Accused of Using AI? What to Say and What to Do Next

What to do in the first twenty-four hours

Answer the email. Do not argue in it, and do not confess in it. Those two sentences cover most of what goes wrong on day one.

An accusation usually arrives with almost no detail: a line saying your submission was flagged, a request to meet, sometimes a zero in the gradebook. The instinct is to fire back a long defense. That is a mistake: you are answering a case you have not seen, so you will defend the wrong thing.

Do three things instead. Reply within a day, politely, and say you want to meet. Ask in writing what specifically prompted the concern. Then preserve your files: export the version history and save your outline, notes and sources. Stop editing the original, because later edits muddy the timestamps that are about to become your best evidence.

The clock is real. Arizona State University gives a student five business days to respond to a notice of alleged academic dishonesty. Syracuse University allows seven business days to request an appeal. Miss a window and the process closes over you.

The first email: wording you can adapt

This template is short on purpose. It neither admits nor denies, and it asks for the one thing you need before you can say anything useful. Copy between the markers and replace the bracketed parts.

--- start of email ---

Subject: Re: [assignment name] — request to meet

Dear Professor [Name],

Thank you for telling me about your concern regarding [assignment]. I take this seriously and want to resolve it properly.

Before we meet, could you tell me what specifically prompted the concern: an AI detection report, particular passages, citation problems, or something else? If there is a report, I would appreciate a copy so I can respond to it directly.

I have the version history, drafts and research notes for this assignment, and am happy to walk you through how I wrote it. I am free [two or three specific times].

Thank you,

[Full name, student ID, course and section]

--- end of email ---

That does two things: it puts your request for the evidence in writing, which matters if you later have to show you asked and were not answered, and it signals that documentation exists.

If you used AI in a way the course did not permit, send the same email, then get advice from a student advocate before the meeting. Institutions treat early honesty differently from a denial that falls apart under questioning.

Ask what the evidence is before you explain anything

You are usually entitled to know. The University of New Mexico's student disciplinary procedure gives students "the right to know the nature and source of the evidence used in a hearing process" and "the right to have access to the evidence, subject to some limitations under FERPA and other applicable privacy laws." That is UNM's wording, not a national rule; find the equivalent clause in your own handbook.

Ask three questions. What is the evidence? If it is a detector score, which tool and what did it report? Is there anything besides the score: specific passages, citations, a comparison against your earlier work?

That third question matters most, because a number on its own is thin. Turnitin's own guidance states that its "AI writing detection model may not always be accurate (it may misidentify human-written, AI-generated, and AI-paraphrased text), so it should not be used as the sole basis for adverse actions against a student." Quoting that is not calling anyone a liar. It is quoting the vendor. If the case rests on a score alone, our breakdown of how accurate Turnitin's AI detection actually is is worth reading first.

Understand too why ordinary human writing gets flagged. Careful, evenly structured academic prose sits statistically close to machine output, which is why genuine essays get flagged as AI. A 2023 study by Weixin Liang and colleagues in the journal Patterns tested seven detectors and found they "consistently misclassify non-native English writing samples as AI-generated, whereas native writing samples are accurately identified." If English is not your first language, that belongs in your response.

What the process looks like from here

Terminology differs by institution; the shape is remarkably consistent.

A concern is raised, usually by the instructor. You get written notice. There is a first meeting, often framed as informal. The case then resolves by agreement with the instructor or by referral to a panel or hearing officer. A written decision follows with a sanction, and a short window to appeal on narrow grounds.

The timelines are specific, so copy yours out of the handbook. UNM requires notice of a hearing "at least seven (7) days prior," says a party "may provide relevant information no later than three (3) days prior to the scheduled Hearing," and issues a written decision within fifteen days. Rutgers gives ten working days to appeal. Stony Brook requires appeals "in writing not later than two weeks after receipt of notification of the accusation."

Write down every date the moment you learn it, and ask early which track you are on, since your rights differ.

