Guide for students and teachers

Is Turnitin’s AI detector accurate?

Turnitin’s AI detector is not 100% accurate, and Turnitin does not say it is. Turnitin says it aims to keep false positives under 1% for documents with more than 20% AI writing, and that it may miss some AI text to keep that rate low. Those are Turnitin’s own figures, and Turnitin says the AI score should not be the sole basis for action against a student.

Updated · Plagiarism Checker Plus editorial team

Need to check a text? Use our free plagiarism checker or AI detector.

The short answer

Turnitin’s AI writing detector is a prediction tool. It estimates how much of the prose in a paper was likely written or modified by AI. Turnitin publishes a goal for one kind of error, wrongly flagging human writing, and is open about the limits of its model. No independent body has confirmed Turnitin’s figures.

  • Turnitin’s claim: under 1% false positives for documents with over 20% AI writing, tested on more than 700,000 papers written before ChatGPT.
  • Turnitin’s caveat: to keep false positives low, it may miss some AI text. A paper reported as 50% AI could contain as much as 65%.
  • Turnitin’s rule for users: the percentage “should not be used as the sole basis for action or a definitive grading measure”.
Every statement about Turnitin on this page comes from Turnitin’s own help centre and blog, read on 17 November 2026, and the Sources section links each one. Accuracy figures are Turnitin’s claims about its own product. We have not tested them and cannot confirm them.

Turnitin’s own accuracy figures

Turnitin’s AI writing detection FAQ answers the accuracy question directly. It says Turnitin strives to keep its false positive rate under 1% for documents with over 20% AI writing. A false positive here means fully human-written text identified as AI-generated. Turnitin puts it as possibly flagging one in every 100 fully human-written documents.

It says that before every model update it tests more than 700,000 academic papers written before ChatGPT was released, to check that rate.

Three conditions sit inside that figure, and they matter:

  • It is a document-level figure. It describes whole documents, not individual highlighted sentences.
  • It applies above 20%. Below 20%, Turnitin shows only an asterisk (*%) and no highlights, to avoid false positives in that range.
  • It is a target Turnitin measures itself. The tests are Turnitin’s own, on its own data.

The sentence-level figure

In a June 2023 blog post, Turnitin’s Chief Product Officer gave a second number: a sentence-level false positive rate of around 4%. That means a sentence highlighted as AI has about a 4% chance of being human-written. Turnitin said these errors are more common in documents that mix human and AI writing, especially at the transitions, and that 54% of the time they sit right next to actual AI text. That post is from 2023, and Turnitin has updated its model since. In October 2026 it says it changed the English model to improve accuracy at exactly those transition points.

What it can miss

Accuracy has two sides. Flagging human writing is one error. Missing AI writing is the other. Turnitin says it accepts more of the second to keep the first low: “there is a chance that we might miss some AI written text in a document.” Its example is a document reported as 50% AI that could contain as much as 65%.

The FAQ also lists text the model does not score at all:

  • Non-prose. Lists, headers, references, data tables, formulas and code are removed before scoring. Turnitin says the model does not reliably detect AI text in bullet points or code.
  • Short papers. Depending on the language, documents with fewer than 300 words of evaluated sentences are not processed.
  • Unsupported languages. Only long-form English, Spanish, Japanese, Modern Arabic, Dutch, German and Turkish are processed. Detection of AI text changed by paraphrasing or “bypasser” tools is English only.

The percentage is therefore a share of the qualifying prose, not of the whole file. Our guide Does Turnitin detect AI? walks through the score states and file requirements in detail.

Where false positives happen

Turnitin names the kinds of human writing that are more likely to be flagged. Its FAQ lists three:

  • content without a lot of structural variation;
  • text that literally repeats itself;
  • text that has been paraphrased without developing new ideas.

It adds a fourth limit for short documents. With only a few hundred words, the prediction is mostly “all or nothing”, so a mix of AI and original writing could be flagged as entirely AI-generated.

On bias, Turnitin says it built its training data to include second-language learners, English users from non-English-speaking countries and less common subject areas, to reduce bias. Outside researchers have raised the concern that detectors in general flag non-native English more often. Our guide to AI detector false positives covers the common causes across tools.

What universities and researchers say

Turnitin’s figures come from Turnitin. Some institutions have looked at the same numbers and reached a different conclusion.

Vanderbilt University announced on 16 August 2023 that it had disabled Turnitin’s AI detection tool “for the foreseeable future”. Its reasoning used Turnitin’s own 1% claim: Vanderbilt submitted about 75,000 papers to Turnitin in 2022, so around 750 could have been wrongly labelled as partly AI-written. It also said Turnitin gave no detailed information on how the tool decides, and pointed to research finding that detectors are more likely to label writing by non-native English speakers as AI. That notice is from 2023, and Turnitin has since published more about its model.

Independent testing. A study by Weber-Wulff et al. (2023) tested 14 detection tools, including Turnitin, and concluded that the tools available were “neither accurate nor reliable”. It also found that obfuscation, such as paraphrasing, made the tools perform significantly worse. The study tested versions from 2023, not today’s model.

Whether your school uses the feature at all is a policy choice. See do colleges use AI detectors?

How to read a Turnitin AI score

In its 2023 post, Turnitin said there is no “right” or “target” AI score, just as there is none for similarity. A score is a likelihood, not proof. A few points help read it fairly:

  • *% means 1% to 19%. Turnitin hides the number and highlights in that range to avoid false positives.
  • 20% or more shows a number and highlights. It is the share of qualifying prose the model predicts was AI-generated or AI-modified. It is not a confidence level.
  • Highlighted sentences are “areas of interest”. That is Turnitin’s phrase. It tells instructors to use them to start a conversation, not to draw a conclusion.
  • It is separate from the similarity score. Turnitin says the two are completely independent. Our guide to the Turnitin similarity score explains the other number.

What counts as too high is set by your course, not by Turnitin. Our guide what percentage of AI is acceptable? covers how schools approach it.

If your work is flagged

Turnitin’s guidance for instructors on what to do when the AI percentage is high starts with the institution’s policy and treats the score as a way to open a conversation, not a punishment. It says the final decision on misconduct rests with the instructor. If you are the student:

  1. Ask to see the report. Students cannot see the AI indicator, but instructors can download it as a PDF and share it.
  2. Keep and bring your drafts. Version history in Google Docs or Word, notes, outlines and earlier drafts show how the work developed. See how to prove you didn’t use AI.
  3. Talk to your instructor. Be ready to explain your argument and your sources, and say which tools you used, such as a grammar checker.
  4. Follow your school’s process. Your academic integrity policy sets out the steps and how to appeal.

Getting a second opinion

Students usually cannot run Turnitin’s AI detection themselves. The Plagiarism Checker Plus AI detector is a different tool from the one your school uses. It gives a likelihood score with a sentence-level view, so you can see which passages read as AI-like and reread them. The first check is free without an account, up to 1,000 words, and a free account includes 3,000 words a month. Text is never used to train AI models.

It will not tell you what Turnitin will show, and no tool can. Different detectors are trained differently and can disagree. If you want to know which detector Turnitin runs, see what AI detector does Turnitin use?, and for the general method behind these tools, see how AI detectors work.

Sources

FAQ

Questions

No, and Turnitin does not claim it is. Its own FAQ says the AI percentage should not be used as the sole basis for action, sets a goal of keeping false positives under 1% for documents with over 20% AI writing, and says it may miss some AI text to keep that rate low.

Get a second opinion on your draft

Paste up to 1,000 words for a free check, no account needed. It is a different tool from the one your school uses, so it will not predict your Turnitin result. Treat its score as a likelihood, not proof.