How to Prove Your Writing Is Human If You're Falsely Accused of Using AI

If a detector flags work you wrote yourself, your strongest defense is evidence of how you wrote it: version history, drafts, notes and sources, plus a calm written account of your process. Turnitin itself says its AI score should not be the sole basis for action against a student, and in England and Wales the student complaints ombuds says the burden is on the institution to prove misconduct, not on you to disprove it. Here is what to gather, what to ask for, and how to appeal.

What an AI score can and cannot prove

Start with what the detector's own maker says. Turnitin's AI detection FAQ states that "Turnitin does not make a determination of misconduct" and that the AI percentage "should not be used as the sole basis for action or a definitive grading measure by instructors." Its release notes of 15 December 2023 say it more bluntly: "an AI Writing score should not be used as the sole basis for adverse actions against a student."

Other rules from the same documents that can matter in your case:

Low scores are hidden. Since 16 July 2024, results from 1% to 19% show only as *%, with no highlights, because of the risk of false positives (release notes).
Short work is less reliable. A report needs at least 300 words of prose, and in documents of a few hundred words the prediction is "mostly 'all or nothing'" (FAQ).
The error rate is small, not zero. Turnitin aims for under 1% false positives on documents with over 20% AI writing (FAQ). Vanderbilt worked out that at 1%, about 750 of the 75,000 papers it submitted in 2022 could have been wrongly labeled, and switched the detector off on 16 August 2023.
Some writers are flagged more. In Liang et al. (2023), seven detectors (not Turnitin) labeled an average of 61.22% of TOEFL essays by non-native English speakers as AI-written. If English is not your first language, read AI detection and ESL students.

Other detectors have their own thresholds and limits, which is why the first thing to find out is which tool was used. The research and documented cases are collected in our explainer on AI detection false positives.

Your authorship evidence pack

Build this before anyone asks. If you have already been flagged, gather as much of it as still exists.

Evidence	What it shows	How to get it
Google Docs version history	Gradual drafting, with timestamps	At the top right, click "Last edit" (Google help). Name versions at milestones so Google does not merge them.
Word version history	The same, for Word	Works only for files stored in OneDrive or SharePoint: click the file's title, then Version history (Microsoft support).
Saved drafts	How the argument developed	Keep dated, labeled files, such as Essay1_outline and Essay1_draft1.
Notes and outlines	Thinking before the writing	Brainstorms, outlines, concept maps, photos of handwritten notes.
Sources	Real engagement with the material	Annotated PDFs, library loans, the pages you quoted.
Browser and search history	When and how you researched	Your browser's history page; screenshot the relevant dates.
A process-recording tool, if you use one	Typing, pasting and AI use over time	Grammarly's free Authorship feature records whether text was "typed, pasted, AI-assisted", in Google Docs, Microsoft Word and the Grammarly Editor.
Disclosure of permitted tools	Honest use of allowed help	A note of any translator, grammar checker or AI tool your course allows, and where you used it.
Earlier graded work	Your normal voice and level	Past essays from the same course or year.
A writing-process statement	Your account, in your words	A short paragraph: how you started, what changed as you revised, where you struggled.

A few practical notes:

Google Docs only lets people with edit permission browse version history (Google help), so plan to show it from your own account on screen or in a meeting.
Evidence like this has cleared students before. A UC Davis student was cleared after giving officials "step-by-step instructions on how to open Google Docs and review history" (Rolling Stone). At Johns Hopkins, a student's drafts and "PDFs with highlighter over them" convinced the instructor the flag was wrong (The Markup).
The University of New Hampshire's handout on documenting your writing process adds good habits: write directly in the document instead of pasting large blocks, keep the file in one place so its history stays intact, and keep earlier versions instead of overwriting them.

Two things not to do

Do not run finished work through a rewriting tool to make it "look human." The same UNH handout lists paraphrasing and "humanizing" tools among those to avoid unless your instructor allows them, because they can "eliminate visible drafting and revision history." Full disclosure: SupWriter makes an AI humanizer. If you wrote the work yourself, a humanizer is not your defense; your process evidence is.

Do not create evidence after the fact. Backdated or reconstructed "drafts" turn a weak allegation into a real misconduct case. Show what exists, and explain gaps honestly.

