Can Turnitin Detect ChatGPT? What 280 Million Papers Show

The answer is yes, and it has been yes since April 2023. That is the easy part. The hard part is that “can it” and “will it” are separated by a gap wide enough to have ended pilots at several large universities and produced a peer-reviewed 2026 result in which Turnitin missed every single fully AI-written paper it was shown. Here are the numbers on both sides of that gap.

1 Yes — and the famous 98% is not an accuracy rate

Turnitin built its detector specifically to catch ChatGPT. The 98% figure everyone quotes is a confidence threshold for flagging, not a measure of how often the tool is right.

Read the second-to-last row twice. Turnitin’s own whitepaper says it “does not use accuracy as a metric as it is too easily manipulated,” and publishes recall and false positive rates instead. So when a blog tells you Turnitin is “98% accurate”, that number has been lifted from a sentence about flagging confidence and quietly relabelled. A separate 98% figure circulates too — the share of Turnitin institutions that had the feature switched on in mid-2023 — which is an adoption statistic, not a performance one.

2 What actually gets scanned

Under 300 words, no score. Wrong file type, no score. Wrong language, no score. And you will never see the result yourself.

Two of those rows cause most of the confusion students bring to us. First, the AI score is not the similarity score — they are separate systems, and the AI check does not match your text against a stored library of chatbot answers. It is a transformer classifier making a per-sentence prediction. Second, you cannot see your own AI percentage; only your instructor can, which is why pre-submission checking is guesswork by design unless your institution has enabled a self-check assignment.

That 300-word floor is also why short assignments behave unpredictably: a 280-word discussion post returns nothing at all, while a 310-word one returns a full score computed from a very small sample of sentences. We dug into that asymmetry in why short essays get flagged more often.

3 Independent studies: best in class, and still not good enough

Turnitin has finished first in multiple peer-reviewed comparisons. The same papers conclude that no detector should be used as evidence in a misconduct case.

The 2026 result is worth stating carefully, because it is the strongest single data point on this page. Van Vlasselaer, Van Droogenbroeck and Spruyt generated papers with GPT-4o and GPT-4o Deep Research in May 2025 and ran them through four detectors. Turnitin scored every fully AI-generated paper between 0 and 20% AI — a 100% false negative rate on that set — while Pangram hit 97.5% inclusive accuracy on the same documents. Turnitin also produced zero false positives on genuinely human papers, so the model has not become reckless. It has become blind to current-generation output.

Set that against Weber-Wulff’s 2023 finding that Turnitin was the only tool to correctly classify every unmodified AI document, and the trend line is not subtle. Detection worked reasonably well against GPT-3.5-era text and has been losing ground with each model generation since, which matches what we found tracking how far detectors now diverge from each other.

4 What actually breaks detection

Perkins et al. measured six evasion techniques individually. Every one of them cut mean detector accuracy below 40%; the worst dropped it to 12.9%.

First bar: Turnitin’s accuracy on unmodified AI text. Remaining bars: mean accuracy across all seven detectors tested after each evasion technique. Source: Perkins et al. (2024).

The uncomfortable row in that chart is “ESL-style writing” at 27.7%. Perkins et al. included it as an evasion technique, which means prompting an LLM to write like a non-native speaker defeats detectors more reliably than paraphrasing does. The same property that makes it an evasion route is what makes genuine non-native writing risky to submit — Liang et al. at Stanford found seven detectors flagged 97.80% of TOEFL essays as AI-written by at least one tool, against near-perfect classification of native-speaker essays. Turnitin was not among the seven tested, but the pattern is why we wrote up safer revision habits for ESL writers.

Turnitin’s response has been to ship counter-models: AI paraphrasing detection in July 2024, and AI bypasser detection in August 2025, which it says was trained on “the signals and patterns of leading humanizers.” Both are paid add-ons requiring Turnitin Originality, both are English-only, and neither has a published accuracy figure. The one independent measurement we have — the June 2026 study — put Turnitin at 50.0% on humanized text. We tracked the rollout in Turnitin now detects AI humanizers and the model update in what the 2026 model claims to catch.

None of this is a recommendation. It is the published state of the evidence, and the practical takeaway runs the other way: the techniques that suppress a score are the same ones that damage a piece of writing, and a 50-50 detector is still a coin flip you have to live with. The durable protection is a documented drafting process, which is exactly what a defensible writing workflow is for.

