Direct answer
Checking a piece of writing for AI text and for plagiarism means running two separate tests, because they measure different things. A plagiarism or similarity check compares your words against published sources and previously submitted papers; an AI-writing check estimates how much of the text a detection model classifies as machine-generated. Neither result is a verdict on its own — both are indicators that a human reader, usually your instructor, interprets in context [1]. Students who check before submitting get something more useful than reassurance: a list of specific passages they can still revise.
What Is The Difference Between An AI Writing Check And A Plagiarism Check?
The two checks answer different questions, and conflating them is the single most common source of panic around Turnitin results. A similarity check asks "does this text match other text that already exists?" while an AI-writing check asks "does this text look like it was produced by a language model?" A document can score 0% similarity and still be flagged for AI writing, and a document can score 34% similarity with no AI flag at all [2].
Similarity is computed by matching strings of your text against a comparison database of web pages, journal articles, and student papers, then reporting the proportion of your document that matched something. Turnitin's own guidance is explicit that the resulting percentage is a similarity index, not a plagiarism percentage — the tool reports matches, and a human decides whether a match is improper [2]. Direct quotations with correct attribution, a bibliography, a standard methods sentence, and a module title all produce matches that are entirely legitimate.
AI detection works on statistical properties of the prose rather than on matching. The model evaluates sentence-level patterns and reports an estimated share of the document it considers AI-generated, together with a confidence signal [1]. Because it is a classification estimate rather than a lookup, it behaves differently from similarity: it can move several percentage points when you rewrite a single paragraph, and it can flag formulaic human writing such as lab report boilerplate or heavily templated essays.
The practical consequence is that you need both numbers before you can make a judgement about your own draft. A clean similarity score tells you nothing about AI detection, and a low AI score tells you nothing about whether you have copied a source without citing it. Treat them as two independent diagnostics on the same file [1][2].
How Do You Read A Turnitin AI Score And A Similarity Score Without Misinterpreting Them?
Read the AI score as an estimate with a confidence band, not as a measurement. Turnitin reports the percentage of the document its model classifies as AI-generated and also shows how much of that text sits in a lower-confidence range; below its confidence threshold the score is presented as an asterisk rather than a precise figure, signalling a weak signal rather than a clean bill of health [3]. A score of 12% and a score of 60% are qualitatively different problems: the first usually points to a few sentences, often an introduction or a transition paragraph, while the second suggests whole sections were generated or heavily paraphrased by a model.
Read the similarity score by opening the matches, never by reading the headline number alone. Click through each highlighted passage and ask three questions: is this a quotation I have attributed, is this a reference or a title, and is this phrasing genuinely mine or did I absorb it from the source? Turnitin's documentation stresses that the same percentage can mean very different things depending on what produced the matches, which is why instructors review the underlying report rather than the number [2]. A 20% score made up of bibliography entries and one quoted definition is a non-issue; a 20% score made up of unattributed paragraphs from two sources is a serious one.
Both reports also need to be read against your own writing process. If you drafted in your own words and used an AI tool only to tighten sentences, a moderate AI estimate is a signal to rewrite those sentences in your own voice and keep your earlier drafts as evidence of how the text evolved [3]. If large parts of the document were generated, the honest fix is revision, not arguing with the score — and the fastest way to know which situation you are in is to see the actual report before your instructor does.
Finally, remember that the report is not the decision. Detection models produce false positives on formulaic or non-native-English prose, and similarity engines produce false positives on common phrasing; both are documented limitations rather than edge cases [2][3]. What protects you is being able to point at your sources, your drafts, and your own reasoning for any passage that gets flagged.
Can You Check A Draft For AI Text And Plagiarism Before You Submit It?
Yes — checking a draft before final submission is normal practice, and the useful version of it happens early enough that you still have time to revise. Turnitin's student-facing guidance frames pre-submission checking as part of drafting: you review the flagged passages, rewrite what needs rewriting, and keep a record of the process so the final version is defensible [4]. Checking the night before a deadline technically counts, but it leaves you no room to act on what you find.
The mechanics matter as much as the timing. Institutional policies differ on whether a draft submitted through a course or library route is retained in a paper database, so where and how you check affects what happens to your file [4]. A non-repository check that returns both an AI report and a similarity report on the same document lets you compare the two diagnostics side by side and confirm that the file you revised is the file you are about to submit.
What you do with the report is the part that actually changes your outcome. Work through flagged passages in order of severity, rewrite rather than paraphrase-with-a-thesaurus, and re-check after revision — a rewrite that only swaps synonyms often keeps the same detectable pattern. Where a similarity match is a legitimate quotation, add or fix the citation instead of deleting the sentence. Where the AI estimate is concentrated in one section, that section is telling you something about how it was produced [3][4].
A pre-submission check is also the cheapest form of risk management available to you. The alternative is discovering the problem in your instructor's report, at which point the conversation is about intent rather than about a paragraph you could have rewritten in ten minutes.
If you want to see both reports on your own draft — the AI estimate and the similarity matches, in the same format your instructor sees — turnitin0 runs that check for you before you submit, returning a Turnitin AI detection report and a similarity report together, usually in under 15 minutes.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does a plagiarism checker also detect AI writing?
No. Similarity checking matches your text against existing sources, while AI detection classifies how the text was likely produced; Turnitin reports them as two separate reports [1]. You need both to understand your draft.
Is a high similarity score automatically plagiarism?
No. The similarity percentage is the share of your document that matched something in the comparison database, including correctly cited quotations and references [2]. Instructors review the matches in context before drawing any conclusion.
Why does my AI score change when I only edit a few sentences?
AI detection is a statistical estimate over the text, so rewriting a flagged passage can shift the reported figure noticeably [3]. That sensitivity is also why re-checking after revision is worth doing.
Can I check my draft before submitting it to my university?
Yes, pre-submission checking is a normal part of drafting, and it is most useful when done early enough to revise [4]. Check the retention policy of whichever service you use, since not all routes are non-repository.
What should I fix first if both reports flag the same paragraph?
Rewrite that paragraph from your own understanding and add any missing citation, then re-check. A passage that is both matched and AI-flagged is the highest-risk text in the document [3][4].