Does Turnitin check for AI?

Does Turnitin have an AI detector?

Yes. Turnitin added an AI writing indicator to its Similarity Report in April 2023, and it now sits in an AI Writing tab in the newer report view. Turnitin's AI writing detection FAQ says the feature identifies content generated by tools such as ChatGPT and "instances where AI-generated text may have been modified by AI paraphraser or bypasser (also called humanizers) tools to evade detection."

It is not a separate product you can buy as a student. The FAQ says AI writing detection is "only available to customers that license Turnitin Originality." Schools on Turnitin Feedback Studio, Similarity or Originality Check need to upgrade through their account manager, and iThenticate 2.0 customers need the AI writing add-on. Turnitin does not sell individual licenses at all.

Whether your paper is checked for AI depends on your school's license and settings, not on Turnitin alone. Turnitin says it only processes submissions for AI writing "if the institution has the feature enabled," and institutions can opt out in their admin settings. Vanderbilt University did so in August 2023, and Curtin University switched it off from 1 January 2026 while keeping text matching.

If your school uses Turnitin through Canvas, Moodle, Blackboard or another learning platform, the AI check still happens inside Turnitin. The FAQ says "there is no AI writing indicator or score embedded directly in the LMS user interface"; instructors open the report to see it.

My take: if you want to know whether your work is checked, don't ask a forum. Look in your syllabus or ask your instructor or academic integrity office which Turnitin features are switched on. It is a reasonable question and asking it signals nothing.

How does Turnitin detect AI?

Turnitin uses a trained deep learning model, not a list of rules. The steps it describes in its FAQ:

1. It strips out non-prose. Lists, headers, references, data tables, formulas and code are removed. Only sentences in paragraphs count.
2. It scores every sentence. Each sentence gets a score between 0 and 1, judged together with the sentences around it.
3. It applies a threshold. Sentences above the threshold are labeled AI; the rest are labeled human.
4. It reports a percentage. The document score is the share of evaluated sentences labeled AI.

Turnitin says the model is transformer-based and trained on its own corpus of academic writing, including second-language learners and less common subjects. It says the model "is not explicitly programmed to evaluate specific signals such as burstiness, perplexity," and that a single prediction may not be explainable feature by feature. In October 2026 Turnitin updated the English model architecture to improve accuracy where text switches between human and likely AI writing.

The model changes over time. Turnitin's AI writing model release notes list updates on 12 February 2026 and a Spanish update on 5 May 2026, and every note says updates do not change reports already generated. A paper only gets rescored if it is resubmitted.

Example. Here is why sentence-level scoring matters, in a short passage I wrote for illustration. Imagine an essay paragraph where the first two sentences describe a specific lab session ("Our group's titration kept overshooting because the burette tap stuck") and the third is a generic summary ("Overall, this experiment highlights the importance of precision in scientific research"). A sentence-level model can score the first two as human and the third as likely AI, because the third is the most predictable kind of closing sentence. The document percentage is then the share of sentences above the threshold, not a verdict on the whole essay. That is also why Turnitin's October 2026 update targets the transitions between human and likely AI text.

For the wider picture of perplexity, classifiers and watermarks, see my guide to how AI detectors work.

What Turnitin can detect: ChatGPT, Claude, Grammarly, QuillBot

Turnitin publishes a list of the language models its English detector covers. Here is what it says about the tools students ask about most.

Can Turnitin detect ChatGPT? Turnitin lists every major GPT release up to GPT-5.4, which came out in March 2026. The list on 29 October 2026 names no model released after March 2026. In my view, that is not a gap you should plan around: Turnitin updates its model several times a year, and a resubmitted paper is scored on the current one.

Is Claude detectable by Turnitin? Claude models up to Opus 4.6 and Sonnet 4.6 are on the list. The same point about newer releases applies.

Does Turnitin detect Grammarly? Not for spelling, grammar and punctuation, in most cases. Turnitin's FAQ says changes from Grammarly free and premium "were not flagged as AI-written" in most of its tests, but text from Grammarly's generative features will likely be flagged.

