Related guides: Blackboard and SafeAssign, Moodle and its plugins and Google Classroom originality reports.
What Turnitin's report shows
Turnitin's AI Writing Report guide explains its indicator and highlights. The AI writing result is separate from the similarity score, which compares text with other sources. The two numbers answer different questions.
The report applies to qualifying text that meets Turnitin's file requirements. Short passages, lists and non-prose material may not be assessed in the same way as a longer document. Check the vendor's current requirements rather than assuming every submission is scored.

Can the report be wrong?
Yes. Turnitin says the model can misidentify human-written, AI-generated and AI-paraphrased text. It says the score should not be the sole basis for adverse action against a student. Its review guidance tells educators to weigh the result with their knowledge of the student, the work and institutional policy.
That caution matters in both directions. A high score does not prove misconduct, and a low score does not prove that no AI assistance was used. Neither result tells you what a course permitted or what the writer disclosed.
What if your work is flagged?
Keep drafts, notes, source lists and version history that show how the work developed. Ask what policy applies and how the institution reviews an AI report. Explain your process without treating the detector score as the whole case.
If you used a writing tool, compare its output with the original and check every claim, quotation and citation. Rewriting text to chase a score does not settle the authorship or disclosure question.
Does a humanizer guarantee a lower score?
No. We have no controlled Turnitin test that supports a promised result for Walter Writes AI or any other humanizer. Our student guide explains Walter's published terms, and our six rewrite examples document editing changes rather than detector scores.
For a decision about coursework, the course policy and the writer's process matter more than any single percentage.
Turnitin AI detection requirements at a glance
Turnitin only scores a paper when the account, the file and the language all qualify. The table below pulls those rules from Turnitin's own AI Writing Report guide and its help center article on why a report may not appear.
| Requirement | Turnitin's documented rule |
|---|---|
| License | The institution must hold an active Turnitin Originality license. Sub-accounts inherit the main account's status. |
| Minimum length | At least 300 words of prose in a long-form format. |
| Maximum length | No more than 30,000 words; file size under 100 MB. |
| File types | .docx, .pdf, .txt and .rtf. |
| Languages | English, Spanish, Japanese, Arabic, German, Dutch and Turkish. |
| Paraphrase and bypasser detection | English submissions only. |
| Low scores | Results above 0% and below 20% display as an asterisk (*%) with no highlights. |
| Student visibility | The help center says AI reports are not visible to students; instructors can download a report and share it. |
Turnitin's minimum used to be lower. The company's Chief Product Officer wrote in May 2023 that Turnitin raised the AI detection minimum from 150 to 300 words because accuracy improves with more text.

How to read the percentage and the highlights
The headline number is the share of qualifying text that the model judges likely AI-generated, or AI-generated and then run through a paraphraser or bypasser. Qualifying text means prose sentences inside paragraphs. Turnitin states that the model does not reliably assess poetry, scripts, code, bullet points, tables or annotated bibliographies.
That rule explains a common confusion. A lab report full of tables or a paper with a long annotated bibliography can show a percentage that does not line up with the highlighted passages, because only the prose portion was scored. Turnitin's guide names that disparity directly.
The submission breakdown separates likely AI-generated text from text that also shows signs of an AI paraphrasing tool or word spinner; the guide names Quillbot as an example. Turnitin adds that its Spanish and Japanese detectors do not include the paraphrasing and bypasser layer.
The asterisk matters most for students. Turnitin's guide says its testing found a higher incidence of false positives between 0 and 19 percent, so reports created after July 8, 2024 show *% instead of a number in that band.
Turnitin's published false positive figures
Turnitin has put numbers on its error rate, and those numbers are the right starting point for any appeal. In a May 23, 2023 update, Chief Product Officer Annie Chechitelli reported a document false positive rate below 1% for documents with more than 20% AI writing.
The same update gave a sentence-level false positive rate of about 4%. Turnitin also found that 54% of false positive sentences sat right next to actual AI writing, which means mixed documents are where wrong highlights cluster. The company validated the document figure on 800,000 academic papers written before ChatGPT was released.
Scale changes how a 1% rate feels. Vanderbilt University calculated that it submitted 75,000 papers to Turnitin in 2022, so a 1% false positive rate could have mislabeled around 750 student papers.
Language background is the other live issue. The Stanford study by Liang and colleagues (2023) found that GPT detectors consistently misclassified non-native English writing as AI-generated. Stanford HAI reported that seven detectors labeled 61.22% of TOEFL essays as AI-generated. Turnitin was not one of the seven, and Turnitin later published its own evaluation reporting no statistically significant bias against English language learners, while pointing out that the Stanford result rested on 91 TOEFL practice essays.
Universities that switched the AI indicator off
Turnitin's AI detector is a setting an institution controls, and several well-known universities have turned it off. Their public statements are useful evidence if your school is reviewing a flag, because they show how other academic bodies weigh the same tool.
| Institution | Decision | Date |
|---|---|---|
| Vanderbilt University | Disabled Turnitin's AI detection tool "for the foreseeable future," citing reliability and transparency concerns. | August 16, 2023 |
| University of Pittsburgh Teaching Center | Disabled the AI-detection tool in Turnitin and does not endorse any AI-detection tool. | Page last reviewed September 15, 2026 |
| UCLA | Opted out of the AI writing detection preview at launch, alongside peer UC campuses. | April 4, 2023 |
| Curtin University | Disabled Turnitin's AI writing detection across all campuses; text matching stays on. | From January 1, 2026 |
Pittsburgh's Teaching Center states that current AI detection software is not reliable enough to deploy without a substantial risk of false positives. Curtin's notice is equally clear that only the AI feature was switched off, so its students still receive similarity checks on every Turnitin submission.

How to protect your work before and after you submit
The strongest defense against a wrong flag is a record of how the paper was written. Build that record while you work, not after a score appears.
- Draft in a tool with version history. Google Docs keeps earlier versions, and Google's help page explains how to create named versions so they are not merged together.
- Name milestone versions. Label the outline, first draft and final edit so a reviewer can see the paper grow.
- Keep source notes and search history. Notes that match your citations show the research happened before the writing.
- Disclose permitted AI help. Vanderbilt points students to APA and MLA guidance for citing AI tools when a course allows them.
- Check your file against the requirements. A paper under 300 words of prose, or one in an unsupported language, will not receive an AI score at all.
- Ask for the report itself. Students cannot see the AI report by default, so request the downloaded copy and the highlighted passages before you respond.
- Answer with evidence, not a second score. Walk through drafts, notes and your argument; a different detector's number does not settle authorship.
If you want a private read before submitting, the free Walter Writes AI detector returns an AI probability score on pasted text. Use it as a drafting check alongside your own records, since Turnitin's model and settings are separate.