Does Turnitin detect AI? What Turnitin says, and what independent tests found

Does Turnitin detect AI? Yes, but only for schools that license the feature and only in submissions with at least 300 words of prose. It also leans, deliberately, towards missing some AI text rather than wrongly flagging students. Turnitin itself says its AI score should not be the sole basis for action, and independent tests show it can miss a large share of AI writing from newer models.

How does Turnitin's AI writing detection work?

When a paper is submitted, Turnitin's model gives each sentence a score between 0 and 1, judging it in the context of the sentences around it. Sentences above a threshold are labelled AI. The document's AI percentage is the share of evaluated sentences above that threshold.

Only "qualifying" text counts. The model drops anything it predicts isn't a real sentence, such as lists, headers, references, data tables, formulas and code, so the percentage is of the prose, not of the whole file. Turnitin describes the model as a transformer-based deep learning system trained on its own large collection of academic writing. It says the model is not programmed to measure "perplexity" or "burstiness" directly, and that its individual predictions can't always be explained in simple terms.

The model changes over time. Turnitin says it updated the English model's architecture in October 2026 to handle the points where a text switches between human and AI writing more accurately. It also publishes a list of the models it can detect, which at the time of writing runs up to GPT-5.4, Claude Opus 4.6 and Gemini 3.1 Pro preview.

What are Turnitin's limits on length, language and file type?

A submission gets an AI score only if it meets these requirements, according to Turnitin's guides:

• At least 300 words of prose in long-form writing, and no more than 30,000 words.
• A file under 100 MB, in .docx, .pdf, .txt or .rtf format.
• A supported language. Turnitin's FAQ lists long-form English, Spanish, Japanese, Modern Arabic, Dutch, German and Turkish. Detection of AI paraphrasing and "bypasser" tools works for English only.

Submissions that don't meet these rules show a grey indicator with no score. Short documents have another quirk: Turnitin says that with only a few hundred words, the prediction is "mostly all or nothing", so a text mixing human and AI writing could be flagged as entirely AI.

What does the asterisk (*%) on a Turnitin AI score mean?

If Turnitin finds AI writing in 1% to 19% of the qualifying text, it shows an asterisk instead of a number and highlights nothing. Its reason is that false positives are more likely in that range. So "*%" doesn't mean "no AI found". It means "some, but below the level Turnitin is prepared to report".

How accurate is Turnitin's AI detector, according to Turnitin?

Turnitin says it aims to keep the false positive rate, meaning fully human documents flagged as AI, under 1% for documents with over 20% AI writing. Before each model release it tests on more than 700,000 academic papers written before ChatGPT existed.

It is open about the trade-off. To hold the false positive rate down, it accepts missing some AI text, and gives this example: a document scored at 50% AI could contain as much as 65%. It also lists the kinds of human writing most likely to cause false positives: text without much structural variation, text that repeats itself, and paraphrasing that doesn't develop new ideas.

On how the score should be used, Turnitin is direct. It says it does not make a determination of misconduct, and that the percentage "should not be used as the sole basis for action or a definitive grading measure by instructors."

Does Turnitin detect AI reliably? What independent tests found

The most recent independent test we could verify was published in June 2026 by researchers at Vrije Universiteit Brussel. They ran 160 academic papers of at least 4,000 words through Turnitin, GPTZero, Copyleaks and Pangram:

• Fully AI-written papers (40, from ChatGPT's Deep Research mode): Turnitin scored every one under 20%. In the study's terms, it missed all of them.
• Hybrid papers (40, human writing with GPT-4o sections): Turnitin's score fell in the right range for 60%.
• Humanized papers (40, the same AI sections rewritten by a GPT-4o prompt asking for more natural, student-like writing): this dropped to 50%.
• Human papers (40, by non-native English graduate students, written before ChatGPT): Turnitin flagged none.

Pangram did much better on the AI-written papers in the same study (see how accurate is Pangram). The authors note the limits of their design: one generator for the fully AI set, one humanizing prompt and 40 papers per group.

Earlier evidence points the same way. In 2023, Debora Weber-Wulff and colleagues tested 14 detection tools, Turnitin among them, and concluded that they were neither accurate nor reliable and leaned towards calling AI text human. That August, Vanderbilt University disabled Turnitin's AI detector. It pointed out that at a 1% false positive rate, the 75,000 papers it had submitted in 2022 could have included around 750 wrongly labelled ones.

Can Turnitin detect humanized or paraphrased AI text?

Turnitin says yes, for English. It says its detector is trained and tested on leading AI paraphrasing and "bypasser" tools, which it also calls humanizers, and that this check runs automatically for institutions with AI detection turned on. It won't name the tools it detects, to avoid helping people evade them.

The Brussels study suggests the protection is partial: Turnitin's accuracy fell from 60% on hybrid papers to 50% once the AI passages had been humanized. Our article on whether AI humanizers work compares results across detectors.

Does Grammarly trigger Turnitin's AI detector?

Turnitin says that in its tests, spelling, grammar and punctuation changes from Grammarly, free or premium, were mostly not flagged. Content from Grammarly's generative features, such as drafting, paraphrasing and summarising, is likely to be flagged as AI-generated.

Can students use Turnitin's AI checker themselves?

No. Turnitin's AI indicator and report are not visible to students. Instructors can download the AI report as a PDF and share it with you. If you want to check your own draft, you need a different tool.

Ours is free, needs no account, has no word limit and runs in your browser, so your paper is never uploaded. It is also a different detector, so its score won't match Turnitin's, and like any detector it can be wrong in both directions: treat a result as a reason to look closer, not as proof. Our methodology page explains how it is tested. Our guide to Turnitin alternatives for students explains what each option can and can't tell you.

Sources

1. Turnitin Guides: Turnitin's AI writing detection capabilities FAQs, checked October 2026
2. Turnitin Guides: File requirements for an AI Writing Report, checked October 2026
3. Turnitin Guides: How to access the AI Writing Report, checked October 2026
4. Turnitin Guides: What should I do if the AI writing percentage is high?, checked October 2026
5. Van Vlasselaer, Van Droogenbroeck and Spruyt, Who wrote this? Evaluating the reliability of AI detection tools in higher education (International Journal for Educational Integrity, 2026), checked October 2026
6. Weber-Wulff et al., Testing of detection tools for AI-generated text (International Journal for Educational Integrity, 2023), checked October 2026
7. Vanderbilt University: Guidance on AI Detection and Why We're Disabling Turnitin's AI Detector (16 August 2023), checked October 2026