How does Turnitin's AI detection work?

Turnitin's AI writing indicator uses a model trained to recognize statistical patterns typical of AI-generated prose. It segments the submission and estimates the share of qualifying text that matches those patterns, reporting a percentage separate from the similarity score.

Why are low scores hidden or marked with an asterisk?

Per Turnitin's July 2024 update, scores between 1 and 19 percent are suppressed and flagged with an asterisk because false positives are more likely in that band. The design choice is meant to reduce the chance of wrongly flagging human writing.

Can Turnitin flag human writing as AI?

Yes. Turnitin disclosed in 2023 that its sentence-level false-positive rate is around 4 percent. Weber-Wulff et al. (2023), in the International Journal for Educational Integrity, concluded across 14 tools that detectors are neither accurate nor reliable. A flag is not proof, which is why several institutions treat the score cautiously.

Why have some universities disabled it?

Vanderbilt disabled the AI detector in August 2023 over accuracy and fairness concerns. Curtin University announced in September 2025 that it would disable the feature from 1 January 2026. The common reasons are false positives, limited transparency into the model, and the serious consequences of a wrong accusation.

What should students know about a Turnitin AI score?

The score is an estimate, not a judgment. Keep your drafts, notes, and version history so you can show your process, understand your institution's policy, and raise any concern with your instructor. A flag can be discussed and reviewed.

A plain explainer, not a workaround

How Turnitin AI detection works

Turnitin's AI writing indicator estimates how much of a submission matches patterns typical of AI-generated text. This page explains what the score means, how accurate it is, and why some universities have turned it off.

Refine does not help anyone evade detection. This is an informational guide for students and educators who want to understand the tool.

Reading the indicator

What the AI writing score actually reports.

The indicator is separate from the similarity score. It estimates the share of qualifying text attributed to AI, and it is meant to prompt review rather than decide it.

01
Find the AI writing indicator
In the Turnitin Similarity report, the AI writing indicator is separate from the similarity score. It reports an estimated percentage of the submission that its model attributes to AI-generated text.
02
Read the percentage in context
The number is an estimate of qualifying text flagged by the model, not a measure of misconduct. It should be read alongside drafts, the assignment, and your knowledge of the writer.
03
Treat it as a prompt, not a verdict
A flagged score is a reason to look more closely and, where appropriate, to talk with the student. It is not standalone proof that a policy was broken.

Accuracy and thresholds

What the research and the thresholds show.

Independent studies and Turnitin's own disclosures show mixed results, and the scoring design reflects a real risk of flagging human writing.

Suppressed low scores: Per Turnitin's July 2024 update, scores from 1 to 19 percent are suppressed and shown with an asterisk because the model has a higher incidence of false positives in that range.
Controlled submission tests: Perkins et al. (2023), in the Journal of Academic Ethics, found Turnitin flagged 91 percent of experimental GPT-4 submissions as containing AI content, but only 54.8 percent of the text inside those submissions was marked as AI-generated.
Broader reliability warnings: Weber-Wulff et al. (2023), in the International Journal for Educational Integrity, tested 14 tools and concluded detectors are "neither accurate nor reliable"; none exceeded 80 percent accuracy across conditions.
Institutions have disabled it: Vanderbilt disabled Turnitin's AI detector in August 2023 over accuracy and fairness concerns. Curtin University announced in September 2025 that it would disable the feature from 1 January 2026, citing reliability and the risk of false accusations.

False positives matter

A flag is a signal, not a verdict.

Weber-Wulff et al. (2023), published in the International Journal for Educational Integrity, tested 14 detectors and concluded they are neither accurate nor reliable. Perkins et al. (2023), in the Journal of Academic Ethics, found Turnitin flagged most GPT-4 test submissions but still missed substantial portions of AI-generated text.

Turnitin also disclosed in 2023 that its sentence-level false-positive rate is around 4 percent, higher than the less-than-1-percent document-level rate it promoted at launch. Because a wrong accusation carries serious consequences, Turnitin suppresses scores from 1 to 19 percent, and institutions including Vanderbilt (August 2023) and Curtin University (from January 2026) have chosen not to rely on the tool. The score should inform a human review, never replace it.

What students should know

Understand the process and keep your evidence.

Keep your drafts, notes, outlines, and version history. A documented writing process is the clearest way to show your work if a score is ever questioned, and it makes any review fairer and faster.

If your work is flagged, you can ask your instructor how the score was used and share your process. A percentage is an estimate that can be discussed, not a final ruling.

Above all, do your own work and follow your institution's academic integrity policy. Understanding how detection works should support honest writing, not attempts to game it.

An optional readability pre-check

Review your own draft before you submit.

Refine's detector is a readability pre-check, not a Turnitin score and not a way to evade one. It estimates AI-likelihood from surface patterns so you can see where your writing reads as generic and revise for clearer, more specific prose. Use it to improve honest work, then follow your institution's academic integrity policy. If phrasing still reads robotic, try Refine's humanizer to smooth drafts while keeping your meaning intact.

Open the full AI detector

Honest limitations

Refine is not Turnitin.

This page summarizes Turnitin's AI writing indicator from public documentation and research. Refine does not integrate with Turnitin, access its models, or predict submission outcomes.

No third-party checker can prove authorship. Treat every AI percentage as one input in a human review process, and follow your institution's academic integrity policy.

FAQs

Turnitin AI detection FAQs

Straight answers on how the AI writing indicator works, how accurate it is, and how it should be used.

Can't find what you're looking for? Contact our customer support team

Ready to refine your draft?

Humanize your AI text in seconds.

Paste your content into Refine and turn stiff, robotic writing into natural copy that sounds like you.

Try Refine now