Does Turnitin Detect ChatGPT? What Turnitin Actually Says

Turnitin's press release announcing its AI writing detector states it "identifies the use of AI writing tools, including ChatGPT," and says development on the underlying detection started roughly two years before ChatGPT's public release, working from GPT-3, with testing extended to GPT-4 by launch. That's a more direct claim than you'll find from most detectors in this category, which tend to describe catching 'AI-generated text' without naming a specific tool.

Turnitin's own FAQ broadens that scope further, stating the detector is built to catch AI-generated content from a range of underlying models — GPT-4o through the GPT-5 family, Gemini, Claude, LLaMA, and Mistral among them — not ChatGPT alone.

Here's the part that trips people up: Turnitin's own FAQ is explicit that the detector "cannot distinguish between specific AI tools." It can flag a passage as showing patterns consistent with AI generation broadly, but it doesn't attribute a report to ChatGPT versus Gemini versus Claude versus any other model — there's no tool-specific label anywhere on the report itself.

Mechanically, the detector segments a submission into overlapping sections of text and scores each on how probable the word sequence would be for a language model to generate, based on the idea that AI models tend to pick a highly probable next word in a consistent pattern, where human writing is comparatively more varied. It only evaluates long-form prose — bullet points, lists, and code are excluded from scoring.

Turnitin states its false-positive rate is under 1% — roughly one in a hundred — for documents where more than 20% of the text is AI-generated. Scores in the 1–19% range are deliberately left unlabeled on the report rather than flagged, which Turnitin describes as a conservative choice: it would rather under-flag some real AI use than risk a false accusation on a borderline score. Turnitin also says it re-tests each model update against a bank of more than 700,000 pre-ChatGPT academic papers to check the false-positive rate holds on genuinely human writing.

One caveat worth being upfront about: that false-positive figure is Turnitin's own published number, not the result of an outside audit. Turnitin is also explicit that the tool itself doesn't determine misconduct — it says its role is to give instructors data to make an informed decision, not to render a verdict.