Why Some Universities Turned Off Turnitin's AI Detector

Universities turned off Turnitin's AI detector for three main reasons. They couldn't see how it reaches a score, they didn't trust its false-positive rate across tens of thousands of papers, and research showed it flags second-language writers more often. Vanderbilt was the best-known case in August 2023. Since then universities in the UK, the US, Australia and Canada have switched it off, never switched it on, or banned its score as evidence.

Which universities have turned off Turnitin's AI detector?

Ten universities have published a decision to switch off, decline or stop using Turnitin's AI writing indicator. The table lists each one in date order with the reason it gave. Turnitin's similarity (text-matching) check stayed on at every university in the table that said so.

Yale's Poorvu Center also says it doesn't endorse AI detection software or enable such features in Canvas.

The UK opt-outs of April 2023

The first wave came from the UK, within days of Turnitin switching the detector on for everyone on 4 April 2023. Greenwich said it would not adopt new technology without testing it first, especially where it could affect student outcomes. It added that its decision was made alongside other UK institutions, supported by UCISA and Jisc.

Manchester's staff notice gave the timing as a reason too. The release came at the start of the assessment period, so there was no time to evaluate the tool, write guidance or train staff. Manchester also said the opt-out "will not prevent us from enabling this function in the future", so it was a pause and not a permanent ban.

Vanderbilt and the US decisions of 2023

Vanderbilt's announcement is the most detailed. The university said Turnitin switched the feature on with less than 24 hours' notice and with no option to turn it off. Turnitin also said only that the tool looks for patterns common in AI writing, without defining them.

Vanderbilt added two more concerns. Sending student work to a third-party detector raised privacy questions, because its data policies were unknown. And AI detectors have been found more likely to flag text by non-native English speakers. Its conclusion was direct: "we do not believe that AI detection software is an effective tool that should be used."

Georgetown, Montclair State and Alabama followed in the same year. Georgetown's page says its main-campus faculty executive committee and Honor Council raised the accuracy concerns. They held "the belief that the harms of false positives are worse than any advantages of the tool." Montclair's Provost memo cites reliability and bias.

Australia, Canada and the 2025 to 2026 decisions

The second wave came after Turnitin had retuned the detector several times, so these universities decided with two years of evidence behind them. Waterloo's notice cites research on unreliability by Perkins et al., Weber-Wulff et al. and Sadasivan et al., and research on bias against second-language writers. It also says its own IT group tested the tool, found the advantages inconclusive, and saw it flag human-written text as 100% AI-generated in more than one case.

Curtin announced its decision on 4 September 2025 and switched the feature off from 1 January 2026. Its notice recommends writing tools such as Curtin Grammarly and the Authorship function. It also asks students to be open about any AI use, following their unit instructions.

Washington State's memo to instructors is the newest and the most specific, so the numbers deserve their own section.

What reasons did universities give?

Universities gave five reasons, and most gave more than one. The table groups them so you can see which were common.

The rollout reason is worth a note. It applied mainly to 2023, when Turnitin switched the feature on for existing customers with very little notice. Later decisions, such as Waterloo's, Curtin's and Washington State's, rest on evidence from running the tool for a full year or more.

How a small false-positive rate becomes hundreds of papers

Vanderbilt submitted 75,000 papers to Turnitin in 2022. At launch, Turnitin claimed a 1% false-positive rate, so the university worked out that about 750 student papers could have been wrongly labeled as partly AI-written in a single year.

Washington State did the same sum with its own data. Turnitin analyzed 148,547 assessments there in fall 2024. At a 1% rate, almost 1,485 of them were likely flagged as AI-generated when a person wrote them.

The memo adds a second figure. Between 2023 and 2025, 33% of the hearing board's cases about alleged AI use ended in a finding of "not responsible". In those cases the AI detection result was submitted with no other supporting evidence. The memo also says the Turnitin output goes to the instructor and not to the student, so a student can't raise a concern before a referral.

