Does Turnitin detect ChatGPT?

Does Turnitin detect ChatGPT? The short answer

If you're asking "does Turnitin detect ChatGPT", the short answer is yes, as long as your school has the feature switched on. Turnitin has checked submissions for AI writing since April 2023. Its current documentation says it also looks for text from Claude, Gemini and other models, and for AI text run through paraphrasers and "humanizer" tools. The result is a percentage your instructor sees. You don't.

Detecting isn't the same as catching every time. Turnitin tunes its model to avoid accusing people who wrote their own work, so some AI text gets through, and it says the score should never be the sole basis for action against a student. Everything below comes from Turnitin's own documents as of October 2026, with dates where things changed. Our free AI detector shows which sentences read like AI to a different model, so its score won't match Turnitin's.

Does Turnitin have an AI detector? Yes, since April 2023

Turnitin announced its AI writing detector in February 2023, saying it caught 97% of ChatGPT and GPT-3 writing in its lab. On April 4, 2023 it switched detection on inside its existing products, for more than 2.1 million educators at over 10,700 institutions, with no extra steps for teachers. Its first target was GPT-3 and GPT-3.5, the models behind ChatGPT at the time.

Access is narrower today. In 2023 the detector came with products such as Turnitin Feedback Studio and Turnitin Similarity. Turnitin's AI writing FAQ now says it's only available to institutions that license Turnitin Originality, or iThenticate 2.0 with an AI add-on. Administrators can also switch it off, and if a school hasn't enabled it, Turnitin doesn't process submissions for AI at all.

So whether Turnitin checks for AI depends on your school, not just on Turnitin. Some universities have turned the AI detector off while keeping the similarity check; which AI detectors colleges use lists who uses what.

What AI detector does Turnitin use?

Turnitin uses its own model. Its FAQ describes a deep learning system based on the transformer architecture, trained on Turnitin's in-house collection of academic writing alongside AI-generated text. The model learns statistical patterns from examples rather than following fixed rules. Turnitin says it isn't explicitly programmed to measure "burstiness" or "perplexity", and that its individual predictions can't always be explained feature by feature.

The mechanics are simple to describe. Turnitin splits a paper into overlapping segments of sentences and gives each segment a score between 0 and 1. Every sentence takes the scores of the segments it sits in, and the sentence scores are combined into the document's percentage. In July 2026 Turnitin merged what had been several models into a single one.

Can Turnitin detect ChatGPT, Claude and Gemini? The models it lists

Turnitin's FAQ lists the models its English detector covers. As of October 2026 the list has 32 models, "and tools based on these LLMs as well":

So Turnitin says it detects Claude and Gemini as well as the GPT models behind ChatGPT. For a model that isn't on the list, it makes no specific claim. Other languages have shorter lists. According to Turnitin's release notes, the Spanish model was trained to detect GPT-3.5 and GPT-4 text when it launched in September 2024, and the Japanese model, launched in April 2025, to detect GPT-4, GPT-4o and GPT-4o-mini.

Listed doesn't mean always caught. In a study published in August 2026, Turnitin scored under its 20% reporting line on 26% of journal introductions written by GPT-4o and 47% of those written by GPT-5, both models on its list. Turnitin has updated its model since the paper was submitted in November 2025. Is Turnitin's AI detector accurate covers that study and others.

Does Turnitin detect AI-paraphrased or humanized text?

Turnitin says yes, in English. Its release notes show that coverage arriving in three steps:

• December 2023: detection of likely AI text "even if it may have been paraphrased using an AI word spinner."
• July 2024: the report began splitting its score into two categories, "AI-generated only" and "AI-generated text that was AI-paraphrased", naming QuillBot as an example of a paraphrasing tool.
• August 2025: detection of AI bypasser tools, which Turnitin describes as tools that "attempt to modify AI-generated text to appear more human-like." Text it thinks went through one counts toward the "AI-generated only" category.

The FAQ says bypassers are "also called humanizers" and that the detector was trained and tested on "leading paraphraser and bypasser tools". It won't name them, so nobody outside Turnitin knows exactly which tools it catches or how often.

Spelling and grammar fixes are a different case. Turnitin says changes from grammar checkers were mostly not flagged when it tested fully human documents. Text produced by those tools' generative features, such as drafting, paraphrasing or summarizing, likely will be.

Who sees the Turnitin AI score? Instructors, not students

Turnitin's FAQ is direct: "The AI writing detection indicator and report are not visible to students." Only instructors and administrators see the percentage and the highlighted report. An instructor can download the report as a PDF and share it with you, so if a score comes up, ask for that copy.

You won't find the score on your course page either. When Turnitin runs inside a learning management system such as Canvas, Moodle or Blackboard, the AI score isn't shown in the LMS itself; instructors open the Turnitin report to see it. Does Canvas have AI detection explains what students see there.

Two more details from the FAQ. The AI percentage is completely separate from the similarity score, which measures text matching other sources, and neither one affects the other. And schools using the Microsoft Teams Assignment Similarity integration don't get AI detection, because that integration has only a student view.

What Turnitin can't check

Turnitin's AI writing report guide and FAQ set clear limits. A submission gets no AI score, just a gray "--", when it:

• has fewer than 300 words of prose, or more than 30,000 words
• isn't a .docx, .pdf, .txt or .rtf file, or is 100 MB or larger
• is written in a language other than English, Spanish, Japanese or Arabic (and paraphrase and bypasser detection works only in English)
• was submitted before the school enabled AI detection, since old papers are scored only if they're resubmitted

Some writing is processed but not really judged. The score covers only prose sentences in long-form writing, so bullet points, tables, poetry, scripts, code and annotated bibliographies aren't reliably assessed, and reference lists are left out. Turnitin hasn't added AI detection to Gradescope and says it isn't pursuing ChatGPT code detection.

Short papers are the roughest case. In a document of only a few hundred words, the FAQ says the prediction is mostly "all or nothing", so a paper that mixes your own writing with AI text can be flagged as entirely AI.

Using AI under your school's rules

Turnitin's score estimates how much of a paper reads like AI. It can't tell whether your instructor allowed AI for brainstorming, outlining or editing, and Turnitin says it doesn't decide misconduct: it "provides data for the educators to make an informed decision." Read your course policy, and if you used AI in a way it permits, say how.

Running AI text through a paraphraser or humanizer doesn't change what your course allows. Turnitin looks for exactly that, and no tool can honestly promise to get past it.

Students can't run Turnitin on their own drafts. A free checker such as ours uses a different model, so its score won't match Turnitin's, but it can show which sentences read like AI before you hand anything in. Our page on pre-checking a draft without Turnitin explains how.