If you have been flagged: step by step

1. Read the allegation and answer calmly, in writing

A flag is not a finding. In England and Wales, the Office of the Independent Adjudicator (OIA) says a student suspected of AI misconduct "should be told in writing what it is that they may have done that is a breach of the assessment rules, and why the provider thinks this" (OIA casework note, published 15 July 2025). Wherever you study, ask for the same.

2. Request the report and all the evidence

Ask which tool was used, for the full report, the exact score, the highlighted passages, and anything else the decision relies on. Turnitin's AI report is not visible to students, but instructors can download it as a PDF and share it (FAQ; see also can students see their Turnitin AI score?). The OIA says students should be "provided with all relevant evidence, including detection software reports."

3. Check the score against the tool's own rules

Was the document at least 300 words of prose? Is it in a language the detector supports? Turnitin covers only long-form English, Spanish, Japanese and Modern Arabic (FAQ). Was the result a *%, meaning 1% to 19%? Turnitin put its sentence-level false positive rate at about 4% in May 2023, so single highlighted sentences deserve particular scrutiny (Turnitin). For what different scores mean in practice, see what percentage of AI is acceptable on Turnitin.

4. Walk through your evidence in order

Present your pack chronologically: first notes, outline, drafts, revisions, final version. Point to specific moments in the version history where the flagged passages took shape.

5. Offer to discuss the work

The OIA's casework note says providers "may decide to carry out a viva" (an oral discussion of the work) and should "give students the opportunity to explain how they worked to prepare their submission." Offering this yourself shows confidence. Providers may also compare the work with your earlier assessed work, and the OIA says they should tell you which pieces, so you can comment.

6. Know your rights under the policy, and get support

Find your institution's academic integrity or misconduct policy and read it before any meeting. Our university AI-detection policy tracker links the published positions of 148 institutions, including whether they use a detector at all. Look for the right to bring an advisor, the standard of proof, and how appeals work. In a 2026 New York case, Adelphi University's own code guaranteed an "advisor of choice" and a "meaningful opportunity to be heard" (court decision). Many universities have staff who help accused students; the Colorado School of Mines, for example, tells students they "can also meet with Community Standards staff as they navigate a situation where they are accused" (Mines). The OIA suggests a students' union adviser or equivalent.

7. Appeal, then escalate if you need to

Use the internal appeal. In England and Wales, once the internal procedures are finished your university should issue a Completion of Procedures Letter, and the OIA must receive your complaint within 12 months of that letter (OIA). Elsewhere, look for an ombuds office or your institution's equivalent. Courts are a last resort, but they exist: on 28 January 2026 a New York court annulled an Adelphi University finding that began with a Turnitin result of 100% AI-generated and ordered the student's record expunged (decision).

How to write your response or appeal letter

Keep it short, factual and organized:

Your position. Name the assignment and the decision, and state plainly that you wrote the work yourself.
Your evidence. Walk through the pack in order, with screenshots or attachments labeled by date.
Why the score alone is not enough. Quote the vendor: Turnitin says its percentage "should not be used as the sole basis for action." Mention any condition your case falls under, such as a short text, a *% result or writing in a second language, and cite the research.
What you are asking for. Withdrawal of the allegation, removal of any grade penalty, and no record of the incident.
An offer. Say you are happy to discuss the work or answer questions about it.

Avoid attacking the instructor. The strongest letters make it easy for a reviewer to say yes.

What decided real cases

UC Davis, 2023. Clearing Louise Stivers took more than two weeks, and Google Docs history is what did it (Rolling Stone).
England, July 2025. Turnitin indicated a high proportion of AI-generated content in a postgraduate student's essay, which also contained hallucinated references. The student denied using AI and supplied the correct sources and earlier versions of the essay. The provider upheld the allegation, but the OIA was concerned it "had not shown how it had engaged with the student's evidence", and the provider agreed to reconsider the case afresh (OIA case summary). The lesson: make the institution respond to your evidence in writing.
New York, January 2026. The Adelphi finding was annulled because the process failed: the professor's own later email undermined the allegation, the student was not allowed to consult an advisor, his evidence was not considered, and the same official decided the case and the appeal (decision).

None of these outcomes turned on proving that a detector is broken. They turned on evidence and fair process, which is exactly what your pack gives you.

Pre-checking your own work

Running a draft through an AI detector before you submit, SupWriter's included, can show which passages one model reads as formulaic. It cannot tell you what your school's detector will say, and it can be wrong (how we test). If a passage flags, the useful fix is usually the honest one: add the specific example, source or argument that only you would write. And keep the version history that shows you did.