Can Universities Detect ChatGPT? What Professors Actually Notice

The short answer

A university cannot prove that you used ChatGPT with software alone: no AI detector can establish who wrote a text, it only estimates how machine-like the writing looks. Universities still notice AI use, usually through other things: sources that do not exist, a style that does not match your earlier work, and a conversation in which you cannot explain your own argument. Whether using AI is a problem at all depends on what your university allows and whether you disclosed it.

Can software prove you used ChatGPT?

No, not as proof. AI detectors, including Turnitin’s AI indicator and our own check, measure style: how evenly the sentences are built, how predictable the word choice is, how often stock phrases appear. From that they compute an estimate. They do not know how a document was written.

That cuts both ways. Writing you did yourself is sometimes flagged, especially very formal writing or writing in a second language. Edited machine text is sometimes not flagged. Vanderbilt University switched off Turnitin’s AI detection in 2023 for this reason: even a small error rate, applied to tens of thousands of papers, produces hundreds of false suspicions a year.

How often our own check wrongly flags human writing, measured on real theses written before ChatGPT existed, is on our accuracy page.

How do professors detect ChatGPT, then?

Supervisors read a lot of student work and often know yours. The signs they act on rarely come from software:

• Invented or miscited sources. ChatGPT sometimes invents references that do not exist, or attributes claims to a real source that it never makes. This can be checked, because every source can be looked up. Our procedure for verifying AI citations shows how to check yours.
• A voice that is not yours. A sudden change of tone between chapters, or a thesis that sounds nothing like your essays and emails.
• General statements with nothing specific. Paragraphs that read smoothly but say nothing about your material, your data or your research question.
• The conversation. In a viva, an oral defence or a meeting with your supervisor it becomes clear quickly whether you know and can defend the arguments in your work.

What counts as evidence?

Your university’s academic integrity rules decide that, not a detector. A high AI score is a reason for an examiner to look more closely and talk to you. It is not a verdict.

For you, the strongest support is the trail your work leaves: notes, outlines, earlier drafts, annotated PDFs and the version history in Word or Google Docs, which show a document growing over weeks. What convinces most is in How to prove you wrote it yourself.

What is actually allowed?

Many universities do not ban AI outright. Spell-checking, translation help, brainstorming or feedback on an outline are often allowed, sometimes only if declared. Submitting text an AI wrote as your own work is forbidden almost everywhere. The rules differ between universities and even between departments, so read your module handbook and ask your supervisor before you submit, not after. Where the line usually falls is in Is using ChatGPT cheating?

How do I disclose AI use properly?

If you used AI in a permitted way, saying so openly is the safe route. A short AI disclosure statement names the tool, what you used it for and where in the thesis. Three templates you can adapt are in AI disclosure statement for a thesis.

Can I check my own work before I submit?

Yes. Turnitin usually shows its AI score to the university only. Our AI check shows you before submission what an examiner would see after it: a colour from green to amber to red, the marked sentences, and a published error rate. The first 1,500 words of an uploaded document are free, with no account and no name.

A green result does not prove you wrote everything yourself, and a red one does not prove the opposite. It shows you which passages might catch an examiner’s eye, so you can look at them again with your notes and sources. If your work has already been flagged, read Accused of using AI when you didn’t.