What happens when Turnitin flags AI in your work
The report goes to your institution, not to you. Here is what is on it, what it is based on, and what part of the process you can still influence.
The sequence is the same almost everywhere. You submit, the AI indicator runs automatically, a percentage appears on a report visible to staff, and you find out only if somebody decides to raise it with you.
What follows from that is set by your institution's own procedures, which vary a great deal and which we are not the right people to describe. What we can describe is the number itself, since that part is the same everywhere.
What is on the report
Two separate scores that look alike and work nothing alike.
Similarity matches your text against a database and shows you the source. It is a lookup.
The AI indicator matches nothing. It is a classifier's probability that your writing has properties characteristic of generated text. There is no source, because nothing was found.
They sit side by side, in the same visual style, on the same page. That design does a lot of work in how people read them. See how Turnitin AI detection works.
What the percentage is a percentage of
The share of qualifying prose the classifier scored as AI written. Not a confidence level, and not a share of your ideas.
Two figures published by Turnitin change how it should be read:
Wrong on human writing 4% of the time. At department scale that is hundreds of documents a term where the indicator is wrong about somebody's own writing.
Below 20% is thinner still. Turnitin states that documents scoring under 20% show a higher incidence of wrong flags, because a small flagged region is the least evidence a report can rest on.
Why the flagged region is usually too wide
Turnitin's own analysis found 54% of falsely flagged sentences sit directly beside a sentence scored as AI written, and 26% two sentences away.
Four in five wrong flags land next to a genuine one. The practical effect is that a document with one assisted paragraph does not produce one flagged paragraph, it produces a spread across the human writing around it. The boundaries of a highlighted region are soft. See documents that are part human, part AI.
The part you can influence
Not the report. The writing that produces it, and only before you submit.
That is the whole asymmetry of this situation: after submission you are discussing a number somebody else generated, and before submission you can simply look at what your text does to a classifier and change the paragraphs that read flat.
The changes are mechanical and do not touch your argument. Vary sentence length. Cut connective openings. Keep the specific detail. See what makes text read as machine written and check your essay before you submit it.
Ours is free and unlimited with no account, takes .pdf and .docx, scores every paragraph separately so you can see exactly which ones would draw a flag, and never stores your text. Our wrong-flag rate is 0.4%, published, measured on 15,900 documents of real academic writing.