Does Turnitin detect ChatGPT?
It reports a percentage, and the percentage is wrong about human writing 4% of the time. Both halves of that matter if you are the one submitting.
Yes, Turnitin runs an AI indicator, it runs on almost every submission that passes through it, and you do not get to see the result. Your institution does.
The part worth your attention is not whether it detects. It is how often it is wrong, and in which direction.
The number Turnitin publishes
4%. That is the rate at which genuine human writing gets flagged.
Ours is 0.4%, which is 99.6% specificity, measured across 15,900 documents of real academic writing.
Ten times the wrong flags, on the error that arrives with your name attached to it.
What it is actually looking at
Not a database. The AI indicator does not match your essay against anything, which is the difference between it and the similarity score sitting beside it on the same report. Similarity points at a source you can open. The AI indicator points at nothing.
What it reads is regularity. Predictable word choice, and sentences that stay close to the same length. Generated text has both. So does a lot of careful human writing, which is the whole problem. See why AI detectors flag human writing.
Which of your writing is most exposed
Formal register. The more correctly you write, the more conventional your phrasing, and conventional is what the classifier reads as machine-like.
Even sentence rhythm. Human writing in our corpus varies its sentence length with a standard deviation of 8.7 words. Generated text sits at 6.4. Write every sentence to a similar length and you land in generated territory without a model involved.
Heavy editing. Every revision smooths, and smoothing removes variation. A fifth draft scores higher than a first.
English as a second language. Repeatedly measured as an elevated wrong-flag population. See why non-native English writing gets flagged.
Where wrong flags land
Turnitin's own analysis found 54% of falsely flagged sentences sit directly beside a sentence scored as AI written, and another 26% two sentences away.
Four in five wrong flags cluster around a genuine one. Which means a document with a single assisted paragraph does not show one flagged paragraph. It shows a smear across the human writing around it. See documents that are part human, part AI.
What you can do about it
You cannot see the report before it is generated. You can see what your writing looks like to a classifier before you submit, which is the only part of this you control.
Run the document, look at which paragraphs come back flat, and fix those. It takes a few minutes, and the changes are mechanical: vary your sentence lengths, cut connective openings, keep the specific detail you were about to generalise. Full list in what makes text read as machine written.
Ours is free and unlimited with no account. It takes .pdf, .docx and .tex, has no word cap, scores every paragraph separately, and never stores your text.