Turnitin AI detection review

Best for

Instructors and integrity officers at institutions that already pay for Turnitin Originality and want a second signal on long-form prose. The report is strongest as a map: it shows which sentences the model scored as likely AI, which helps a teacher decide what to ask about in a meeting. It is weakest as a verdict.

Main limitation

The score is a model estimate, not evidence of copying, and it can be wrong in both directions. Turnitin's AI writing detection FAQ says the company aims to keep its document false positive rate under 1% for documents with over 20% AI writing, and that it accepts missing some AI text to hold that rate. A document flagged at 50% "could contain as much as 65% AI writing," in Turnitin's own example. Neither figure tells a teacher whether a specific student is honest.

Turnitin aims for a document false positive rate under 1% for documents with over 20% AI writing, according to its AI writing FAQ checked 17 October 2026.

What the Turnitin AI detection report shows

The guide to accessing the AI Writing Report describes four states for the indicator in the classic Similarity Report:

• Blue with a percentage: the submission processed, and the number is the share of qualifying text the model labels as likely AI.
• Asterisk (*%): detection fell between 1% and 19%. No number and no highlights are shown, "to reduce misinterpretation."
• Gray with no number: the file could not be processed, often because it predates detection or misses the file rules.
• Error icon: processing failed.

In the newer report view, the same result sits under an AI Writing tab. The percentage is separate from the Similarity score: a paper can have high text matching and no AI flag, or the reverse.

"Qualifying text" matters. Turnitin's FAQ says the model drops lists, headers, references, data tables, formulas and code, and scores only prose sentences in long-form writing. A lab report that is half tables is scored on the remaining paragraphs, so the percentage can look large while the highlighted area looks small.

How do AI detectors work at Turnitin?

Most AI detectors estimate how predictable a passage is to a language model. Turnitin says its current system is a transformer-based deep learning model trained on its own corpus of academic writing. Each sentence gets a score between 0 and 1 in the context of the sentences around it, sentences above a threshold are labeled AI, and the document score is the percentage of sentences above that threshold. Turnitin says the model does not rely on hand-written rules such as "perplexity" or "burstiness," and that individual predictions "may not always be explainable in simple feature-by-feature terms."

That last point is the practical one. A student cannot be told which feature of a sentence triggered the label, so the fair response to a flag is to look at the writing process, not to argue with the model.

The model also changes. Turnitin's AI writing model release notes list English model updates on 12 February 2026 and 14 October 2026, and say updates do not change reports generated earlier. A submission scored before 14 October 2026 keeps its old score unless it is resubmitted.

Documented thresholds and limits

Turnitin shows no number and no highlights for AI scores from 1% to 19%, marking them with an asterisk instead, according to its report guide checked 17 October 2026.

Short documents carry extra risk. The FAQ says that in documents of only a few hundred words the prediction is "mostly all or nothing," so a mix of original and AI text can be flagged as entirely AI.

Published false positive statements and independent studies

Turnitin's own numbers. In a May 2023 post, Turnitin's chief product officer reported a sentence-level false positive rate of about 4% and said 54% of false positive sentences sat right next to actual AI writing. The same post explained why scores under 20% were flagged as less reliable and why the minimum length rose from 150 to 300 words. These are vendor figures on a vendor test set of 800,000 pre-ChatGPT papers.

Weber-Wulff and colleagues, 2023. The study Testing of detection tools for AI-generated text in the International Journal for Educational Integrity ran 54 test documents through 14 tools, including Turnitin. Turnitin scored highest of the 14 on every accuracy method, yet the authors concluded that the tools are "neither accurate nor reliable" and that even the highest accuracy values under their binary scoring method were below 80%.

Weber-Wulff and colleagues ranked Turnitin first of 14 AI detectors in a 2023 International Journal for Educational Integrity study, but no tool reached 80% accuracy under their binary method.

Liang and colleagues, 2023. The Stanford study GPT detectors are biased against non-native English writers found that seven detectors misclassified human-written TOEFL essays as AI at an average rate of 61.3%. Turnitin was not one of the seven tools, but the mechanism the authors describe, plainer vocabulary reading as machine-like, is why non-native writers deserve extra care with any detector.

Institutional decisions. Vanderbilt University disabled Turnitin's AI detector in August 2023, noting it had submitted 75,000 papers in 2022, so a 1% rate could have mislabeled about 750. Curtin University announced that Turnitin's AI writing detection would be disabled from 1 January 2026 across all campuses, while text-matching checks stay on.

