Turnitin AI Detection for Instructors: What the Score Tab Does and Doesn’t Tell You

A student’s paper opens, and there it is in the corner: a tab labeled “AI” with a percentage next to it. Maybe it reads 64%. Maybe it’s a single asterisk where you expected a number. Most instructor guides argue about whether to trust AI detection in the abstract, then skip the thing you’re actually staring at. So let’s read that screen together, element by element, and name what the tab quietly leaves out.

Key takeaways

• Turnitin’s AI writing indicator lives in its own tab, separate from the Similarity Report, and shows the estimated percentage of *qualifying* text it predicts was AI-generated, not a measure of cheating.
• Turnitin generally surfaces a percentage only when a submission has roughly 300 words or more of long-form prose. Short answers, bullet lists, and code largely fall outside what the indicator scores.
• An asterisk (*%) appears instead of a number when the score lands between 1% and 19%. Turnitin hides those low scores and their highlights on purpose, because that range produces more false positives.
• The tab tells you nothing about a student’s writing process, their known voice, draft history, or whether they’re a non-native English speaker more likely to be falsely flagged.
• Use the score to decide whether to look closer, never as standalone evidence. Pair it with drafts, version history, and a conversation before any integrity decision.

Where the AI score lives, and why it’s not the Similarity Report

First, a distinction that trips up a lot of faculty. The familiar Similarity Report, with its color-coded matches against other papers and web sources, is a different feature from the AI writing report, and they sit in separate tabs. Similarity asks whether this text matches something that already exists. The AI indicator asks something completely different: does this text *read like* it was generated by a language model?

That second question has no source document behind it. Nothing to click through to, no original to compare against. It’s a statistical guess about the *style* of the prose, not a match against a database. Once that clicks, the percentage stops looking like a smoking gun and starts looking like an estimate.

Reading the AI writing indicator percentage

The headline number is the percentage of qualifying text Turnitin predicts was generated by AI. Two words in that sentence carry the weight.

“Qualifying” matters because Turnitin doesn’t score everything in the document. Its model is built for long-form English prose, and Turnitin’s own documentation says a submission generally needs about 300 words before the tool returns a result. Quotations, references, lists, and code typically get excluded. So a “64%” isn’t 64% of the whole file. It’s 64% of the slice the tool decided it could evaluate.

“Predicts” matters because the number is a probability, not a count. It’s tempting to read 64% as “the machine wrote roughly two-thirds of this.” Wrong reading. The figure reflects how much of the qualifying text fell on the AI-like side of the model’s decision boundary, and a confident model can still be confidently wrong. OpenAI shut down its own AI-text classifier in July 2023, citing a low rate of accuracy. If the company building the language models couldn’t reliably detect their own output, no single vendor’s percentage deserves blind faith.

For the machinery underneath this, we wrote a deep dive on the statistics behind how detectors score text. The short version: these tools reward unpredictable, uneven writing and penalize smooth, predictable prose. Hold that thought.

The asterisk, and what it really signals

Sometimes you won’t see a percentage at all, just an asterisk, like *%, where the number should be. That’s not a glitch, and it isn’t a high score in disguise. Turnitin shows the asterisk when its model puts the document in the 1% to 19% range, and there it withholds both the figure and the highlights. Its stated reasoning: scores under 20% produce more false positives, so a precise number would imply confidence the tool doesn’t have.

So an asterisk is close to a non-result. The model saw little or no AI-like text and won’t commit to a number, which is not the same as quietly flagging a borderline cheater.

The highlighted segments

When a document does score 20% or higher, Turnitin highlights the passages it flagged as AI-like. That’s genuinely useful: it shows you *where* the signal sits instead of handing you one blunt figure. But notice the limit. The highlights tell you which sentences *read* predictable to the model, not who typed them. A flawless, evenly paced topic sentence gets highlighted whether a model produced it or a careful student wrote it that way on purpose.

A mini-scenario: the 71% on an honest paper

Picture Priya, a sophomore who turns in a 600-word reflection. The AI tab reads 71%, no asterisk, highlights and all.

The reflexive move is to file a referral. The better move is to read what the screen is telling you. The highlights cluster in two paragraphs of tidy, abstract prose, exactly the kind of smooth writing detectors over-flag. So you open the Google Doc version history and watch the paper grow over two evenings: deletions, a reordered paragraph, a typo fixed at 10:40 p.m. You pull Priya’s last two assignments and the voice matches the slightly formal register that, you now remember, is just how she writes, because English is her second language.

