Quick answer: Turnitin’s AI detector is one of the more accurate tools available, but it is not accurate enough to prove misconduct on its own. Turnitin says its document-level false positive rate is under 1% for documents with more than 20% AI writing, and that about 4% of sentences it highlights may actually be human-written. In an independent 2023 study of 14 detectors, Turnitin ranked first with 76% accuracy. Turnitin itself says a high score is not necessarily evidence of misconduct.
Key takeaways
- Turnitin reports a document false positive rate under 1%, but only for documents with more than 20% AI writing.
- About 4% of highlighted sentences may be human-written, by Turnitin’s own figure.
- An independent study ranked Turnitin first of 14 tools, at 76% accuracy.
- Scores under 20% are treated as unreliable and shown with an asterisk.
- Only instructors and administrators see the AI score, and it is meant as a starting point for a conversation.
What Turnitin says about its accuracy
In a May 14, 2023 update, Turnitin’s Chief Product Officer, Annie Chechitelli, shared these figures from Turnitin’s own testing, including a test of 800,000 academic papers written before ChatGPT existed:
| Turnitin figure | What it means |
|---|---|
| Document false positive rate under 1% | For documents with more than 20% AI writing, fewer than 1 in 100 fully human documents are wrongly labeled. |
| Sentence-level false positive rate about 4% | A sentence highlighted as AI has roughly a 4% chance of being human-written. |
| More false positives below 20% | When Turnitin detects less than 20% AI writing, false positives are more common, so these scores get an asterisk. |
| 54% of false positive sentences sit right next to real AI writing | Most wrongly flagged sentences are close to AI-written text, often where human and AI writing meet. |
These are Turnitin’s own numbers, measured on its own tests. They are useful, but they come from the company that sells the product.
What independent research found
| Source | What it found |
|---|---|
| Weber-Wulff et al. (2023) | Tested 14 detectors on 54 documents, including human writing, machine-translated text, AI text, edited AI text and paraphrased AI text. Turnitin ranked first, classifying 41 documents correctly (76% accuracy). No tool reached 80%, and the authors concluded that detection tools are “neither accurate nor reliable.” |
| Vanderbilt University (2023) | Turned off Turnitin’s AI detector. It noted that even a 1% false positive rate could have meant around 750 of the 75,000 papers it submitted in 2022 being wrongly flagged. |
| RAID benchmark (ACL 2024) | Found that AI text detectors in general are easily fooled by adversarial attacks and changes in how text is generated. |
The picture is consistent: Turnitin does better than most detectors, especially on unedited AI text, but edited, paraphrased and mixed writing are much harder for it, as they are for every tool.
Limits you should know about
Based on a university guide to Turnitin’s AI detection in Canvas (University of Wisconsin-Superior):
- Length: the AI check needs between 300 and 30,000 words of qualifying text. Shorter submissions get no AI result.
- Language: submissions in unsupported languages get no AI result.
- Low scores: scores from 1% to 19% show an asterisk and no highlights, because false positives are more likely in that range.
- Who sees it: only instructors and administrators, not students.
- Not proof: a high score is not necessarily evidence of misconduct, and a low score does not prove AI was not used.
When Turnitin is more and less reliable
| Situation | Reliability |
|---|---|
| Long, unedited text pasted from a chatbot | Higher |
| Fully human writing in a long essay | Higher (Turnitin reports under 1% false positives above the 20% threshold) |
| Scores under 20% | Lower: Turnitin itself marks these as less reliable |
| Human writing next to AI-assisted parts | Lower: most false positive sentences sit near real AI writing |
| AI text that was edited or paraphrased | Lower |
| Short submissions under 300 words | No AI result at all |
❌ Turnitin says 35% AI, so the student cheated.
✅ Turnitin found AI-like writing in about 35% of the text, so the next step is to look at drafts and talk with the student.
What to do if Turnitin flags your work
- Ask what the score was and how it is being used. A percentage alone is not proof, and the detector companies say so themselves.
- Show your version history. In Google Docs, go to File > Version history > See version history. Timestamped drafts are strong evidence of your process.
- Bring your notes and sources. Outlines, research notes and earlier drafts show how the work came together.
- Offer to talk through your work. Explaining your argument in your own words is often the most convincing evidence.
- Point to the research. Independent studies show detectors make mistakes. See our guide are AI detectors accurate?
For teachers
- Treat the AI score as information for your review, not a verdict: Turnitin says it does not determine misconduct.
- Be cautious with scores under 20% and with mixed human and AI writing, where false positives cluster.
- Ask for drafts or version history before drawing conclusions.
- Make your AI policy clear in the syllabus and in each assignment.
Check your writing before you submit
You can paste your text into our free AI detector to see which sentences read most like AI. It is a different tool from Turnitin, so treat it as a second opinion. If your own writing looks generic to a detector, our list of AI words to avoid and the naturalness check in our AI humanizer can help you revise in your own voice.
Related tools and guides
- AI detection guide: everything in one place
- Are AI detectors accurate?
- Does Canvas detect AI?
- Is QuillBot’s AI detector accurate?
- Is GPTZero accurate?
- Free AI detector
Frequently asked questions
How accurate is Turnitin’s AI detector?
Turnitin reports a document false positive rate under 1% for documents with more than 20% AI writing, and an independent 2023 study ranked it first of 14 tools at 76% accuracy. It is among the better detectors, but it can still be wrong.
Can Turnitin’s AI detector be wrong?
Yes. Turnitin says about 4% of the sentences it highlights may be human-written, and false positives are more common when it detects less than 20% AI writing.
Does Turnitin detect AI in short assignments?
No. The AI check needs at least 300 words of qualifying text, so shorter submissions get no AI result.
Can students see their Turnitin AI score?
No. Only instructors and administrators can see the AI writing indicator and report.
Is a Turnitin AI score proof of cheating?
No. A high score is not necessarily evidence of misconduct, and Turnitin describes the score as information to help educators make a decision.
Sources
- Turnitin: AI writing detection update from Turnitin’s Chief Product Officer, Annie Chechitelli (May 14, 2023)
- Weber-Wulff et al.: Testing of detection tools for AI-generated text (International Journal for Educational Integrity, 2023)
- Vanderbilt University: Guidance on AI detection and why we’re disabling Turnitin’s AI detector (August 16, 2023)
- University of Wisconsin-Superior: How to use and interpret Turnitin AI writing detection in Canvas
- Dugan et al.: RAID benchmark (ACL 2024)
