Turnitin AI Detection: What Teachers See in the Report, and What the Number Means
Most students never see the Turnitin AI detection report. They hear about it second-hand, in a sentence like "the report says 40 percent," and have to guess what sits behind the number. This guide describes the instructor's side of the screen using Turnitin's own published documentation: the percentage, the highlighted sentences, the asterisk, the states the indicator can show, and the vendor's own statement that none of it is proof of misconduct.
Knowing what the report contains changes the conversation. A student who understands the number can answer it; a student who imagines it as a verdict cannot.
In short: Instructors see an AI writing indicator inside the Similarity Report: an overall percentage of qualifying prose the model thinks was AI-generated or AI-paraphrased, plus a report that highlights those sentences. Students cannot see it. Turnitin says the percentage should not be the sole basis for action, and scores under 20 percent show only an asterisk.
What the Turnitin AI detection report actually shows
Turnitin's AI writing detection is a feature of the Similarity Report, the same report that has flagged copied text for two decades. In the new report view the instructor opens an "AI Writing" tab; in the classic view an indicator sits in the right-hand panel. Either way, the indicator carries a single number: the percentage of qualifying text that the model predicts was generated by an AI tool, or generated and then run through a paraphraser. Selecting it opens a report in which the suspect sentences are highlighted.
Two facts about visibility matter to students. First, Turnitin's FAQ states that "only instructors and administrators are able to see the indicator." The student view of the Similarity Report does not include it. Second, an instructor can download the AI report as a PDF and share it, and a student in a misconduct conversation is entitled to ask for exactly that. If you have been told a number, ask to see the highlighted sentences behind it.
The vendor also frames its own product carefully. The same FAQ says Turnitin "does not make a determination of misconduct" and that the percentage "should not be used as the sole basis for action or a definitive grading measure." Those are Turnitin's words, and they belong in any conversation that starts with the report.
How Turnitin AI detection calculates the percentage
According to Turnitin's documentation, a submission is split into sentences, the sentences are grouped into overlapping segments, and each segment is scored between 0 and 1 for the probability that it is AI-generated. Each sentence inherits its segment's score; sentences that fall in more than one segment have their scores pooled. The sentence scores are then aggregated into the document percentage. The company says its current model is a transformer-based classifier trained on its own corpus, and that it is "not explicitly programmed" to compute the perplexity and burstiness measures that public discussion of how AI detectors work tends to focus on. It learns statistical patterns from data and does not explain individual predictions.
The percentage counts only "qualifying text," which Turnitin defines as prose sentences in long-form writing. Lists, bullet points, headings, code and other non-sentence material are excluded. This is why the number and the amount of highlighted text can disagree: a report can say 40 percent while the highlights cover far less of the page, because the denominator is prose, and only prose.
| What the instructor sees | What it means | What a student should know |
|---|---|---|
| Blue indicator with a percentage (20 to 100) | The submission processed; the number is the share of qualifying prose predicted to be AI-generated or AI-paraphrased | Highlights exist and can be shared as a PDF; ask for them |
| Asterisk (*%) | AI writing was detected in the 1 to 19 percent range; Turnitin shows no exact number and no highlights because false positives are more likely here | An asterisk is the weakest possible signal and should not lead to an accusation |
| Gray dashes (- -) | The file could not be processed: under 300 words of prose, over 30,000 words, over 100 MB, an unsupported language or file type | No AI score was produced at all |
| Error (!) | Processing failed on Turnitin's side | Nothing can be inferred from an error |
What the number means, and what it does not
Turnitin publishes a document false-positive rate of under one percent for documents in which it detects more than 20 percent AI writing, and it says it re-tests every model release against more than 700,000 papers written before ChatGPT existed. The company is candid that keeping false positives low means missing some machine text: its FAQ gives the example that a document scored at 50 percent "could contain as much as 65% AI writing."
That percentage score is easy to misread in both directions. One percent of documents is small until multiplied by an institution's volume; Vanderbilt University, explaining in August 2023 why it disabled the feature, calculated that one percent of its 75,000 annual submissions would be around 750 wrongly flagged papers. And the sentence-level rate is higher than the document rate. Turnitin's chief product officer reported in May 2023 that roughly four percent of highlighted sentences are human-written, most of them sitting next to genuine AI text in mixed documents.
What the number does not do is identify a tool, a prompt, or a date. It cannot distinguish a paragraph pasted from a chatbot from a paragraph a student wrote after reading a chatbot's summary. It cannot tell an instructor whether AI use was permitted for the assignment; Turnitin's own statistics on AI writing in submissions include courses where the tool was assigned. And it is fully independent of the similarity score. A paper can show zero percent similarity and a high AI percentage, or the reverse, and neither influences the other.
