Accused of Using AI on an Assignment? Read This Before You Respond

An email arrives: your paper was “flagged for AI.” A meeting is scheduled. Maybe the notice cites Turnitin’s AI indicator, maybe a professor’s gut, maybe a GPTZero screenshot. Before you write back, before you explain, before you “just clear this up” — read this. What you do in the next week matters more than anything you did while writing the paper.

What an AI detector actually measures

AI-writing detectors do not detect AI. They estimate statistical patterns — how predictable your word choices are, how uniform your sentences run — and produce a probability score. That distinction decides cases. A probability is not an eyewitness; it is a guess with a percentage attached, produced by tools whose error rates their own makers acknowledge, that flag non-native English writers, technical writing, and heavily edited prose at higher rates. No detector vendor will testify that its score proves your paper was machine-written, and most publish disclaimers saying scores should not be used as sole evidence of misconduct. Your school’s case frequently rests on a number the number’s own manufacturer disowns.

The burden problem — and why schools skip past it

In an honor-code case the school carries the burden of proving misconduct, usually by a preponderance of the evidence. A bare detector score plus a professor’s suspicion is a thin record. But process moves fast, panels are not statisticians, and a confident accusation can shift the burden in the room: suddenly you are being asked to prove you didn’t use AI. Naming that move — politely, in writing — is often the single most effective thing a response can do: the policy defines the violation, the school bears the burden, and a probability score from an unvalidated tool does not carry it.

Why “just explain your process” is a trap

The first meeting is usually pitched as informal — “just walk us through how you wrote it.” Understand what that meeting is: evidence collection. Innocent students hurt themselves in it constantly. They misremember dates, guess at details, minimize (“I only used Grammarly”), or over-explain until something contradicts a timestamp. Every inconsistency becomes “shifting explanations” in the final report. You are generally entitled to know the specific allegation and see the evidence before you respond substantively — ask for both, in writing, and prepare your account against your documents before you give it. Preparation is not guilt; it is what every institution does before its meetings.

Your best evidence already exists — preserve it tonight

False-positive cases are won with artifacts, not adjectives:

• Version history. Google Docs and Word (with AutoSave) keep time-stamped edit histories showing a document growing over hours and days — the single most persuasive exhibit a wrongly accused student can produce. Export it now; do not keep editing the file.
• Drafts, outlines, and notes — handwritten pages, annotated sources, earlier files, even texts to a friend about the assignment.
• Research trail — browser history, library database logs, PDFs with highlights.
• Your writing baseline — earlier graded papers in your voice, which make “this doesn’t sound like you” testable instead of vibes.

Preserve everything before you respond to anyone, and preserve it outside school-controlled accounts — portal access has a way of vanishing mid-dispute.

What honest use looks like in the gray zone

Many cases are not fabrication cases at all — they are policy-line cases: Grammarly’s rewrite suggestions, a brainstorming chat, translation help, an outline generated then written over. Course policies vary wildly, syllabi contradict university-wide rules, and “unauthorized assistance” often goes undefined. If your use sat in a gray zone, the response strategy is different — definition, notice, and consistency of enforcement — and admitting to “using AI” in the abstract, without that framing, converts a defensible case into a signed confession. This is exactly where a misconduct-defense review earns its keep before the meeting, not after.

Sanctions are not standardized — and that is leverage

Outcomes for the same allegation range from a warning, to a zero on the assignment, to course failure with a transcript notation, to suspension — sometimes at the same school in the same semester. How comparable cases were handled, what the syllabus actually authorized, and whether the process followed the school’s own written rules all bear on sanction even when the finding itself will not move. A transcript notation is the difference between a bad week and a disclosure obligation that follows you onto every graduate, medical, and bar application — fight hardest there.

When to bring in counsel

Earlier than feels natural. The highest-value window is before your first substantive response, when the record is still yours to build: the preservation steps above, the written request for the evidence, the framing of your account, the comparator questions. Most of that work is invisible to the school — a flat-fee record review with the file and the syllabus tells you whether this is a fight, a negotiation, or a misunderstanding, and the engagement terms are published so you know exactly what you are buying. If the finding is already in, appeals run on short windows measured in days — move.

This is general information about matters I handle, not legal advice about your situation. AI-detection policies and honor codes differ by institution and change frequently. Last reviewed August 2026.