Turnitin AI Detection: How It Works and What It Flags
Who can see the score, what the percentage means, and how well it handles ChatGPT, humanizers and human writing.
By the AI Checker Humanizer team. Published . Updated .
Does Turnitin detect AI? The short answer
Yes, if your school pays for it and switches it on. Turnitin's AI writing detection comes with its Originality product, and administrators can turn it off. If your school hasn't enabled it, Turnitin won't generate an AI writing report for your submission.
When it is on, you probably won't see the result. Turnitin says the AI score and report aren't shown to students; your instructor sees them and can choose to share a PDF copy.
If you don't know whether your course uses it, look at the syllabus or the assignment's submission page, or ask your instructor; many schools also say on their teaching center pages whether the feature is on.
We checked Turnitin's product details against its own guides and release notes on November 11, 2026; the article also draws on an independent evaluation and on university policies. Turnitin changes this feature often, so details may have moved on.
How Turnitin's AI detection works
Turnitin uses a deep learning model, of the same broad type as the language models it is trying to detect. The model gives each sentence a score between 0 and 1, reading it in the context of the text around it. Sentences above a set threshold are labeled AI. Turnitin says the model isn't programmed to look for specific signals like “perplexity” or “burstiness”; it learned its own patterns from training data.
A few practical limits apply:
- The file needs at least 300 words of prose, and at most 30,000 words.
- Accepted formats are .docx, .pdf, .txt and .rtf.
- Lists, headings, code, the reference list and non-prose table content are left out before scoring. Prose written inside a table is still scored.
- Supported languages, as of November 2026: English, Spanish, Japanese, Arabic, German, Dutch and Turkish. Detection of paraphrased or humanized text works in English only.
The 300-word minimum matters more than it sounds. A short reflection or discussion post isn't scored at all, and a paper that barely clears the minimum gives the model very little to work with. Turnitin says that on short submissions it scores the text as a single segment without overlap, so the result tends to be all or nothing: either most of it is labeled AI or none of it is.
The exclusions matter too. Because the reference list, headings and lists are removed, the percentage is calculated only on your prose. A paper that is mostly lists and data tables is judged on the sentences that are left.
| When | What changed |
|---|---|
| April 2023 | AI writing detection launches for institutions |
| December 2023 | Starts detecting text run through AI word spinners |
| July 2024 | Adds an “AI-paraphrased” category; scores from 1% to 19% shown only as an asterisk |
| August 2025 | Detects text changed by “AI bypasser” tools, also called humanizers (English only) |
| August 2026 | Merges the paraphrased category into one “AI-generated” category |
| August to October 2026 | Adds Arabic, then German, Dutch and Turkish |
What the AI score means
The percentage is the share of the qualifying text that Turnitin's model labeled as AI-written. It isn't the chance that the whole paper was written by AI, and it doesn't count the parts that were left out, such as the reference list.
- 0%: nothing was labeled AI.
- 1% to 19%: shown only as an asterisk (*%), with no highlights. Turnitin hides these because low scores are more likely to be wrong.
- 20% and up: the percentage is shown and the labeled text is highlighted.
Older reports had two highlight colors, one for AI-generated text and one for AI-generated text that had been paraphrased. Since August 2026 there's a single “AI-generated” category.
Your instructor sees the percentage and any highlighted sentences.
There's no official “acceptable” percentage. Turnitin says the score should not be used as the sole basis for adverse action against a student and that it doesn't decide whether misconduct happened. What counts as too high is up to your school and your instructor, and many schools don't set a number at all.
Can Turnitin detect ChatGPT, Claude and Gemini?
Turnitin lists the models it covers, including GPT, Claude and Gemini models such as GPT-5.4, Claude Opus 4.6 and Gemini 3.1 Pro. Our own detection rates for four Claude models we tested are on our Claude AI detector page. Unedited text from these models is the case detectors handle best, because nothing has disturbed the patterns they learned.
It gets harder when a person has rewritten the AI text, or when AI and human writing are mixed in one paper. An evaluation by Temple University's teaching center tested 120 papers. Turnitin judged 28 of 30 fully human papers correctly and gave a full AI score to 23 of 30 AI-written papers. Of 30 papers mixing human and AI writing, 13 got a partial score and six were scored as entirely AI, and the highlighted sentences often didn't match the parts that were actually AI-written. That evaluation predates several updates to Turnitin's model.
Mixed papers are exactly what Turnitin says its October 2026 update to the English model was aimed at: it describes the change as improving accuracy “at transition points”, where a passage moves from human writing to AI writing or back. Turnitin hasn't published new figures for mixed text, so it's too early to say how much that changed. If you wrote some parts yourself and used AI for others in a way your course allowed, the safest assumption is that the boundaries in a report won't line up exactly with what you did.
Turnitin and humanized or paraphrased text
Turnitin has added coverage of rewriting tools step by step. It started detecting AI “word spinners” in late 2023, added a separate category for AI-paraphrased text, naming QuillBot as an example, in 2024, and in August 2025 began detecting text changed by what it calls “AI bypasser” tools, also known as humanizers. The bypasser detection is for English only, and Turnitin doesn't name the tools it trained on.
