Turnitin AI Detector: How It Works in 2026
You cannot run Turnitin yourself — only your institution can. What you can do is understand what its AI report measures and check your draft with a free detector first. Humanit's is above; how Turnitin works is below.
Turnitin's AI writing detector estimates what share of a submission was likely written by AI and shows it to instructors, not students, by default. Since August 2025 it also flags text it thinks was run through an AI bypasser tool. Turnitin says the score should not be the sole basis for action against a student.
Works with every major AI model
How it works
- 1Paste your text. Drop in the writing you want to check against Turnitin-style detection.
- 2Run the free detector. Humanit returns a 0–100 AI-likelihood score, a verdict, and the signals behind it.
- 3See what reads as AI. The subscores show which patterns — predictable phrasing, uniform rhythm, stock vocabulary — are pulling the score up.
- 4Fix it if needed. If it reads as AI, send it to the humanizer to rewrite those signals, then re-check.
What the Turnitin AI percentage counts
Turnitin's number is a share of the submission: the proportion of qualifying text its model judges likely to be AI-generated. That is a different convention from GPTZero, whose headline figure is a probability for the whole document, and it is why the same essay can read 40% in one tool and something else entirely in another. Scores from 1 to 19% are not shown as a number at all — Turnitin displays an asterisk instead, because low scores produced too many false positives. The report sits alongside, but separate from, the similarity (plagiarism) score most students already know.
Turnitin now flags AI bypasser tools
The change most students have not heard about: on August 27, 2025, Turnitin released an update that detects the likely use of AI bypasser tools — software that rewrites AI text to look human, which is exactly what humanizers do. Its AI-generated category now includes the share of text that may have been modified by such a tool. Reports generated before the update were not changed; a submission has to be resubmitted to be scored by it. The honest implication for anyone using a humanizer, this one included: rewriting to dodge Turnitin is a moving target, and Turnitin is now specifically looking for it.
Languages and recent model updates
Turnitin's AI writing detection covers English and has expanded to other languages: detection for Japanese submissions arrived on April 2, 2025, and the Spanish-language model was updated on May 5, 2026 to improve detection while, in Turnitin's words, minimizing false positives. Its release notes also describe repeated updates to improve recall while keeping a low false-positive rate. Each update changes how the same text might score, which is one more reason a score from months ago says little about a score today.
What students can and cannot see
Students cannot run Turnitin's AI detector themselves, and the AI writing report is not visible to them by default — the instructor sees it. If you have been told a Turnitin report flagged your work, ask in writing for the PDF of the AI writing report before responding. Turnitin's own documentation says the company does not make a determination of misconduct and that the percentage should not be used as the sole basis for adverse action. The step-by-step response is in flagged by Turnitin for AI?.
How accurate is Turnitin's AI detector?
Turnitin publishes model updates aimed at catching more AI text while keeping false positives low, and it hides low scores precisely because they were unreliable. Independent research on AI detectors as a class is sobering: a 2023 peer-reviewed study in Patterns found more than 61% of genuine TOEFL essays by non-native English speakers misclassified as AI by the detectors it tested. The fuller look at Turnitin's error rates is in how accurate is Turnitin?. The practical reading: a high score on long, unedited AI text is meaningful; an asterisk or a low score is designed not to be acted on.
Checking your draft before you submit
Because you cannot preview Turnitin, the useful move is to check your draft with a detector that shows its reasoning. Humanit's detector above is free with no account for up to 300 words per check and breaks its 0–100 score into sub-signals — perplexity, burstiness, AI vocabulary, structure — with the exact phrases pulling the score up. It is a different model from Turnitin, so the numbers will not match; what transfers is which passages read as machine-written. Keep your drafts and version history either way. They settle more academic-integrity cases than any score.
Who it's for
| Feature | Turnitin | Humanit detector |
|---|---|---|
| Who can run it | Institutions only | Anyone, free, no account up to 300 words |
| What the number means | Share of the submission likely AI-written | 0–100 likelihood for the whole text |
| Low scores | 1–19% shown as an asterisk | Shown as a score with sub-signals |
| Flags AI bypasser / humanizer use | Yes, since August 2025 | No |
| Languages | English, Spanish, Japanese | English-tuned |
| Who sees the result | The instructor, by default | You |
Frequently asked questions
Can I check my paper with Turnitin myself?
No. Turnitin sells to institutions, and its AI writing report is shown to instructors by default. You can check a draft with other detectors — Humanit's is free with no account for up to 300 words — but no tool can show you Turnitin's own score.
Can Turnitin detect AI humanizers?
Turnitin says so: since August 27, 2025 its AI detection includes the likely use of AI bypasser tools, and its AI-generated category includes the share of text that may have been modified by one. No humanizer can promise to get past it.
What does the asterisk mean on a Turnitin AI report?
Scores from 1 to 19% are shown as an asterisk rather than a number, because Turnitin found low scores unreliable. An asterisk is not a finding of AI use.
Is Turnitin's AI detector accurate?
Turnitin updates its model to catch more AI text while keeping false positives low, and says its score should not be the sole basis for action against a student. Independent studies of detectors in general find real false positives, especially for non-native English writers.
Does Turnitin detect AI in other languages?
Yes, beyond English: Japanese detection launched on April 2, 2025, and the Spanish-language model was updated on May 5, 2026.
Does Turnitin re-check old submissions with new models?
No. Turnitin says updates do not retroactively change reports already generated; a submission has to be resubmitted to be scored by a newer model.
Will my text be used to train AI?
No. Your text is sent to the model only to produce your result and is never used to train any AI. No content is stored beyond the request.
Does it keep my original meaning?
Yes. Humanit preserves your facts, numbers, names, and citations while changing the phrasing and structure. It never invents new claims.