The rights these policies give you, including the ones students forget

You are presumed not responsible. Rutgers University's adjudication procedures say it plainly: "An accused student is initially presumed to be not responsible for the alleged violation(s)." Washington University in St. Louis promises students "a presumption of non-responsibility."

The burden is not yours. Arizona State's policy states that "the Instructor shall have the burden of showing it is more likely than not that the student engaged in Academic Dishonesty." Preponderance of the evidence is a lower bar than a criminal court's. It is still a bar, and the instructor has to clear it.

You can bring someone. At the University of Texas at Austin an advisor "can be a friend, parent, family member, attorney, or anyone whom the student would like to have present," although "an advisor cannot advocate on behalf of the student during a meeting or hearing." UNM allows one advisor on the same terms. Rutgers offers trained Campus Advisors, "members of the University community trained to aid complainants and accused students navigating the academic integrity process."

Bring someone even if they must sit silent: frightened people remember meetings badly.

Think hard before signing an informal resolution

This is where students quietly lose cases they could have won.

An informal resolution is a signed agreement that a violation occurred, with a sanction attached. It ends things fast, which is what a panicking student wants. What it costs varies. UNC Charlotte is explicit: "If you accept the resolution, the sanction(s) assigned is final and cannot be appealed." Syracuse takes the opposite line: "students whose cases are resolved through a waiver process retain the right to request an appeal."

Read your own policy on this exact point before signing. If the form is in front of you and you do not know the answer, say you would like time to review it; UNC Charlotte builds in three business days for that reason.

Signing is sometimes right: if you did it, if the sanction is proportionate, and you want it finished. Just make it a decision, not a reflex.

How to make your writing process credible

The most persuasive thing you can produce is a record of how the document was built, not an assertion that you built it.

A record cuts both ways, which is exactly why it carries weight. In Harris v. Adams, decided in the federal district court in Massachusetts in November 2024, juniors at Hingham High School were disciplined over an AP US History project. The school's evidence included revision history showing the student "had only spent approximately 52 minutes in the document, whereas other students spent between seven and nine hours," plus footnotes citing books that do not exist. The court refused to undo the discipline at the preliminary stage, and the case remained pending. It was a public high school due-process case, not a university one, so it does not govern your hearing, but it shows how document history reads to someone looking hard.

Build a bundle of your own, organized as one narrative.

Version history, exported or screenshotted, showing the document growing across sessions.

The outline you started from.

Research notes, annotated PDFs, library search history, photos of handwritten work.

Earlier drafts saved as dated files.

Feedback from a tutor, a writing center, a classmate or the instructor.

A one-page timeline: when you started, which days you worked, what changed.

Then rehearse the content: be ready to explain without notes why you structured the argument as you did. Fluency about your own reasoning is evidence no score can touch.

You can also bring a second, independent reading. Run the same text through our free AI detector: no account, no cost, up to 300 words per check. It uses a different model from whatever your institution used, so it cannot tell you what their report says, and one probabilistic result proves nothing on its own. It is still worth having: in the Adelphi case below, the student's own detector evidence went unaddressed, and a court held that against the university.

What to say in the meeting

Calm, specific, short. You are not there to win an argument about machine learning. An opening you can adapt:

"Thank you for meeting me. I want to say clearly that this work is mine, and I would like to show you how I wrote it. I have the version history here, with my outline, notes and earlier drafts. Could I take you through the timeline, then answer any questions about the argument or the sources?"

Walk the timeline. Then stop talking and let them ask.

Four things not to do. Do not attack the detector as a scam; you want the conversation to be about your evidence, not their judgment. Do not bring up other students. Do not invent a detail you are unsure of, because one wobbly claim contaminates everything else. And do not agree to anything on the spot: "I would like to take this away and think about it" is a complete sentence.

Afterwards, email a short summary of what was discussed and agreed. It creates a record whether or not anyone else writes one.

Student advocates and ombuds offices do different jobs

These get confused constantly, and the difference decides who to call first.