Then you ask her to walk you through it. She does, easily, and tosses in a detail about her grandmother that isn’t even in the essay. The 71% hasn’t moved; your reading of it has. The number was real; the cheating wasn’t. That review took fifteen minutes, and it’s the gap between responsible instruction and a wrongful accusation.

What the Score tab does NOT tell you

Here’s the section every instructor needs and the interface won’t give. The tab is silent on:

• Process. It can’t see drafts, outlines, or version history. A document that grew messily over three nights and one pasted in as a single block look identical to the indicator.
• The student’s known voice. You’ve got years of writing samples in your head. The model has none, so it can’t tell that this *is* how Priya always writes.
• Who’s at higher risk of a false flag. Its errors aren’t random. A 2023 study from Stanford researchers, with James Zou as senior author, found several AI detectors disproportionately misclassified writing by non-native English speakers as AI-generated, because second-language writers often lean on steadier, more predictable phrasing. The percentage describes how *regular* the writing looks, never how *honest* the writer is.
• Intent. Even if AI was involved, the tab can’t tell a student who used an allowed grammar checker from one who pasted a whole essay. That call is yours.

A useful mental model: the Score tab is a smoke alarm. A beep means look around the room. It doesn’t mean the building’s on fire, and it can’t tell you who lit the match.

Turning the score into a fair process

You don’t have to abandon the tool. Just keep it in its lane, as one input among several.

1. Set the policy before anyone writes. Put your AI rules and your use of detection in the syllabus, in plain language, before the first assignment.
2. Treat a flag as a question, not an answer. A high percentage triggers a review, never an automatic penalty.
3. Gather what the tab can’t show. Version history, in-class writing samples, earlier drafts, and a short conversation tell you more than any number.
4. Document what you reviewed, either way. A record protects an innocent student from the next suspicion, and it’s what actually survives an appeal, where a lone percentage won’t.

If your department is weighing tools and tiers, our guides for educators and writers go deeper on detection and integrity, and you can compare what each plan covers.

Frequently asked questions

Does the Turnitin AI percentage mean that much of the paper was copied or cheated?

No. The percentage is Turnitin’s estimate of how much of the *qualifying* long-form text matches patterns it associates with AI generation. It’s a prediction about prose style, not a measurement of plagiarism or intent. A student can hit a high percentage on writing they composed themselves, and a low one on work a model produced. Treat it as a prompt to investigate, never as a finding of misconduct.

Why does Turnitin show no AI score on some submissions?

Turnitin’s AI writing detection is built for long-form English prose and generally needs about 300 words before it returns a percentage. Short answers, heavily bulleted documents, lists, code, and lab data often fall below that threshold or outside what the model scores, so the indicator stays blank. A missing score isn’t a clean bill of health, and it isn’t a flag. It usually just means the document didn’t meet the conditions the tool was designed for.

What does the asterisk next to a Turnitin AI percentage mean?

The asterisk (*%) appears in place of a number when Turnitin’s model puts the document in the 1% to 19% range. Turnitin withholds the figure and the highlights there on purpose, saying low scores produce more false positives. So an asterisk is closer to “nothing to see here” than to a borderline accusation. Don’t read it as a hidden high score.

Can I open an academic integrity case using only the Turnitin AI score?

You shouldn’t, and most institutions’ integrity policies don’t let a single probability carry a case on its own. The Score tab reports a likelihood with an acknowledged error rate; it’s evidence about text, not proof about a person. Before any formal step, gather what it can’t show: earlier drafts, version history, the student’s voice from past work, and a conversation about how they wrote the piece.

Are some students more likely to be wrongly flagged by Turnitin’s AI indicator?

Yes. A 2023 study from Stanford researchers, with James Zou as senior author, found that several AI detectors disproportionately misclassified writing by non-native English speakers as AI-generated, because second-language writers often use steadier, more predictable phrasing that reads as low perplexity. Students who write in plain prose, or were drilled on rigid essay templates, can trip the same wires. The percentage describes how regular the writing looks, not how honest the writer is.

Reading the screen, not fearing it

The Turnitin AI tab is a tidy little box that wants to feel like a verdict. It isn’t one. It’s a statistical estimate of qualifying prose, with an asterisk that admits its own uncertainty and highlights that show patterns, not people. Weigh it against process and voice, and keep the final call human.

Want to see how slippery these scores really are? Run a passage through a checker yourself, starting with a paper you trust, and watch how honest writing actually behaves.