Turnitin also sells a separate product, Turnitin Clarity. Where a school uses it alongside Originality, the writing report includes a note when a block of more than 300 words was pasted in, which is a different kind of signal from the AI score.
On Grammarly, Turnitin says ordinary spelling and grammar corrections were not flagged in most cases. Grammarly's generative features, which draft, rewrite or summarize text, will likely be flagged, because the result is AI-written text.
No detector catches everything, and no humanizer can promise a result either. If your course doesn't allow AI-written work, rewording AI text doesn't make it your own, and getting past a detector doesn't change what the rules say.
Where AI help is allowed, the safest approach is to say what you used and how. Many courses now ask students to note whether they used AI for brainstorming, outlining or language editing. Keep your drafts as well, so you can show which parts you wrote and revised yourself. Our academic integrity page sets out how we expect our own tools to be used.
How accurate is Turnitin's AI detection?
Turnitin publishes two figures, both measured by Turnitin itself:
- A document-level false positive rate under 1%: fewer than 1 in 100 fully human documents get labeled 20% AI or more. Turnitin says it tests each release on 700,000 papers written before ChatGPT existed.
- A sentence-level figure from 2023: of the sentences it highlighted as AI, about 4% were actually human-written. More than half of those sat right next to real AI writing.
The 4% figure is easy to misread. It describes the highlighted sentences, not a human essay: it doesn't tell you how many sentences in a fully human paper would be wrongly highlighted. At a hypothetical 1% false positive rate, 10,000 human-written papers would produce about 100 false flags at 20% AI or more. Turnitin reports under 1% on its own tests; the rate at a given school may differ.
For an independent check, see the Temple University evaluation discussed above. Our article on how accurate AI detectors are rounds up the other studies.
Some universities have decided the tool isn't worth the risk. Vanderbilt turned off Turnitin's AI detector in August 2023, noting that its own volume of papers would mean hundreds of wrong flags even at Turnitin's stated rate. UCLA opted out when the feature launched, and Yale, UC Berkeley and the University of Pittsburgh say on their teaching pages that it is switched off. The Australian National University disabled it from January 2024, the University of Waterloo from September 2025, Curtin University from January 2026, and Washington State University cancelled it in February 2026. Our table of 32 college policies has more.
Others keep it on with conditions. The University of Georgia, Texas State University, Lone Star College and the University of Bristol all use Turnitin's AI detection, and all tell staff in their own guidance that the score can't be the only evidence against a student. (University pages checked November 11, 2026.)
What to do if Turnitin flags your work
- Ask to see the AI report and which parts were highlighted.
- Read your school's academic integrity policy, especially anything on AI and on how a detector score can be used.
- Gather your drafts, notes, sources and the document's version history. Don't edit the submitted file.
- Offer to walk your instructor through how you wrote it.
Keep in mind what the report can and can't show. It shows which sentences Turnitin's model labeled, not how they were written. If those sentences are the ones you worked hardest on, or the ones you polished with a grammar tool, say so. If the highlighted parts are quotations or a definition you paraphrased from a source, point to the source.
Our guide to AI detector false positives goes through each step in detail.
Checking your draft before you submit
We aren't Turnitin and we don't have access to its model, so your score here won't match a Turnitin report, sometimes by a lot. What our AI text detector can do is show which sentences read like AI, so you can look at them again before you hand the work in. Paste your draft below, or use the full AI essay checker.
0 / 2,000 words
Frequently asked questions
Can students see their Turnitin AI score?
- Not by default. Turnitin says the AI indicator and report aren't visible to students, but instructors can download the report as a PDF and share it.
What AI percentage is too high on Turnitin?
- Turnitin doesn't set one. It says the score shouldn't be the only basis for action against a student. Your school or instructor decides what, if anything, a score leads to.
Does Turnitin detect AI in Spanish?
- Yes. As of November 2026 Turnitin lists English, Spanish, Japanese, Arabic, German, Dutch and Turkish. Detection of paraphrased and humanized text is English only.
Can Turnitin detect Claude or Gemini?
- Turnitin lists the models it covers, including Claude and Gemini models such as Claude Opus 4.6 and Gemini 3.1 Pro. It doesn't publish a detection rate for each model.
Does Turnitin flag QuillBot or Grammarly edits?
- Turnitin says AI paraphrasing tools such as QuillBot are covered. Grammarly's spelling and grammar fixes usually aren't flagged, but its generative features, like rewriting or drafting, likely are.
Can Turnitin detect humanized text?
- Since August 2025 it has tried to, for English. It doesn't name the tools it was trained on, and no detector catches everything.
Is there a free Turnitin AI checker?
- No. Turnitin sells to schools, not to individuals, and has no free public version. Sites calling themselves a “Turnitin checker” aren't Turnitin.
Is this site affiliated with Turnitin?
- No. We're an independent AI detector with our own detection engine. Turnitin is a trademark of its owner.