Does turnitin detect AI?
Yes. Turnitin launched AI writing detection in April 2023, and it gives instructors a percentage estimate of how much of a paper it believes is AI-written. Turnitin claims under 1% false positives at the document level but admits missing about 15% of AI text. Some universities, including Vanderbilt, disabled it.
Why — the first-principles explanation
Turnitin's AI detector is a separate system from its plagiarism checker, and confusing the two causes most of the panic around it. The plagiarism checker does something conceptually simple and verifiable: it compares your text against a database of existing documents and shows you the matching source. You can click the link and see it. The claim is checkable.
The AI detector does something fundamentally different. There is no database of AI-written papers to match against — the same prompt produces different text every time. So instead it makes a statistical judgment: it estimates how predictable your word choices are compared to how a human typically writes. Fluent, plain, formal prose scores as more AI-like, because that's what predictable writing looks like. There is no source to click. There is no receipt. It is an opinion expressed as a number.
That difference is why the two tools deserve different levels of trust. A plagiarism match is evidence. An AI percentage is an inference — and every inference has an error rate that lands on real people.
Turnitin's own framing shows the tradeoff explicitly. It says it keeps document-level false positives under 1%, and to buy that low rate it accepts missing roughly 15% of AI writing. That's a deliberate dial: catch less, accuse fewer innocents. But even a 1% error rate isn't small at scale — Vanderbilt pointed out that across its 75,000 annual submissions, 1% means about 750 falsely flagged papers, and cited that, plus bias against non-native English writers and Turnitin's lack of transparency about how the tool works, when it disabled the detector in August 2023.
An example that makes it click
Think of two kinds of security guard. The first guard has a photograph of a stolen bicycle. He stops you, holds the photo next to your bike, and either they match or they don't. Anyone watching can judge for themselves. That's the plagiarism checker.
The second guard has no photo. He just watches how you walk and says, "78% of people who walk like that stole their bike." Maybe he's right on average. But you can't inspect his reasoning, you can't show him a receipt that changes his mind, and if you happen to walk that way because of an old injury, he'll stop you every single time. That's the AI detector — and that second guard is the one being pointed at student essays.
Key facts
- Turnitin released its AI writing detection feature in April 2023, enabled for customers with less than 24 hours' advance notice (Vanderbilt University).
- Turnitin states a document-level false positive rate of under 1% for documents with more than 20% AI writing, and says it validated this against a large corpus of pre-ChatGPT academic papers.
- Turnitin acknowledges the tradeoff: it accepts missing roughly 15% of AI-generated text in order to keep false positives low.
- Vanderbilt University disabled Turnitin's AI detector in August 2023, noting that a 1% false positive rate across its 75,000 annual submissions would falsely flag about 750 papers.
- A Washington Post test on a smaller sample found a far higher error rate than Turnitin's claimed figure, and independent research documents elevated false positives for non-native English writers.
- The AI score is separate from the similarity/plagiarism score and, unlike a plagiarism match, points to no clickable source.
▶ The 60-second explainer (script)
Does Turnitin detect AI? Yes — since April 2023, it gives your instructor a percentage estimating how much of your paper it thinks was AI-written. But here's what matters, and almost nobody explains it: that's a completely different tool from the plagiarism checker, and it deserves completely different trust. The plagiarism checker compares your text to a database and shows the matching source. You can click it. It's evidence. The AI detector has no database — the same prompt makes different text every time. So it just measures how predictable your writing is. Plain, formal, careful prose scores as more AI-like. There's no source to click. It's an opinion, printed as a number. Turnitin says its false positive rate is under one percent, and to get that low it accepts missing about fifteen percent of real AI writing. But one percent isn't nothing. Vanderbilt did the math: across seventy-five thousand papers a year, that's about seven hundred fifty students wrongly flagged. They disabled the tool. If you're flagged, the score isn't proof. Your drafts and version history are the real evidence.
What authoritative sources say
People also ask
Can students see their own AI score?
Generally no. The AI writing indicator is shown in the instructor view, not on the student similarity report, so most students never see the number that gets discussed about them.
Is a high AI score proof of cheating?
No. It's a probabilistic estimate with no underlying source document. Institutions that use it responsibly treat it as a prompt for a conversation, not a finding of misconduct.
What if I wrote it myself and got flagged?
Ask for the specific evidence beyond the score, and offer your Google Docs or Word version history, drafts, and notes. Process evidence is concrete in a way the score isn't.
Does using Grammarly trigger the AI detector?
Basic spelling and grammar fixes generally don't. Generative features that rewrite or produce sentences for you are a different matter and can push a score up, which is why heavy rewriting tools are risky if your school bans AI assistance.
Do all schools use it?
No. Turnitin's AI detection is a setting institutions can turn off, and several — including Vanderbilt — have done exactly that. Check your own school's policy rather than assuming.
The same question, asked other ways
- Does turnitin check for AI?2,400/mo
- Can turnitin detect AI?1,600/mo