The AI writing report is a strange document the first time you see one: a single percentage up top, blocks of highlighted text below, and no explanation of how one produced the other. Since most students never get to see theirs at all, here's a guided tour — so when you do have one in hand, you know exactly what you're looking at and what to do about it.
1The overall score — the share of qualifying prose flagged, not of the whole file. Above 0% and below 20% it displays as *% instead.
2Highlighting comes in runs — usually whole passages, not scattered words. This is where to focus your revision.
3Non-prose text — bullet fragments, references, equations — isn't reliably analyzed; the score is about the document's long-form prose.
The percentage: what's actually being counted
The headline number is not "how much of this file is AI." It's the share of qualifying prose that the detector classified as likely AI-generated. That distinction hides two things worth knowing:
- Only long-form prose is reliably analyzed. Bullet fragments, code, equations, references, and other unconventional text aren't dependably scored — though continuous prose sitting inside a table can still be processed. A lab report that's half data tables is judged mostly on its written prose, so the percentage can feel disconnected from the document's size.
- Short documents may not score at all. The detector needs a few hundred words of continuous prose to produce a result. Below that, there's no score — not a zero.
0% vs. the asterisk score
Two low results look similar and mean different things. A 0% means the detector did not identify any qualifying text as likely AI-generated. An asterisk (*%) means the detector registered something, but the result fell above 0% and below roughly 20% — a range Turnitin treats as too uncertain to state precisely, because false-positive risk rises there (the display rules are laid out in Turnitin's report documentation). An asterisk is still a decent outcome — it's the range Turnitin tells instructors to treat with the most caution, and arguing about whether you're a "4%" or an "11%" is meaningless below the threshold — but it is a suppressed low number, not a zero.
The highlighting: where the signal lives
Below the score, the report highlights which passages drove it. This is the genuinely useful part, and the part most people skip. The detector scores segments of text, so highlighting comes in runs — often whole paragraphs — rather than scattered words. When you review your own report:
- Look for clustering. One flagged section in an otherwise clean paper tells you exactly where to revise. Flags spread evenly across the document say more about your overall style than any single passage.
- Notice what kind of text flagged. Summary and transition paragraphs — "In conclusion, this essay has examined…" — flag far more than argumentative or narrative passages, because formulaic prose is precisely what the model is trained to spot.
Turning the report into edits
If a section is flagged and the writing is yours, the fix isn't synonym-swapping — mechanical rewording tends to make things worse. Rewrite the flagged passage the way you'd explain it out loud: vary the sentence rhythm, cut the boilerplate transitions, anchor claims in specifics only you would include. Then re-check the final version, since edits move scores in both directions. And read the AI report alongside the similarity report — they measure completely different things, and confusing the two percentages is the most common report-reading mistake there is.
One last note on access: instructors see this report by default; students almost never do. If you want yours before submission day, a real Turnitin check delivers both PDFs — the AI writing report with its highlighting intact, and the similarity report next to it — about 10 minutes after you upload.