Being accused of using AI in college is frightening, especially when you wrote the work yourself. The good news is that an accusation is the start of a process, not a verdict, and that process gives you specific rights. This guide walks through what to do in the first 48 hours, who actually handles the case, what you can expect at a meeting or hearing, and what to do if the outcome feels unfair.
If you have just been flagged and you are still trying to catch your breath, our overview hub on being falsely accused of using AI is the calmer place to start. This page is the procedural one: how the institution's machinery works and how to move through it without making your situation worse.
A detector score is not a finding of misconduct. It is a number that prompted a human to ask a question. At almost every accredited college, a flag from Turnitin, GPTZero, or any similar tool cannot by itself decide your grade or your standing. It has to enter a defined integrity process, where a person reviews it, where you get to respond, and where someone other than the software makes the final call.
Turnitin says this plainly in its own guidance: its AI writing score "should not be used as the sole basis for adverse actions against a student." That sentence is worth memorizing, because it is the institution's own vendor telling faculty not to treat the percentage as proof. Several universities went further and turned the feature off entirely. Vanderbilt was among the first to disable Turnitin's AI detection, citing reliability and fairness concerns. You are not arguing against settled science. You are responding to a probabilistic tool that the field itself treats with caution.
So the first mental shift is this: you are not guilty because a tool said so. You are being asked to participate in a fair process, and your job is to participate well.
The early hours matter more than almost anything else, mostly because of evidence. Move calmly, but move.
Anxiety pushes people to over-explain, apologize for things they did not do, or agree to a quick resolution just to make the discomfort stop. Resist that. A panicked admission to "make this go away" can be treated as a confession even when you did nothing wrong. You are allowed to say, "I did not use AI to write this, and I would like to understand the process before I respond in detail." That is a complete and reasonable answer.
This is the single most important practical step. The strongest evidence that you wrote your own work is the record of you writing it. Before you do anything else, protect it.
Do not edit, "clean up," or reorganize these files. Their value is that they are untouched. We go deeper on building this record in our guide to how to prove you didn't use AI, and it is worth reading before your first meeting.
You cannot defend against an accusation you cannot see. Politely ask, by email so there is a record, for the details:
A vague "the system flagged your essay" is not enough to act on. You are entitled to know what you are responding to, and asking for it in writing is normal, expected, and not an admission of anything.
One of the most common mistakes is arguing with the detector or the company that makes it. Turnitin and GPTZero are vendors. They do not adjudicate your case, set your school's policy, or have any say in your outcome. Do not email them. Your case lives inside your institution.
Your learning platform is not the source either, and establishing which product actually produced the number is a fair first question to ask. Canvas does not detect AI and neither does Moodle; both pass submissions to an outside vendor, or to nobody at all. Blackboard's SafeAssign is a similarity checker rather than an AI detector, and Anthology has published its decision not to build one. Knowing which company issued the score tells you which published error rate applies to it.
The exact owner varies by school, so check your student handbook or the integrity policy on your college website, but it usually flows like this:
Your handbook is the authority here, not this page. Read the section on academic integrity and AI before your first conversation so you know which stage you are at and who has the decision.
Most institutions, especially those bound by due-process expectations, give accused students a recognizable set of rights. Yours may be named differently, but look for these in your policy.
If your school is skipping any of these, that itself is a procedural problem worth raising calmly and in writing.
The first contact is often an informal meeting with your instructor or an integrity officer. A formal hearing, if it gets that far, is more structured, but neither is a courtroom. The tone is closer to a serious academic conversation, and preparation is what changes the outcome.
Walk in with an organized file, not a verbal protest. A strong pack usually includes:
A reasonable reviewer may ask you to talk through your own paper. This is one of the fairest checks available, because someone who genuinely wrote a piece can explain their argument, their sources, and why they made certain choices. Re-read your own work beforehand. Be ready to explain your thesis, define any term you used, and describe how you got from your research to your conclusions. Confident, specific answers about your own writing are powerful evidence in your favor.
Be polite, factual, and unrattled. You are not begging; you are demonstrating that the work is yours. Stick to evidence and process. Avoid attacking the professor personally, and avoid sweeping claims that "detectors are always wrong." The stronger move is precise: this specific result is unreliable, and here is the record of me doing the work.
People often ask whether running their paper through another AI detector will clear them. It is worth being honest about what that can and cannot do. No detector can prove you did not use AI, because no tool can prove a negative, and every detector is probabilistic, including ours. What a second, independent check can do is add a data point.
If one tool flags your work at a high percentage and a second, independent tool returns a low score on the same text, that disagreement is meaningful. It shows the result is not stable across tools, which undercuts the idea that the original flag is reliable. Used that way, alongside your version history and your ability to defend your work, a second opinion supports your case. It is not the proof. Your writing process is the proof.
TextSight's AI detector gives you a probability score with sentence-level highlights, so you can see exactly which lines another model reads as machine-like and bring that breakdown to your meeting. We are upfront about the limit: it is a second opinion, also probabilistic, and false positives are real, particularly for non-native English writers. The reliability data backs this up. A 2023 Stanford study (Liang et al.) found that seven GPT detectors flagged essays by non-native English writers as AI at an average false-positive rate of 61.3 percent, against a near-zero rate for native writers in the same study. Turnitin documents two rates of its own: under 1 percent at document level above its 20 percent reporting threshold, and roughly 4 percent across the full distribution of documents. Those numbers are the strongest reason to insist on human review rather than a tool verdict.
We publish the same kind of number for our own detector rather than only citing other people’s. We ran TextSight over 1,180 human-written papers, all published in 2018, years before ChatGPT existed, so every AI flag is a false positive by construction. It flagged 5.85 percent of them. We also looked specifically for the penalty against non-native English writers that the Stanford study found in other tools, and did not find one in ours: the gap between the two groups was −3.5 percentage points with a 95 percent confidence interval of −7.8 to +0.7, which contains zero. That means no significant difference between the groups — not an advantage for either, and not a claim that our detector is unbiased. The scope limit matters and we would rather state it plainly than have you over-read the result: those are academic abstracts of 150 to 400 words, from a single year, scored by one version of our detector. It does not transfer to a student essay. The dataset, the per-document results and the analysis script are all published, so you can re-run the numbers rather than take our word for them.
Sometimes the process gets it wrong. If you are sanctioned despite genuine work and solid evidence, you usually still have options.
Persistence inside the proper channels, backed by your evidence, is what tends to turn an unfair early decision around. The process exists precisely because tools are fallible, and using it fully is your right.
Once the case is behind you, the useful next move is making the next paper harder to misread. Our founder wrote a free book on writing with AI without losing your voice, including how to keep a draft history that speaks for itself.
TextSight is an AI-detection and writing-trust tool. We help you check and defend your own work; we do not help anyone disguise AI writing or evade an institution's detection. No detector can prove who wrote a piece, and all detector scores, including ours, are probabilities that can be wrong, especially for non-native English writers. This page is general guidance, not legal advice; for a formal or high-stakes case, consult your student handbook and, where appropriate, a qualified advisor or attorney.
TextSight's free tier gives you daily scans with sentence-level highlights, so you can see which lines carry the AI signal and why. Every score is a probability, not proof, and we say so plainly.