Does Turnitin detect QuillBot? Turnitin's guide to using the AI Writing Report names QuillBot as an example of an AI word spinner whose changes to AI-generated text the model can detect. That detection is English only.

Can Turnitin detect AI on a PDF? Yes. PDF is one of the four accepted file types, alongside .docx, .txt and .rtf.

Turnitin's coverage claims are the vendor's own. Independent testing has found weaker results, covered below.

My take: the practical line is between editing help and writing help. Spelling and grammar fixes are very unlikely to matter. Anything that generates or rewrites whole sentences for you is exactly what the model is built to find, and more to the point, it is what your course's AI policy is written about.

Turnitin AI score and report

The AI score appears as an indicator on the Similarity Report and opens a separate AI Writing Report. Turnitin's guide to accessing the AI Writing Report describes four states:

• Blue with a percentage (20% to 100%): processed. The number is the share of qualifying prose the model labels likely AI or AI-modified.
• Asterisk (*%): the model found AI writing between 1% and 19%. No number and no highlights are shown, because "results are less reliable and false positives are more likely."
• Gray, no number: not processed. The paper may predate the feature or miss the file rules.
• Error icon: processing failed.

Inside the report, a submission breakdown shows where the flagged sentences sit, and highlights them in the text. AI-generated text that was further modified by a paraphraser or bypasser is folded into the same "AI-generated" category.

Three things often confuse students:

• The AI score and the similarity score are separate. Turnitin says they are "completely independent and do not influence each other." My Turnitin similarity score guide explains the matching score.
• The percentage is not of the whole paper. It is a share of qualifying prose only, so a paper full of tables can show a high number with few highlights.
• Students cannot see it. The AI indicator and report are visible only to instructors and administrators. An instructor can download a PDF and share it. There is no way for a student to see their own AI score in Turnitin before submitting.

Here is how I would read each state if an instructor shares the report with you:

What percentage of AI is acceptable on Turnitin?

Turnitin sets no acceptable percentage, and neither do most schools. The rule that matters is your course's AI policy, not a number.

• Turnitin does not decide misconduct. Its FAQ 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."
• Scores under 20% are hidden. Turnitin treats 1% to 19% as too unreliable to show, so an asterisk is not a finding of anything.
• Some AI use may be allowed. Many courses permit AI for brainstorming or editing with disclosure. If AI use was allowed and disclosed, a score reflects that use rather than a breach.
• Universities say the report alone is not enough. The University of Melbourne tells students that 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."

So there is no safe percentage to aim for, and trying to engineer one misses the point. Follow your course's AI rules, disclose permitted use, and keep drafts and version history so you can show how the work was made. My guide to AI detector false positives explains what to do if a score is wrong.

My take: the students I'd worry about are not the ones with a 30% score and a folder of drafts. They are the ones with a 0% score, no drafts and a policy they never read.

How accurate is Turnitin's AI check?

Turnitin's target is a document false positive rate under 1% for documents with over 20% AI writing, tested on over 700,000 academic papers written before ChatGPT. It accepts misses to hold that rate: a document scored at 50% "could contain as much as 65% AI writing."

Independent studies found lower accuracy than vendor figures suggest, on older versions of the model:

• Weber-Wulff and colleagues, 2023. Turnitin scored highest of 14 tools in the International Journal for Educational Integrity study, but no tool reached 80% accuracy, and the authors concluded the tools are "neither accurate nor reliable."
• Perkins and colleagues, 2024. In tests run in September and October 2023, Turnitin detected 61% of unedited AI samples, second of seven detectors, but showed the largest drop, 42.1%, once the AI text was manipulated, according to the International Journal of Educational Technology in Higher Education paper.

Both studies predate Turnitin's December 2023 paraphrase detection and August 2025 bypasser detection, so current performance may differ in either direction. My guide on whether Turnitin can detect humanized AI text covers the bypasser update and what independent studies found, and my guide on whether teachers can detect AI covers the checks instructors use beyond the score. My Turnitin AI detection review covers accuracy, appeals and the 300-word rule in more depth.

