Does Turnitin Detect AI? How the AI Writing Indicator Works and What Instructors See

Does Turnitin detect AI? Yes, if your institution licenses the feature and switches it on. Turnitin adds an AI writing indicator to the Similarity Report your instructor already uses, and it links to a report that highlights the sentences its model predicts were written or modified by an AI tool. What the number means, which text it ignores and how often it is wrong are all in Turnitin's own guides. This page walks through them, then covers what to keep if your work is flagged. It does not cover ways around the detector.

How does Turnitin detect AI writing?

The score is a share of sentences, not a verdict.

Turnitin's AI writing detection FAQ describes the method in a few steps. A model assigns each sentence a score between 0 and 1, judged in the context of the sentences around it. Sentences above a set threshold are labeled AI and the rest are labeled human. The document score is the percentage of evaluated sentences above that threshold. Turnitin says it updated the English model in October 2026 to improve accuracy where text switches between human and AI writing.

The model, the FAQ says, is a transformer-based deep learning system trained on Turnitin's in-house corpus of academic writing. It is not explicitly programmed to measure signals such as "burstiness" or "perplexity"; it learns statistical patterns, so a single prediction may not be explainable feature by feature. There is no list of reasons behind a flag.

The detector also looks for generated text that was run through a paraphraser or bypasser tool (the FAQ notes these are also called humanizers). Turnitin says that check covers English only, and it will not name the tools it trained against.

What text Turnitin checks, and what it skips

The indicator does not read your whole file. Turnitin's guide to using the AI Writing Report calls the text it evaluates "qualifying text": prose sentences in paragraphs that make up a longer piece of work, such as an essay, a dissertation or an article. The FAQ lists what is removed before scoring: lists, headers, references, data tables, and text that looks like formulas or code. The report guide adds that it does not reliably detect AI text in poetry, scripts or annotated bibliographies.

Because skipped text drops out of the count, the percentage is not a share of your whole submission: a paper with a long bulleted appendix is scored on its prose alone.

What the indicator shows, including the asterisk

The indicator can land in a handful of states. Since a July 2024 release recorded in Turnitin's AI writing detection model release notes, scores from 1% to 19% are not shown: the report displays *%, with no number and no highlights, because Turnitin found a higher incidence of false positives in that range. Reports made before July 8, 2024 may still show a number below 20%.

• 0%. No qualifying text was identified as likely AI-generated or AI-modified.
• An asterisk. Something between 1% and 19% was detected, and Turnitin attributes no score or highlights to it.
• 20% to 100%. The share of qualifying text identified as likely AI-generated, or AI-generated and then modified by a paraphraser.
• Gray with no percentage. The file could not be processed, usually because it misses a requirement in the table above or was submitted before the feature existed.

The AI percentage is separate from the similarity score. The FAQ says the two are completely independent: one measures matching text, the other predicted AI writing.

False positives: what Turnitin itself says

Turnitin's FAQ states its target as a false positive rate under 1% for documents with over 20% AI writing, and translates it: the detector might flag one out of every 100 fully human-written documents. Turnitin says it checks that rate before each release against more than 700,000 academic papers written before ChatGPT existed. To hold it down, Turnitin accepts missing some AI text; the FAQ's example is that a document scored at 50% could contain as much as 65% AI writing.

The same FAQ names the kinds of human writing most likely to be flagged: content without much structural variation, text that literally repeats itself, and text paraphrased without new ideas. In documents of only a few hundred words, it says, the prediction is mostly "all or nothing", so a short piece mixing your writing with AI text may be flagged as entirely AI. The May 2023 release notes add that generic opening and closing sentences had a higher false positive rate before a fix.

What instructors see, and what students see

The FAQ says the indicator is visible to instructors and administrators, and that an institution's admin can switch it off. Instructors can download the AI report as a PDF and share it. The newer report guide adds that an instructor can make the AI Writing Report available to students, who then see the percentage and the highlighted text for their own submission.

On grammar tools, the FAQ says Grammarly's spelling and grammar fixes were in most cases not flagged in Turnitin's tests, but text from its generative features, such as drafting or paraphrasing, will likely be flagged.

What to do if your work is flagged

A flag starts a conversation. What helps is evidence of how the work was made, and most of it already exists if you drafted in a document editor. These steps assume you wrote the work yourself.

1. Ask what was flagged

   Ask your instructor for the highlighted passages and the course's written policy on AI use.

2. Open your version history

   In Google Docs, the Editors Help page on finding what changed in a file explains that you click Last edit at the top right to browse earlier versions, and that you need edit permission on the file. Version history shows the text growing over days. How to use version history as an authorship record covers Word and other editors.

3. Gather your working material

   Collect outlines, notes, annotated sources and earlier drafts.

4. Explain the flagged passages

   Say where each highlighted idea came from. If a passage is generic, say so: Turnitin's FAQ lists repetitive, low-variation text among likely false positives.

5. Disclose any tool use plainly

   If you used a grammar checker, translator or chatbot for any step, say which and for what.

For a fuller checklist, see how to show your own authorship when a detector disagrees, and for building that evidence from the first draft, annotating drafts to show your process.

A worked example: reading a flagged report

Here is an example for illustration, not a real report. A student submits a 1,400-word essay with a bulleted timeline and a reference list. Those drop out, leaving about 1,000 words of prose in roughly 50 sentences. The report shows 22%: about 11 of the 50 sentences scored above the threshold.

The highlights sit in the opening and closing paragraphs, written last and in broad summary sentences such as this example: "The causes of the war were complex and had a lasting impact on the region." The body, full of dates and quotations, is not highlighted. The student shows version history with the body written across four sessions. That proves nothing alone, but it gives the instructor what the score cannot: context.

Common misreadings of a Turnitin AI score

• Reading 30% as "30% of my paper." It is 30% of the qualifying prose sentences, which can be much less than the whole file.
• Treating the asterisk as a hidden low score you failed. Turnitin shows no number in that range precisely because it is less reliable there.
• Assuming a second detector predicts Turnitin. Detectors use different models and often disagree.

Tips for students and teachers

• Draft in one document. One file with continuous history is easier to show than scattered copies; the AI writing guides cover marking drafts and sources fairly.
• Teachers: read before you ask. Compare the highlighted sentences with the student's earlier work and the course policy first.

Some students run a finished draft through a second detector to see which passages read as generic. Walter Writes AI's AI detector is one option: Walter Writes AI is an AI humanizer with a built-in AI detector, it supports over 80 languages, and every plan includes voice matching and the detector. A second opinion does not tell you what Turnitin will show, because detectors disagree with one another, and no detector, Walter's included, can guarantee 100% accuracy. Use it to find passages to reread, then revise them yourself.

The number on the report needs its own reading: what an AI percentage measures. For a second look at a single passage, checking one paragraph for AI sets out the reading checks, and a fair markup process covers what should happen next.

So does Turnitin detect AI? It predicts which prose sentences look AI-written and reports their share, with a 300-word minimum, no number below 20%, and a warning against using the score alone. The best protection is ordinary: draft in one place, keep your history, and be able to explain your sentences. For what detectors in general can establish, read what AI text detectors can tell you.