Guide for students and teachers

What AI detector does Turnitin use?

Turnitin uses its own AI writing detection model, built into its Similarity Report. Turnitin describes it as a transformer-based deep learning model trained on its in-house academic data, and its documentation does not name any outside detector such as GPTZero. It predicts which sentences were likely written by AI tools such as ChatGPT and, for English, which were likely AI-generated and then changed by paraphrasing or “bypasser” tools.

Updated · Plagiarism Checker Plus editorial team

Need to check a text? Use our free plagiarism checker or AI detector.

The short answer

Turnitin’s AI checker is its own. Turnitin describes its own model and names no outside detector, and the result appears as an AI writing indicator inside the Similarity Report that instructors already use.

  • Who built it: Turnitin. Its FAQ says submissions are evaluated “via our proprietary algorithms” for both similarity and AI writing.
  • What it detects: text likely generated by large language models, and in English, AI text likely modified by paraphrasing or bypasser tools.
  • When it runs: only if your institution licenses and enables the feature, on files with at least 300 words of prose in a supported language.
  • Who sees it: instructors and administrators, not students.
Everything this guide says about Turnitin comes from Turnitin’s own help centre, read on 17 November 2026. The Sources section links each page. Turnitin updates these pages often, and your school decides whether the feature is on.

What the model is

Turnitin’s AI writing detection FAQ has a question titled “Which model is Turnitin’s AI detection based on?”. Its answer, in short:

  • A transformer-based deep learning model. Turnitin says it chose this approach because transformer models perform well across many language tasks.
  • Not off the shelf. Turnitin says it does not rely solely on a standard transformer architecture and keeps improving it with work from its own research lab and the wider research community.
  • Trained on Turnitin’s data. The models are trained on Turnitin’s “vast inhouse corpus of academic data”. Elsewhere the FAQ says the training sample includes both AI-generated and authentic academic writing across regions and subjects, including second-language learners.
  • No single signal. Turnitin says the model is not programmed to check specific measures such as “burstiness” or “perplexity”. It learns patterns from training data, so a single prediction may not be explainable feature by feature.

How it scores a paper

Turnitin says the model gives each sentence a score between 0 and 1, judged in context with the text around it. Sentences above a threshold are labelled AI, and the document percentage is the share of evaluated sentences above that threshold. In October 2026 Turnitin updated the English model’s architecture to improve accuracy where text changes between human and likely AI writing. Our guide on how AI detectors work explains this kind of classifier in general terms.

What it detects

Turnitin describes two things the indicator looks for:

  1. AI-generated text. Text likely produced by AI writing tools such as ChatGPT. The FAQ lists the large language models its English detector covers, including GPT-4o and GPT-5 versions, Gemini, Claude, Llama, DeepSeek, Grok and Mistral models. Turnitin says it keeps adding models as new ones are released.
  2. AI text changed by paraphrasers or bypassers. Turnitin says the detector also looks for likely AI-generated text that was then modified by an AI paraphrasing tool, a word spinner or a bypasser (often called a humanizer). This runs automatically wherever AI detection is enabled. Turnitin does not publish which tools it detects, and this part is available for English submissions only.

Turnitin also says what the model does not target. Grammar, spelling and punctuation fixes from tools such as Grammarly were, in most of its tests, not flagged. Content made with Grammarly’s generative features, such as drafting or paraphrasing, will likely be flagged.

Only prose sentences count. Lists, headers, references, data tables, formulas and code are removed before scoring, so the percentage is a share of the qualifying prose, not of the whole file.

Languages, word count and file types

These requirements are from Turnitin’s FAQ for English submissions. If a file misses any of them, no AI percentage is produced.

RequirementWhat Turnitin states
LanguagesLong-form English, Spanish, Japanese, Modern Arabic, Dutch, German and Turkish
Minimum lengthAt least 300 words of prose text in a long-form writing format (depends on the language)
Maximum lengthNo more than 30,000 words
File types.docx, .pdf, .txt, .rtf
File sizeLess than 100 MB

A paper in an unsupported language is not processed and gets no AI report. The Similarity Report still runs as usual.

Where it runs and who sees it

  • Licensing. AI writing detection is available to institutions that license Turnitin Originality, or as an add-on for iThenticate 2.0. Administrators can turn it off.
  • Inside an LMS. In Canvas, Moodle, Blackboard and similar systems, there is no AI score in the LMS screens. Instructors open the Turnitin report to see it.
  • Visibility. The AI indicator and report are not visible to students. Instructors can download the report as a PDF and share it.
  • Separate from similarity. Turnitin says the similarity score and the AI percentage are completely independent. See our guide to the Turnitin similarity score.

Turnitin also offers Authorship, which it says is a different technology. Authorship uses metadata and forensic language analysis to check whether someone other than the student wrote the work. It does not tell you whether text was AI-written.

How it differs from other AI checkers

Because Turnitin uses its own model, a score from another detector does not tell you what Turnitin will show. GPTZero, Copyleaks, ZeroGPT and other checkers are separate products from other companies. Turnitin’s documentation does not say it uses any of them. Two detectors can give the same text very different scores.

That applies to our tool too. The Plagiarism Checker Plus AI detector is a different tool from the one your school uses. It can be a second opinion: a likelihood score with a sentence-level view of which passages read as AI-like. It does not predict or reproduce a Turnitin result.

How much to trust Turnitin’s own score is a separate question. Our guide is Turnitin’s AI detector accurate? sets out Turnitin’s published false positive figures and their conditions, and the concerns universities have raised. For which schools use detectors at all, see do colleges use AI detectors?

If your work is flagged

Turnitin says it does not decide whether misconduct happened. Its guidance for instructors starts with the institution’s policy and treats the score as the start of a conversation. The final decision rests with the instructor. If you are the student:

  1. Keep your drafts and version history. Google Docs or Word history, notes and outlines show how the work developed.
  2. Talk to your instructor. Ask to see the report, and be ready to explain your argument and sources.
  3. Follow your school’s process. Your academic integrity policy sets out the steps and how to appeal.

An AI score is a likelihood, not proof. For what different scores tend to mean in practice, see what percentage of AI is acceptable? and AI detector false positives. For the full walkthrough of Turnitin’s score states, see does Turnitin detect AI?

Sources

FAQ

Questions

Turnitin’s documentation does not say it uses GPTZero or any other outside detector. It describes its own transformer-based model, trained on Turnitin’s in-house academic data and improved by its own research lab. GPTZero is a separate product from a different company.

Check a draft with a free AI detector

Paste up to 1,000 words for a free check, no account needed. It is a different tool from the one your school uses, so treat the result as a second opinion and a likelihood, not a preview of your Turnitin report.