My take: Turnitin is tuned to avoid accusing people, and it pays for that by missing AI text. That is the right trade for a tool used in misconduct cases, but it means a low score proves nothing and a high score still needs a human to look.

Are Turnitin and JustDone the same?

No. JustDone is a separate company and product. Its own site describes it as "an all-in-one workspace designed for students and learners" that includes an AI detector, AI humanizer, plagiarism checker, summarizer, paraphraser and grammar checker. Nothing on Turnitin's or JustDone's pages, as read on 29 October 2026, says the two are related.

The practical differences:

• Who uses it. Turnitin sells only to institutions, and its AI score is visible to instructors. JustDone sells to individuals.
• What it does. Turnitin checks submitted work. JustDone also rewrites text, including a humanizer.
• Whether results match. A score from any third-party checker, JustDone included, is a different model with different thresholds. It tells you nothing reliable about what Turnitin will report.

Does Turnitin detect JustDone AI? Turnitin does not name the humanizer or paraphrasing tools it detects. It says only that its model is trained on "leading" bypassers in English. The University of Melbourne warns that many free checker sites flag human work to sell humanizer tools, and that those tools "usually use AI and are often picked up by the University's detector tool."

Schools set their own AI rules. Using a humanizer to hide AI use that a course does not allow is an integrity issue regardless of any score. My guide on whether using an AI humanizer is cheating quotes real course policies, and where AI help is allowed, such as emails, marketing copy or smoothing a second language, my ranking of AI humanizers compares the tools on price, languages and privacy terms.

How we checked this

• Read on 29 October 2026: Turnitin's AI writing detection FAQ (English), Using the AI Writing Report, How to access the AI Writing Report, the AI writing model release notes and Turnitin's 27 August 2025 bypasser press release.
• Also read: JustDone's home page and FAQ; Weber-Wulff et al. (2023) and Perkins et al. (2024); Vanderbilt University's August 2023 notice; Curtin University's September 2025 notice; the University of Melbourne's student advice on Turnitin and AI writing detection.
• Method: I quoted model lists, thresholds and file rules as Turnitin listed them on that date, and kept my own judgement in the "My take" lines. The site's general approach is on the HumanRank ranking methodology page. More detection guides are in the HumanRank guides hub.

Frequently asked questions

Does Turnitin automatically check for AI? Yes, if your institution licenses Turnitin Originality and has AI writing detection enabled. Every submission is then checked for AI writing, AI paraphrasing and AI bypassers without any extra step.

What AI detector does Turnitin use? Its own. Turnitin built a transformer-based deep learning model trained on its corpus of academic writing. It does not use GPTZero, Copyleaks or any outside detector.

Can I see my Turnitin AI score as a student? No. Only instructors and administrators can see the AI indicator and report. Your instructor can download the report as a PDF and share it with you.

Can Turnitin detect AI in other languages? Turnitin processes long-form English, Spanish, Japanese, Arabic, Dutch, German and Turkish. Paraphrase and bypasser detection works on English only.

Does Turnitin check old papers for AI? Only if they are resubmitted. Turnitin cannot score submissions retroactively, and model updates do not change existing reports.

Can Turnitin detect AI 100% of the time? No. Turnitin says it accepts missing some AI text to keep false positives under 1%, and a 50% score could hide 65% AI writing.

Does Turnitin flag Grammarly as AI? Spelling and grammar fixes are mostly not flagged, according to Turnitin's tests. Text produced by Grammarly's generative features, such as drafting or paraphrasing, will likely be flagged.

Why does my school not use Turnitin's AI detection? Institutions can switch it off. Vanderbilt disabled it in 2023 over false positive concerns, and Curtin University disabled it from 1 January 2026.

REVIEW NOTEI checked this guide on 29 October 2026 against Turnitin's own instructor guides, FAQ, release notes and press releases, two independent studies and published university guidance. Every number on this page belongs to the named source, and the opinions are mine.