How Turnitin detects AI writing

Updated June 10, 2026

Turnitin checks more student writing than any system on earth, and its AI score is the one you can't see before submission. Here's how it works, what Turnitin itself admits about accuracy, and how to be on the right side of it.

Where the check happens

Turnitin's AI detection launched in April 2023 inside the Similarity Report — the same interface instructors already used for plagiarism. When your assignment goes through the LMS portal, the instructor sees an estimated percentage of AI-generated prose alongside source matches. You, the author, see nothing. That asymmetry shapes everything about how to handle it.

The mechanics

Turnitin segments your document into chunks of running prose and scores each with a classifier trained to distinguish model output from student writing — the standard signals: token predictability, structural evenness, stock phrasing. It needs a few hundred words of qualifying prose to score at all, ignores non-prose elements, and reports the AI percentage at document level with flagged segments.

Turnitin has also said it targets some AI-paraphrased text specifically — synonym-swapped model output keeps model structure, which is the thing being measured.

What Turnitin admits

Turnitin publishes that its false-positive rate is low but real, that short documents and certain genres are less reliable, and — crucially — that scores should prompt conversations, not automatic accusations. Multiple universities have publicized cases of honest work flagged; some institutions disabled the feature over it. If you're flagged on your own writing, that institutional ambivalence is your context: see the false-positive playbook.

The student playbook

  • Know your course policy precisely — allowed-with-disclosure, brainstorm-only, or banned changes everything downstream.
  • Keep your process: drafts, outlines, version history. The strongest counter to any score is evidence of work.
  • Where AI help is allowed, humanize the stiff drafting and verify your score on your side before submitting — it's the only pre-check you'll get.
  • Write with your own rhythm — formal doesn't have to mean uniform, and uniform is what gets flagged.

Frequently asked questions

What AI percentage gets you in trouble with Turnitin?

There's no universal threshold — institutions set their own responses, and many treat any score as conversation-starter rather than evidence. The flagged-segments view matters more than the single number.

Can Turnitin see ChatGPT history or know which AI I used?

No — it analyzes only your submitted text. It has no access to your accounts, and it doesn't identify which model produced text.

Does Turnitin AI detection run on everything submitted?

Where institutions license and enable it, generally yes. Assume anything submitted through a portal may be scored.

Humanize it — then verify it

Paste your text, get a rewrite that reads like a person wrote it, and check the AI-probability score yourself before anyone else does. 3-day free trial.