Falsely Accused of Using AI? How to Prove You Wrote It

If you were falsely accused of using AI, the evidence that detector companies themselves point to is your writing process: drafts, version history, notes and outlines, plus your ability to explain the work. Turnitin also says its AI detection "should not be used as the sole basis for adverse actions against a student" (Turnitin).

Key points

• Turnitin and GPTZero say in writing that their AI detectors make mistakes, and OpenAI withdrew its own classifier in 2023 for low accuracy.
• Turnitin's stated document false positive rate of less than 1% applies only to documents with over 20% AI writing.
• In a 2023 study, seven detectors on average labeled more than half of 91 essays by non-native English writers as AI-generated.
• In Google Docs, only people with edit access can browse version history.
• Vanderbilt (2023), UC Berkeley (2025), Waterloo (2025) and the University of Cape Town (2025) each announced they were dropping Turnitin's AI detection.

What AI detector companies say about false positives

Three companies that have built AI detectors say their tools make mistakes and that a result is not proof of misconduct.

Turnitin. As of November 2026, Turnitin's help centre says its model "may not always be accurate" and may misidentify human-written text. Its FAQ adds that "Turnitin does not make a determination of misconduct." In a March 2023 post, its chief product officer told instructors: "If the evidence is unclear, assume students will act with integrity."

GPTZero. GPTZero's FAQ says "these results should not be used to punish students." After a positive detection, it suggests asking the student for "drafts, revision histories, or brainstorming notes."

OpenAI. OpenAI launched an AI text classifier on January 31, 2023. It now says that as of July 20, 2023, the classifier "is no longer available due to its low rate of accuracy." At launch, OpenAI reported that it labeled human-written text as AI-written 9% of the time on an English "challenge set," and that "It should not be used as a primary decision-making tool."

What is Turnitin's false positive percentage?

Turnitin says that for English submissions its document false positive rate is less than 1%, but only for documents with over 20% AI writing. Its FAQ describes this as possibly flagging "one out of every 100 fully-human written documents."

In a May 2023 update, Turnitin said real-world use was "yielding different results from our lab." After testing 800,000 academic samples written before ChatGPT, it found a higher incidence of false positives when it detected less than 20% AI writing. The same post put the sentence-level false positive rate at about 4%, mostly near actual AI-written text.

As of November 2026, Turnitin shows no number for scores between 1% and 19%, only an asterisk, "to call attention to the fact that the score is less reliable" (reports generated before July 8, 2024 may still show a number). For more, see what Turnitin's percentages mean and whether Turnitin can detect ChatGPT.

These figures are not directly comparable with others, because each was measured on different texts, tools and dates.

Why AI detectors flag human writing

Many detectors look for predictable wording (Liang and colleagues), and some human writing is predictable. Turnitin's FAQ says AI picks "highly probable" next words, while human writing "tends to be inconsistent and idiosyncratic." It says false positives can include text without much structural variation, text that repeats itself, or paraphrase "without developing new ideas." See why human writing gets flagged and how AI detectors work.

Non-native English writers

A 2023 paper in Patterns by Liang and colleagues tested seven widely used detectors on 91 TOEFL essays from a Chinese forum and 88 US eighth-grade essays. The US essays were classified accurately, but more than half of the TOEFL essays were labeled AI-generated, an average false positive rate of 61.3%. At least one detector flagged 97.8% of them.

The authors wrote that they "strongly caution against the use of GPT detectors in evaluative or educational settings." These results describe those tools in 2023. Turnitin says it included second-language learners in its training data to minimize bias.

An independent test of 14 tools

Weber-Wulff and colleagues tested 14 tools, including Turnitin, on 54 test cases between March and May 2023, and concluded they were "neither accurate nor reliable." On the 18 human-written or machine-translated texts, false positives ranged from 0% for Turnitin to 50% for GPTZero. The authors caution that Turnitin knew the documents were part of a test. Machine-translating your own writing "can lead to a higher number of false positives."

Universities that stopped using AI detection

Several universities have switched off Turnitin's AI detector and explained why on their own sites.

• Vanderbilt University disabled it in August 2023. It calculated that at a 1% rate, "around 750 student papers could have been incorrectly labeled" from its 2022 submissions.
• UC Berkeley opted out after a pilot, noting that unlike plagiarism matching, "the AI detection tool provides no such conclusive proof."
• University of Waterloo discontinued it in September 2025. Its internal testing found that "in more than one instance the product flagged human written text as 100% generated by AI."
• University of Cape Town announced that the AI Score "will no longer be available at UCT from 1 October 2025."

Can a professor fail you based on an AI detector alone?

That depends on your institution's rules, not the detector's. But a detector vendor, a university faculty and researchers have each said a detector result should not stand alone.

• Turnitin's FAQ says the AI percentage "should not be used as the sole basis for action or a definitive grading measure by instructors."
• Waterloo's Faculty of Science tells instructors that allegations "need to be supported by more than just GenAI detection tool reports."
• Weber-Wulff and colleagues wrote that detector reports "cannot be used as the only basis for reporting students for cheating."

Your school's academic integrity policy and your course's AI rules decide what counts as evidence.

How to prove you didn't use AI to write a paper

You show how the paper came to exist. Turnitin's downloadable guide for students says that if the video playback in Turnitin Clarity is unavailable, you should "share any notes, outlines, version histories, or drafts you used during the writing process." GPTZero's FAQ says that if a document "was typed out with several edits over a reasonable period of time, it is likely the student work is authentic."

