It tries to — and it's unreliable in both directions. Turnitin's AI writing indicator estimates how much of a document reads as machine-generated. It does not identify ChatGPT specifically, and it produces both false positives and false negatives.
So the honest answer to “will it catch me?” is that nobody can promise you either way — which cuts against students who cheated and students who didn't.
What Turnitin is actually measuring
There is no signature in ChatGPT's output. No watermark, no hidden token, nothing that says “a model wrote this.” So Turnitin isn't recognising ChatGPT the way its similarity checker recognises a copied sentence — that one genuinely matches your text against a corpus of real documents.
The AI indicator does something statistical instead. It scores how predictable your writing is: how closely each word follows the word a language model would have expected next. Text that is highly predictable gets scored as machine-like. That's the entire mechanism, and it explains every strength and every failure of the tool.
Because it measures predictability rather than origin, it cannot distinguish “a machine wrote this” from “a careful human wrote this in plain, correct, well-structured prose.” Those look the same to the maths.
How wrong does it get?
Wrong enough that its own output shouldn't decide anything. A Stanford study found detectors wrongly flagged roughly 61% of essays by non-native English speakers. In 2026 the Office of the Independent Adjudicator upheld complaints from students — including non-native speakers and an autistic student — accused on the strength of a detector score and later cleared. HEPI's position that summer was blunt:
A detector score should never be sufficient evidence of misconduct on its own.HEPI, August 2026
It errs the other way too. A tool that reports both false positives and false negatives isn't measuring what its name implies, and treating any single percentage as a verdict — in either direction — is a mistake.
Who actually sees the score
Usually your marker, and usually not you. The AI indicator sits alongside the similarity report on the staff side of Turnitin. Depending on how your module is configured you may see a similarity figure, see it only after the deadline, or see nothing at all; the AI percentage is frequently withheld from students entirely.
Which produces the situation most students find themselves in: the first you hear about a flag is an email about a meeting, long after the last point at which you could have changed a sentence.
What to do with that
Two things are within your control. Know your course's actual rule on AI use — it's on the assignment brief, and it varies between modules, let alone universities. And keep your working: draft in one continuously-edited document so its version history shows the essay being built. If you're ever asked how you wrote something, that answers it faster than any argument about percentages.
If you've already been flagged, start with what to do when you're wrongly accused. For the mechanics in more depth, we've written up how Turnitin's AI detection works.
SafeGrade shows you which passages read as machine-like and why, before you submit. Analysed privately on our own infrastructure — never stored, never used for training.
Run one free scan →No card. No catch. First scan free.