Turnitin says it can, for English text and the Claude versions on its list. The list, updated 2026-11-03, runs from Claude 3 Haiku to Claude Opus 4.6. What Turnitin reports is how much of a document is likely AI-written, not that Claude wrote it. Everything below is from Turnitin's own pages, and its accuracy statements are vendor claims.
What Turnitin states about Claude
- Claude 3 Haiku, Sonnet 3.5, Sonnet 3.7, Sonnet 4.5, Haiku 4.5, Sonnet 4.6 and Opus 4.6 are on its English model list.
- It says it is "continuously updating" the model to detect more LLMs. The list shows no Claude version after Opus 4.6, while the current Claude versions run well beyond it.
- The report does not name the model behind a passage.
- Turnitin says it can also flag text modified by AI paraphrasers and bypassers, but only for English submissions.
How the score works
Each sentence gets a score between 0 and 1. Sentences over a threshold count as AI, and the document score is the percentage of qualifying text above it. In documents of a few hundred words Turnitin says the prediction is mostly all or nothing. Scores from 1% to 19% are not shown, because Turnitin says false positives are more likely there.
Limits Turnitin states
- English files need at least 300 words of prose. Lists, headers, references, tables, formulas and code are skipped, and Turnitin says it does not reliably detect non-prose or code.
- Its goal is a false positive rate under 1% for documents with over 20% AI writing. This is Turnitin's own target, not independently verified.
- It says it can miss AI text: a document it scores at 50% could contain as much as 65% AI writing.
- Text with little structural variation, text that repeats itself and text paraphrased without new ideas are more likely to be flagged wrongly.
- It "does not make a determination of misconduct".
- Access is only through institutional licences, with no individual free tier and no public paste box.
What an independent second check adds
A Turnitin score is one signal inside a school's own system, and only instructors and administrators see it. A second check from a different engine helps in three ways. It is available to the writer as well as the marker. It can cover text under the minimum length that Turnitin applies. And it can show writing-style similarity to Claude and other families, which Turnitin's report does not.
Paste the text into ClaudeDetect for a quick second reading from Veread, Fileverity's text model. Where two checks disagree, treat the case as open and look at the drafts, sources and the writer's earlier work. In published tests across seven languages Veread wrongly flagged 0.5% of human-written text; see the facts page for how that was measured.
Sources: Turnitin AI writing detection capabilities FAQ, AI writing detection model release notes, how to access the AI writing report.