For the wider picture across vendors, see our guide to AI detector false positives.

What a student can do if flagged

Process rules belong to each institution, so read your own academic integrity policy first. The University of Melbourne's published guidance is a clear model of how a fair process runs.

1. Ask for the report. Turnitin hides the AI report from students, but the University of Melbourne's advice for students notes that an instructor can share a PDF copy. Ask for it and for the date the report was generated.
2. Know what an informal meeting is. Melbourne says being asked to explain your work "is not an allegation of academic misconduct" and that "an AI writing detection report alone is not sufficient evidence for an allegation."
3. Bring process evidence. Drafts, version history, outlines, notes and sources show how the work was built. Be ready to explain your argument and why you chose your sources.
4. Ask questions in writing first. The University of Melbourne Student Union's academic misconduct guide advises asking for the purpose, scope and relevant policy section of any informal meeting before attending.
5. Respond to a formal allegation on time. At Melbourne, students must say how they will respond within 5 business days of an allegation notice, submit a written response 2 days before the hearing, and can bring a support person with 24 hours' notice.
6. Appeal if needed. Melbourne students receive an outcome within 5 business days of a hearing and then have 20 business days to appeal. Most universities publish similar deadlines; a student union advocate can help with yours.

No AI detector score should be the only evidence in an academic misconduct case.

What a teacher should do with a high score

Turnitin's guide on what to do if the AI writing percentage is high starts with institutional policy and frames the score as a prompt for a conversation. In practice:

• Check the length, genre and share of qualifying text before reading the number.
• Note the report date against the model release notes.
• Compare with the student's earlier supervised writing.
• Ask about the process before making any finding, and record the reasoning.

Our AI tools page for educators covers classroom policy options beyond detection.

Data and privacy

Turnitin's FAQ says AI writing detection runs only where the institution has enabled it, and admins can switch it off. There is no separate AI repository: results are stored with the Similarity Report, and whether papers enter the Turnitin repository remains the institution's choice. Institutions can request full deletion of their submissions, but not deletion of only the AI component.

Alternatives worth comparing

Institutions reviewing their options usually compare Turnitin with standalone detectors. Our AI detector rankings set out the field, our GPTZero review covers the detector most often used by individual teachers, and the Copyleaks review covers a combined plagiarism and AI checker. The same caution applies to all of them: a score starts a review, it does not finish one.

How we checked this

On 17 October 2026 we read Turnitin's AI writing FAQ, report access guide, model release notes and guidance for high scores; Turnitin's May 2023 product update; the 2023 studies by Weber-Wulff and colleagues and by Liang and colleagues; Vanderbilt's and Curtin's announcements; and the University of Melbourne's student guidance and student union process guide. Thresholds and limits are quoted as Turnitin listed them on that date. Our general method is on the HumanRank ranking methodology page.

Frequently asked questions

Can students see their Turnitin AI score? No, the AI writing indicator and report are visible only to instructors and administrators. An instructor can download the report as a PDF and share it.

What does an asterisk mean in Turnitin's AI score? It means the model detected AI writing between 1% and 19%. Turnitin shows no number or highlights in that range because false positives are more likely there.

How many words does Turnitin need to check for AI writing? At least 300 words of prose in long-form writing, and no more than 30,000 words, for English submissions.

Does Turnitin flag Grammarly spelling and grammar fixes as AI? Usually not, according to Turnitin's FAQ. It says grammar and spelling corrections were mostly not flagged in its tests, but text produced by generative features such as draft generation will likely be flagged.

Is the Turnitin AI score the same as the similarity score? No, the two are independent. The similarity score measures matching text, while the AI percentage estimates how much qualifying prose was likely generated by AI.

Will my Turnitin AI score change if Turnitin updates its model? Only if the paper is resubmitted. Turnitin says model updates, including the one on 14 October 2026, do not retroactively change existing reports.

Which languages does Turnitin AI detection support? Long-form English, Spanish, Japanese, Arabic, Dutch, German and Turkish. Papers in other languages are not processed and show an empty or error state.

Can a university turn off Turnitin AI detection? Yes, administrators can disable it in account settings. Vanderbilt did so in 2023, and Curtin University disabled it from 1 January 2026.