Turnitin gives every submission two different percentages, and confusing them is the single most common mistake students make about the whole system — it leads people to relax when they shouldn't and panic when they needn't. Two minutes here saves both errors.
The similarity score: a measurement
The similarity score answers one mechanical question: how much of this text matches existing sources? — journal articles, websites, and other students' stored papers. It's a lookup against databases, it comes with an itemized match list, and it carries no judgment: quotes, bibliographies, and templates all match innocently. That's why a "bad" similarity score is entirely about what the matches are, not the number.
The AI score: a prediction
The AI writing score answers a completely different question: how much of this prose looks machine-generated? There's no database and no match list — it's a trained classifier reading statistical patterns, which is why it can be wrong in both directions, why low results display as an asterisk rather than a number, and why "acceptable" works differently for it. Instructors see it in the same report viewer, but students almost never see it at all.
The four combinations, decoded
- Low similarity, low AI. The normal result for original work. Submit in peace.
- High similarity, low AI. A citation problem, not an AI problem: quotes, paraphrase too close to sources, boilerplate — or a self-match against your own earlier work. Fixable by auditing the match list.
- Low similarity, high AI. The one that fools people. Generated text matches nothing — it's new — so a pristine 2% similarity alongside a confident AI score isn't an alibi; it's the classic signature of a ChatGPT essay. "But my plagiarism score was zero" convinces no one.
- High similarity, high AI. Usually AI text that was paraphrased from sources, or heavy tool-rewriting on top of copied material. Both systems firing at once invites the closest reading of all.
One document, two lenses
The practical takeaway: clearing one score tells you nothing about the other, because they share no machinery. Any pre-submission check worth doing has to show you both — a real Turnitin check returns the two reports as separate PDFs, similarity with its match list and the AI report with its highlighted passages, so you can audit each on its own terms before anyone else does.