What does Turnitin say about its own false positives?

Turnitin says its document-level false-positive rate is under 1% for documents with more than 20% AI writing, and its sentence-level rate is about 4%. Both figures come from its chief product officer's update of 23 May 2023. The same update says errors are more common where human and AI writing meet in the same passage.

Turnitin's answer to the low end of the range is the asterisk. Scores from 1% to 19% show as *% with no number. Turnitin found a higher incidence of false positives when less than 20% of a document is detected as AI. The detector also needs at least 300 words of prose before it gives a score at all.

Turnitin's May 2023 update also advises educators on using the result. They should "use the AI score to start a meaningful and impactful dialogue with their students." Turnitin also recommends viewing the highlighted passages in totality. Turnitin says the AI score shouldn't be the only basis for action against a student, and Washington State put the same rule in its policy.

Turnitin has also kept changing the detector. It added a category for text likely changed by an AI humanizer in August 2025 and retuned the model in February 2026. It has added languages too, with Arabic in August 2026 and Dutch, German and Turkish in October 2026. A decision from 2023 therefore doesn't describe today's model. Washington State's reasons in 2026 are about how any score is used, whichever model produces it.

What should you do if your university still uses the tool?

If your university still uses the AI indicator, treat the score as an estimate and keep your own record of how the paper was written. Your instructor decides what the number means, and a record of your drafting gives them something to look at besides the percentage.

1. Read your university's AI policy before you start. It tells you whether AI help is allowed, in what form, and how to disclose it.
2. Save your drafts, notes and source list as you go. Version history in Word or Google Docs is enough, and it shows the paper developing over time.
3. If you used AI for any part, say so in the way your course asks. Our guide to AI-use disclosure statements covers wording for manuscripts, and most courses accept a short note.
4. If your score is flagged, ask to see the highlighted passages and the policy that applies. Answer with your drafts and be ready to talk through the paper.
5. If the case goes further, use a written appeal. We have a template in the AI detection appeal letter for ESL researchers.

If you are preparing an appeal, the first court ruling on an AI-detection accusation is a useful reference. Our summary of the Newby v. Adelphi ruling explains what the court decided.

What replaced the AI score at these universities?

Universities replaced the score with evidence of how the work was produced, plus clearer course rules. The common replacements fall into four groups.

1. Clear expectations. Vanderbilt tells instructors that "clearly outlining expectations and guidelines will go a long way", and asks students to disclose and cite AI use in APA or MLA format.
2. Assessment redesign. Vanderbilt suggests in-class writing, discussion of specific class topics and current events that AI tools weren't trained on. Waterloo points instructors to its teaching centre, writing centre and library for assignment redesign and for citing AI use.
3. A conversation with the student. Vanderbilt advises comparing the paper with the student's earlier writing, checking sources and facts, and talking with the student directly.
4. Process evidence. Curtin recommends writing tools such as Curtin Grammarly and its Authorship function, and Turnitin sells Clarity as a record of how a paper was drafted.

Turnitin Clarity became generally available on 15 July 2025. Turnitin's product page says educators can review the whole drafting process, including pasted text, writing time, construction time and draft history. Our explainer on Turnitin Clarity and its policy tiers shows what each tier records and what it means for your paper.

Washington State's policy is the clearest statement of where this is heading. It does not allow any AI detector as the only support for a misconduct case. It tells instructors to re-engineer assessments so they include elements that are secure and AI-resilient. It also points to the University of Sydney's two-lane approach, which separates assessments that are secured against AI from assessments that are open to it.

This is also why we wrote about process being the new proof. A record of tracked changes and dated drafts answers the question a detector can't, which is who wrote the paper and how.

What does this mean for writers in a second language?

Second-language writers are the group most affected, because detectors flag simpler, more predictable English more often, and that describes most careful non-native academic writing. The strongest evidence is Liang et al. (2023), published in the journal Patterns. The authors ran seven detectors on 91 essays by non-native speakers (TOEFL essays) and 88 essays by US eighth-graders. The detectors were close to perfect on the US essays and flagged an average of 61.22% of the TOEFL essays as AI-generated.