A student advocate works for you. At UC Berkeley, the Student Advocate's Office assigns conduct caseworkers, and a student facing an academic misconduct allegation can ask one to "help you prepare for your hearing and attend with you." The same office helps with grade appeal petitions, but that is a separate Academic Senate process, so ask what your advocate can do inside the misconduct process specifically. Many universities run something similar, often through the students' union; search your institution's name plus "student advocate" before assuming it does not exist.

An ombudsperson works for nobody. Ombuds offices operate under principles of independence, neutrality, confidentiality and informality. Rutgers' Office of the Ombudsperson for Students calls itself "an off-the-record, confidential resource," notes that it "does not conduct investigations, adjudicate" and "has no authority to overrule administrative decisions," and adds that "Speaking to the Ombuds Office does not constitute a formal complaint or legal notice to the university."

That is why an ombuds office is often the right first call: you can learn your realistic options without starting anything. Then take the case itself to a student advocate.

Appeals: narrow grounds, short deadlines, and what actually wins

An appeal is not a second run at the same argument. Most policies restrict it to a short list of grounds, and writing outside that list is how students waste theirs.

Rutgers permits four: a conclusion not supported by the facts of the case; procedural error, where "the error must have substantially impacted the outcome"; a grossly disproportionate sanction; and new information, "previously unavailable or unknown at the time of the initial decision," that "must be sufficient to alter the outcome." UNM's four are almost identical. Syracuse allows only two: new, previously unavailable evidence, or failure to follow required procedure.

Note the reversal: Rutgers states that on appeal "the student has the burden of proof." The presumption that protected you at first instance does not follow you upstairs. Structure the appeal around the ground, not your innocence: name it, state the fact that satisfies it, cite the clause breached, and say what outcome you want.

Procedure is where these cases are won. In Matter of Newby v. Adelphi University, decided on 28 January 2026 by Justice Randy Sue Marber in the New York Supreme Court, Nassau County, a freshman was found responsible for improper AI use after a Turnitin report returned an AI writing score of 100. His family ran the essay through other detectors, which indicated human authorship. The court annulled the finding, ordered the record expunged, and called the violation "without valid basis and devoid of reason." Three procedural failures did the work: the administrator "failed to even consider the Petitioner's evidence in the form of the two AI detection programs, which indicated that the Essay was written by a human"; "the Petitioner was not afforded the opportunity to confer with an advisor of their choice regarding the Violation"; and the person who issued the determination was also "the identical person charged with the responsibility of entertaining the appeal," which rendered the appeal right "inconsequential."

None of that turns on whether detectors work. It turns on whether the institution followed its own rules. That is the appeal a student can actually win.

If the finding goes against you

Get the decision in writing with its reasoning. Then ask three questions: what goes on my record, who can see it, and for how long?

Answers vary. Some findings sit only in an internal conduct file. Some attach a transcript notation. Some, like Rutgers' XF grade, carry their own removal process. Ask whether a record-removal route exists after a period of good standing; at many institutions one does and nobody mentions it unless asked.

Be realistic about the tail. UC Davis senior Louise Stivers was flagged by Turnitin's then-new AI detector in 2023 and was eventually cleared, and she still told Rolling Stone that the investigation remained on her record and that she would have to disclose it to law schools and state bar associations. A short factual explanation offered upfront lands better than a discovered omission.

Then deal with the underlying thing honestly. If your prose reads as machine-generated because it is smooth, uniform and cautious, that is fixable, and a pass through our AI word checker shows which stock phrases are doing it. If AI was involved, learn where your institution's line sits: whether using an AI humanizer counts as cheating depends on your course's rules, not on what any tool reports. If assisted drafting with disclosure is permitted, rewrite the draft properly so the words and reasoning are genuinely yours, then disclose as required. What does not work is treating detection as the thing to outsmart. The people deciding your case are humans reading your evidence, and evidence is the only thing that has ever moved them.