How to prove you didn't use AI on Google Docs

Google's help page says: "To browse earlier versions of a file, you need permission to edit that file." Version history shows who updated the file and their changes. A teacher with view-only access cannot browse it, so you can grant edit access or show it on your screen. See what teachers can see in Google Docs version history. Google notes that revisions "may occasionally be merged."

Microsoft Word and OneDrive

Microsoft says "Version history in Microsoft 365 only works for files stored in OneDrive or SharePoint in Microsoft 365." A personal account keeps the last 25 versions; for a work or school account, it depends on the library configuration.

Writing-process tools

These only record writing done while they are in use.

• Turnitin Clarity captures insertions, deletions and paste events in its own writing space, and instructors see the report on submission (Turnitin).
• Grammarly Authorship separates typed from pasted text and offers a shareable replay, but "By default, Authorship in Google Docs and Microsoft Word does not track your writing activity" (Grammarly).

Turnitin's student guide also suggests preparing to explain how you researched, chose sources and evidence, and what you changed and why.

Falsely accused of using AI in college or university: your rights

Procedures differ between institutions, so treat these as examples only.

University of Toronto. Its Code of Behaviour on Academic Matters (effective July 1, 2025) requires an instructor to inform the student "giving reasons, and invite the student to discuss the matter." Nothing the student says in that discussion can be used against them at a hearing. At the dean's or chair's meeting, the student must be told they may seek advice or be represented by counsel, and that statements made there can be used in evidence. A student can appeal a Tribunal conviction (on a question that is not one of fact alone) or a sanction by filing a notice within 21 calendar days of being notified of the decision.

University of Waterloo. Policy 71 entitles students to "a presumption of innocence unless the contrary is established," to know the case against them, and to bring a support person. In a formal inquiry, students have 5 working days to respond in writing, and can appeal within 10 working days of the decision if a ground for appeal can be established.

Both codes make falsified documents an offence. Toronto makes it an offence for a student knowingly "to forge or in any other way alter or falsify any document or evidence required by the University." Neither code mentions writing histories by name, but presenting drafts or a writing history made after the fact as evidence could reasonably be treated as falling within offences like these.

Wrongly accused of using AI: what to do

1. Stay calm and respond. Turnitin's student guide says to "calmly explain your process." Reply within any deadline you are given.
2. Ask what evidence is being used. Ask which tool, score and passages. Turnitin's AI report is hidden from students, but instructors "can download and share the AI report with students."
3. Gather your record. Collect version history, drafts, outlines, notes, sources and feedback from tutors or peers. Do not delete or edit old versions.
4. Offer to talk through the work. Ask for a meeting where you explain your argument and choices.
5. Read your institution's procedure. Find the policy, deadlines and appeal route, and ask whether you can bring an adviser.
6. Keep everything genuine. Share only evidence that already exists.

Frequently asked questions

What is the AI detector false positive rate?

There is no single rate. Turnitin states less than 1% for documents with over 20% AI writing, and OpenAI reported 9% for its withdrawn classifier. Studies found 0% to 50% across 14 tools and a 61.3% average on TOEFL essays, under different conditions.

Can Turnitin give a false positive on human writing?

Yes, and Turnitin says so. Its help centre states that false positives "are a possibility in AI models." Its 2023 testing found more of them when less than 20% AI writing was detected.

Can students see their Turnitin AI score?

Not directly. Turnitin's FAQ says the AI indicator and report "are not visible to students," but instructors can download the report as a PDF and share it. You can ask for a copy.

Should I tidy up or delete my version history?

No. Google says "Deleting version history is permanent and cannot be undone," and that history is the record of how your work developed.

Sources

1. Turnitin: Understanding false positives within our AI writing detection capabilities (16 March 2023)
2. Turnitin: AI writing detection update from Turnitin's Chief Product Officer (23 May 2023)
3. Turnitin Help Center: Using the AI Writing Report (accessed November 2026)
4. Turnitin Help Center: Turnitin's AI writing detection capabilities FAQs (accessed November 2026)
5. Turnitin: AI conversations: Handling false positives for students (downloadable PDF guide) (April 2026)
6. Turnitin Help Center: FAQs for students using Turnitin Clarity (accessed November 2026)
7. GPTZero: Frequently asked questions (accessed November 2026)
8. OpenAI: New AI classifier for indicating AI-written text (31 January 2023, updated 20 July 2023)
9. Patterns, via PubMed Central: GPT detectors are biased against non-native English writers (10 July 2023)
10. arXiv: Testing of Detection Tools for AI-Generated Text (version 2, 10 July 2023)
11. Vanderbilt University: Guidance on AI Detection and Why We're Disabling Turnitin's AI Detector (16 August 2023)
12. UC Berkeley Research, Teaching, and Learning: Availability of Turnitin's Artificial Intelligence Detection (1 May 2025)
13. University of Waterloo: Discontinuing use of AI detection functionality in Turnitin (September 2025)
14. University of Waterloo Science Undergraduate Office: Policies and Absences (accessed November 2026)
15. University of Cape Town News: Introducing UCT's AI in Education Framework (15 July 2025)
16. Google Docs Editors Help: Find what's changed in a file (accessed November 2026)
17. Microsoft Support: View previous versions of Office files (accessed November 2026)
18. Grammarly Support: Introducing Authorship (accessed November 2026)
19. University of Toronto Governing Council: Code of Behaviour on Academic Matters (effective 1 July 2025)
20. University of Waterloo: Policy 71, Student Discipline (last updated 28 February 2020)

Turnitin is a trademark of Turnitin, LLC. Ryne is an independent service and is not affiliated with or endorsed by Turnitin.