All seven detectors flagged the same 18 of the 91 TOEFL essays, and 89 of the 91 were flagged by at least one detector. The authors found that the unanimously flagged essays had lower perplexity, so their word choices were more predictable. Our explainer on perplexity in AI detection covers why that happens, and why detectors flag non-native writers goes through the research in more detail.

Vanderbilt and Waterloo both named this bias in their announcements. Washington State's memo says other studies place false-positive rates higher for students who are neurodivergent or have English as a second language. A university that has switched the score off removes that risk for these students.

If you write in English as a second language and your university still uses the detector, three habits help:

• Keep dated drafts, so you can show the paper growing.
• Use a proofreader to fix grammar and word choice in your own text. ProofreaderPro.ai shows each change, so the final wording is yours.
• If you use AI to draft part of a paper, the ProofreaderPro.ai humanizer polishes it into natural academic prose and is designed to lower AI scores. It keeps your citations in place. Check your university's policy first, and disclose AI use where it asks you to.

Our wider piece on AI detection accuracy in 2026 compares what independent tests found across detectors.

Frequently asked questions

Which universities have turned off Turnitin's AI detector?

Vanderbilt, Georgetown, Montclair State, Alabama, Waterloo, Curtin, Washington State, Greenwich and Manchester have all published decisions, and Australian Catholic University switched its indicator off in March 2025. The table near the top of this post gives each date and the reason the university stated. Yale says it doesn't enable AI detection features in Canvas.

Why did Vanderbilt disable Turnitin's AI detector?

Vanderbilt disabled it on 16 August 2023 because it had no insight into how the tool works and was given less than 24 hours' notice before it was switched on. It also cited privacy concerns, the difficulty of detection as AI tools improve, and evidence that detectors flag non-native English speakers more often. It worked out that a 1% false-positive rate on 75,000 papers meant about 750 wrongly labeled papers a year.

Is Turnitin's AI detector still available?

Yes. Turnitin still sells the AI writing indicator, and it keeps adding languages and retuning the model. Each university chooses whether to enable it, which is why some students are checked by it and others are not.

What does the asterisk next to a Turnitin AI score mean?

The asterisk marks a score between 1% and 19%. Turnitin shows *% with no number for that range because it found more false positives when less than 20% of a document is detected as AI. It also needs at least 300 words of prose to give a score.

Can a university punish me for a Turnitin AI score alone?

Turnitin says its AI score shouldn't be the only basis for action against a student, and several universities, including Washington State, have written the same rule into their policies. Your own university's academic integrity policy decides what happens in your case, so ask for it in writing if you are flagged.

What should I do if Turnitin flags my own writing?

Ask to see the flagged passages and the policy that applies, and answer with your dated drafts, notes and sources. Be ready to talk through the paper with your instructor. Our appeal letter template gives you wording if the matter goes to a formal review.

Does Turnitin's AI detector flag second-language writers more often?

Research says detectors often do. Liang et al. (2023) found an average false-positive rate of 61.22% on essays by non-native English speakers across seven detectors. Vanderbilt and Waterloo cited this bias when they turned the Turnitin detector off.

What do universities use instead of AI detection?

They use clear course rules, disclosure of AI use, redesigned assignments, conversations with the student and process evidence such as draft history. Turnitin Clarity and Curtin's Authorship function are tools for recording that process. Washington State's policy forbids using any detector as the only support for a misconduct case.

Does switching off the AI detector turn off Turnitin's plagiarism check?

No. Curtin, Washington State and Australian Catholic University each said the text-matching similarity check stayed on. Only the AI writing indicator was removed.

Academic AI Humanizer

Polish AI-assisted drafts into natural academic prose, with citations preserved and a setting for academic writing. Designed to lower AI scores on your own drafts.