Why short and formulaic passages score high
Turnitin's FAQ names the kinds of human writing that produce false positives: "content without a lot of structural variation, text that literally repeats itself, or text that has been paraphrased without developing new ideas." Add the length problem. Documents of only a few hundred words are scored on a single segment with no overlap, so the prediction is, in Turnitin's words, "mostly all or nothing." The company raised its minimum from 150 to 300 words for this reason, and it changed how it aggregates the first and last sentences of a document after finding extra false positives in introductions and conclusions.
Translate that into a literature classroom. A five-paragraph essay on Animal Farm that restates the plot in even, declarative sentences ("Napoleon takes power. The pigs change the commandments. The animals cannot read them.") has no structural variation and develops no idea, and it can score high whoever wrote it. The paragraph that argues a position, quoting Squealer's ever-rising production figures and asking why the animals accept numbers that contradict their hunger, varies its sentence length and moves toward a claim. Our guide to building a thesis that is actually an argument exists for reasons older than detectors, and it happens to describe the writing least likely to be flagged. Grammar checking, for what it is worth, is not the issue: Turnitin states that ordinary spelling and grammar corrections from Grammarly were not flagged in its tests, while Grammarly's generative features were.
How to read "the report says 40 percent"
When an instructor opens with that sentence, hear it as a description of a screen, not a finding. Then ask three questions in order. Which sentences are highlighted? A 40 percent score on a 1,200-word essay means roughly 480 words of prose were flagged, and you can look at every one of them. Were any of the highlighted sentences quotations, block quotes, or the introduction and conclusion, where Turnitin itself reports more false positives? And what does the course policy say about how the score may be used?
Then move the conversation to the essay. Explain where the flagged paragraph came from, show the draft in which it first appeared, and talk about the text it analyses. Turnitin's own advice to educators is to use the score "to start a meaningful and impactful dialogue with their students," and a student who arrives ready for dialogue, with drafts and reading notes, is meeting the instructor on the ground the vendor recommends. If the conversation has already become an accusation, our step-by-step guide for anyone falsely accused of using AI covers the reply, the evidence and the hearing. If you are trying to understand a flag before any conversation has happened, why is my essay flagged as AI works through the usual causes.
Frequently asked questions
Can students see the Turnitin AI detection score?
No. Turnitin's documentation states that the indicator and report are visible only to instructors and administrators. An instructor can download the report as a PDF and share it, and you may ask for it.
What does the asterisk in a Turnitin AI report mean?
An asterisk replaces the percentage when detected AI writing is between 1 and 19 percent. Turnitin shows no exact number and no highlights in that range because false positives are more likely there.
Does Turnitin AI detection flag Grammarly?
Turnitin reports that ordinary spelling, grammar and punctuation corrections from Grammarly were not flagged in its tests on human-written documents. Text produced by Grammarly's generative features, such as drafting or paraphrasing, will likely be flagged.
Is a Turnitin AI detection percentage proof of cheating?
Turnitin says no: it "does not make a determination of misconduct," and the percentage "should not be used as the sole basis for action." Institutions such as Vanderbilt have disabled the feature over false-positive concerns.
Conclusion: reading Turnitin AI detection as a description, not a verdict
Turnitin AI detection gives an instructor a percentage and a set of highlighted sentences, computed over prose only, with an asterisk for weak signals, a published error rate of under one percent per document and around four percent per sentence, and an explicit instruction from the vendor to treat the result as the start of a conversation. A student who knows that can ask for the highlights, account for each flagged sentence, and bring the drafts that show where the writing came from. The rest of the Authorship pillar is about making that evidence exist before anyone asks for it.
The essays on this site are study companions for reading and writing about literature; they are never submission-ready coursework.
Sources
- Turnitin Guides, "Turnitin's AI writing detection capabilities FAQs": what the indicator shows, instructor-only visibility, the qualifying-prose rule, the asterisk range, file requirements, the under-one-percent document rate, the 50-to-65 percent example, the false-positive text types, the Grammarly statement, and the "sole basis for action" language
- Turnitin Guides, "How to access the AI Writing Report": the AI Writing tab in the new report, the indicator in the classic report, and the blue, asterisk, gray and error states
- Turnitin, "AI writing detection update from Turnitin's Chief Product Officer" (May 2023): the 300-word minimum, the first-and-last-sentence adjustment, the four percent sentence-level rate, and the advice to use the score to start a dialogue
- Vanderbilt University, "Guidance on AI detection and why we're disabling Turnitin's AI detector" (2023): the decision to disable the feature and the 750-papers calculation