What AI Detector Does Turnitin Use? Their Classifier, Not GPTZero

It is Turnitin’s classifier, not a rented name

Turnitin® built and ships an AI writing indicator as part of the campus product. Instructors see a percentage (or a noisy-star treatment at low scores) for how much the prose looks like language-model output. Their own guide to using the AI writing report is the primary source for what that badge looks like. The company has described it as a model trained on academic writing, not a screenshot of ChatGPT’s servers.

That is the answer to “what AI detector does Turnitin use.” Their stack. Their training. Their threshold. A professor who says “Turnitin caught ChatGPT” is naming the product on the assignment, not naming GPTZero.

If you need the name confusion cleaned up in one place, does Turnitin use GPTZero is the short no. Different companies. Different reports. A screenshot from gptzero.me will not reprint the campus number.

The AI indicator is not the similarity score

Similarity looks for overlap with an index of papers, sites, and journals. The AI indicator looks at style: even sentences, safe verbs, tidy transitions. A ChatGPT essay that was never posted online can sit near 0% similar and still look generated. A quoted methods section can do the reverse.

What the Turnitin similarity score is is the overlap badge. Do not argue with an AI percentage using a plagiarism number. Instructors who know the product already treat them as two columns.

Low AI percentages are noisy. Turnitin has said as much, which is why some views hide a small number behind a star. That is not a loophole. It is a warning that tiny scores are not a confession. The Washington Post’s test of the detector is a public record of that reliability problem.

What the classifier actually sees

Finished prose in the submitted file. Not your ChatGPT chat log. Not your prompt. Not Google Docs version history unless someone exported a trail into the file. The dropbox sees the document you uploaded. Details of that path are in how Turnitin works.

The tells are the same family public checkers use. Language models pick likely next words. That leaves a rhythm. Academic human writing can share that rhythm, especially lab reports and careful second-language English. False positives exist. False negatives exist. A percentage is a signal.

Paraphrase tools do not automatically wipe the signal. Can Turnitin detect ChatGPT if you paraphrase is often yes, because swapping synonyms is not the same as writing like a person.

Who sees the number, and who does not

Settings sit with the institution. Many courses hide the AI indicator from students. You submit. The instructor opens Feedback Studio. You might never see the percentage people argue about on Reddit.

That is why a public self-check exists. DetectGPT scores how AI-like a draft reads: up to 3,000 words, paste or Analyze URL, highlights on the stiffest lines. It will not match Turnitin®. It will show you the paragraphs that still sound like a model before Sunday.

Do not email a DetectGPT screenshot as a substitute for the assignment report. Do use it to find the cluster of even sentences you can actually rewrite.

Mistakes that waste a week

Calling GPTZero “the Turnitin detector.” Pasting 60 words and trusting 98%. Treating a 12% AI badge like a hearing. Humanizing and submitting without putting your examples back in. Assuming similarity at 4% means ChatGPT is invisible.

• Ask which score the instructor is looking at.
• Keep drafts. Process beats a JPEG.
• Read the syllabus before you “fix” a number.
• Use DetectGPT as rehearsal, not as a courtroom exhibit.

A self-check that is not their software

DetectGPT is independent. Not affiliated with Turnitin, LLC. Same family of question: does this draft read like a language model. Different product, no school login.

Paste the section you actually worry about. If the score is high, rewrite the red and orange lines in your voice. Then submit through the real dropbox. The official path is still how Turnitin works.

FAQs

What AI detector does Turnitin use?

Its own AI writing classifier inside the campus product. It is not GPTZero, ZeroGPT, or DetectGPT.

Is the AI indicator the same as similarity?

No. Similarity is overlap with an index. The AI indicator is a style score. They can disagree on the same paper.

Can students see the Turnitin AI score?

Often no. Many schools keep it instructor-only. DetectGPT is a public style check you can run without that login.

Does Turnitin use GPTZero?

No. Different companies and different reports. See the longer note on that mix-up.

Will DetectGPT match the Turnitin number?

No. Classifiers do not share a scale. Use DetectGPT to find robotic lines, not to predict the official badge.

Does paraphrasing hide ChatGPT from Turnitin?

Not reliably. Synonym swaps can leave the even rhythm the classifier looks at.