce your statistical risk, read ESL Writing Tips to Reduce AI Detection False Positives. For institutional context, see Universities Dropping AI Detection in 2026.*
Humanize AI text to sound naturally human with EvalHub.
Start Free Trial