5 Detection risk estimator

Pick what happened to the text and see the measured detection accuracy for that scenario, with the study it comes from.

6 False positives, and the student who won

Turnitin publishes a sub-1% document error rate. Multiplied by institutional volume, that is hundreds of students a year — and in February 2026 one of them won in court.

The Adelphi case is the one to know. A student’s World Civilizations essay was marked fully AI-written by Turnitin while two other detectors read it as human. A judge ruled the plagiarism finding without merit and the charge was removed from his record — after his family spent six figures on legal fees. That last detail is the real lesson: being right is not the same as being cheap to prove, which is why assembling an authorship packet before you submit costs an hour and an appeal costs a semester.

Turnitin’s own data on where errors cluster helps here too: 54% of false-positive sentences sit directly beside genuine AI writing. If you used a chatbot for one section and wrote the rest yourself, the boundary sentences are the documented risk. And if your instructor is treating the score as proof, it is worth knowing what most institutional policies actually say about that — Perkins et al. concluded flatly that these tools “cannot currently be recommended for determining whether violations of academic integrity have occurred.”

7 280 million papers: what the scale data shows

Turnitin publishes running totals. They show AI-written submissions rising roughly 4.5x in under three years while detection accuracy in the literature moved the other way.

Share of submissions scoring 80% or more AI-generated writing, Turnitin’s own published data. Bars scaled to the highest value. Sources: Turnitin press releases, 2023-2026.

So: can Turnitin detect ChatGPT? It can, it does it hundreds of millions of times a year, and it is measurably worse at it than the marketing suggests — best-in-class against raw output, close to useless against current-generation text that has been touched by a human or a laundering tool, and conservative enough that a clean score proves nothing either way. If you are on the receiving end of a flag you believe is wrong, start with our false positive defence checklist and our guide to what the scores really mean.

Methodology

Compiled 27 July 2026. Every figure is sourced either to Turnitin’s own published material — guides, whitepaper, blog posts and press releases — or to peer-reviewed research in the International Journal for Educational Integrity, the International Journal of Educational Technology in Higher Education, Patterns and the ACL Anthology. Twenty-four sources are cited below. Vendor-published competitor benchmarks were excluded from the data tables because they are not independently replicated.

Freshness distribution: nine cited sources are from 2025 or 2026, seven from 2024, and eight from 2023. The 2023 items are Turnitin’s launch-era technical disclosures and the Weber-Wulff and Liang studies, which remain the most-cited work of their kind. Limitations: Turnitin has published no cumulative paper total since October 2024, no accuracy figures for its paraphrasing or bypasser detection add-ons, and no independent study has yet tested Turnitin specifically on non-native English writing at scale — Liang et al. did not include it. Reviewed quarterly.

Frequently asked questions

Sources

1. Turnitin Press. “Turnitin turns on AI writing detection capabilities for educators and institutions (4 April 2023).” https://www.turnitin.com/press/turnitin-turns-on-ai-writing-detection-capabilities-for-educators-and-institutions Accessed 27 July 2026.
2. K-12 Dive. “Turnitin unveils AI writing detection tool (98% confidence claim).” https://www.k12dive.com/news/turnitin-launches-AI-writing-detection/646632/ Accessed 27 July 2026.
3. Turnitin Guides. “Turnitin’s AI writing detection capabilities FAQs.” https://guides.turnitin.com/hc/en-us/articles/28477544839821-Turnitin-s-AI-writing-detection-capabilities-FAQs Accessed 27 July 2026.
4. Turnitin. “AI writing detection model architecture and testing protocol (whitepaper).” https://iknow.library.uitm.edu.my/249/2/AI%20Writing%20Detection%20Model.pdf Accessed 27 July 2026.
5. Turnitin Blog. “AI writing detection update from Turnitin’s Chief Product Officer (23 May 2023).” https://www.turnitin.com/blog/ai-writing-detection-update-from-turnitins-chief-product-officer Accessed 27 July 2026.
6. Turnitin Blog. “Understanding the false positive rate for sentences (14 June 2023).” https://www.turnitin.com/blog/understanding-the-false-positive-rate-for-sentences-of-our-ai-writing-detection-capability Accessed 27 July 2026.
7. Turnitin Press. “AI detection feature reviews more than 65 million papers (25 July 2023).” https://www.turnitin.com/press/turnitin-ai-detection-feature-reviews-more-than-65-million-papers Accessed 27 July 2026.
8. Turnitin Press. “Turnitin marks one year of its AI writing detector (9 April 2024).” https://www.turnitin.com/press/turnitin-first-anniversary-ai-writing-detector Accessed 27 July 2026.
9. Turnitin Blog. “Does Turnitin detect AI writing? Debunking common myths (31 October 2024).” https://www.turnitin.com/blog/does-turnitin-detect-ai-writing-debunking-common-myths-and-misconceptions Accessed 27 July 2026.
10. Turnitin Press. “New AI paraphrasing detection feature (16 July 2024).” https://www.turnitin.com/press/turnitin-new-ai-paraphrasing-detection-feature Accessed 27 July 2026.
11. Turnitin Press. “Turnitin expands capabilities amid rising threats posed by AI bypassers (27 August 2025).” https://www.turnitin.com/press/turnitin-expands-capabilities-amid-rising-threats-posed-by-ai-bypassers Accessed 27 July 2026.
12. Turnitin Press. “Turnitin Delivers Turnitin Clarity (15 July 2025).” https://www.turnitin.com/press/turnitin-delivers-turnitin-clarity Accessed 27 July 2026.
13. Turnitin Press. “Turnitin Data Shows Transparency About AI Use Benefits Students And Educators (24 February 2026).” https://www.turnitin.com/press/turnitin-data-shows-transparency-about-ai-use-benefits-students-and-educators Accessed 27 July 2026.
14. Weber-Wulff et al. (2023). “Testing of detection tools for AI-generated text, International Journal for Educational Integrity.” https://link.springer.com/article/10.1007/s40979-023-00146-z Accessed 27 July 2026.
15. Perkins et al. (2024). “GenAI Detection Tools, Adversarial Techniques and Implications for Inclusivity in Higher Education.” https://arxiv.org/abs/2403.19148 Accessed 27 July 2026.
16. Perkins et al. (2024). “Open-access full text with per-technique accuracy table, James Cook University.” https://researchonline.jcu.edu.au/86904/1/86904.pdf Accessed 27 July 2026.
17. Van Vlasselaer, Van Droogenbroeck & Spruyt (2026). “Who wrote this? Evaluating the reliability of AI detection tools in higher education, International Journal for Educational Integrity.” https://link.springer.com/article/10.1007/s40979-026-00226-w Accessed 27 July 2026.
18. Liang et al. (2023). “GPT detectors are biased against non-native English writers, Patterns.” https://arxiv.org/html/2304.02819v3 Accessed 27 July 2026.
19. Dugan et al. (2024). “RAID: A Shared Benchmark for Robust Evaluation of Machine-Generated Text Detectors, ACL.” https://aclanthology.org/2024.acl-long.674/ Accessed 27 July 2026.
20. Vanderbilt University. “Guidance on AI Detection and Why We’re Disabling Turnitin’s AI Detector (16 August 2023).” https://www.vanderbilt.edu/brightspace/2023/08/16/guidance-on-ai-detection-and-why-were-disabling-turnitins-ai-detector/ Accessed 27 July 2026.
21. Western Gazette. “Western stops using Turnitin AI written detection software (January 2024).” https://westerngazette.ca/news/administration/western-stops-using-turnitin-ai-written-detection-software/article_f34a50d8-b647-11ee-bc98-678b913cbcc0.html Accessed 27 July 2026.
22. Tom’s Guide. “A major university just banned AI detectors (Indiana University Kelley School, 1 June 2026).” https://www.tomsguide.com/ai/a-major-university-just-banned-ai-detectors-heres-why Accessed 27 July 2026.
23. Inside Higher Ed. “Adelphi student wins AI plagiarism lawsuit (11 February 2026).” https://www.insidehighered.com/news/quick-takes/2026/02/11/adelphi-student-wins-ai-plagiarism-lawsuit Accessed 27 July 2026.
24. PLEASE. “Schools that Banned AI Detectors.” https://www.pleasedu.org/resources/schools-that-banned-ai-detectors Accessed 27 July 2026.

Last updated: 27 July 2026. Reviewed quarterly against Turnitin’s published documentation and new peer-reviewed studies.