Direct answer
Checking for plagiarism and AI means running two separate analyses on the same file: a similarity check that compares your text against published sources and a student-paper database, and an AI-writing check that estimates how much of your prose reads as machine-generated. They are produced by different models, scored on different scales, and read with different rules — which is why one number can look alarming while the other looks clean [1]. A combined check gives you both reports side by side before your deadline, so you can fix citation gaps and rewrite flagged passages while you still have time to do it properly [4].
What Is the Difference Between a Plagiarism Check and an AI Check?
A plagiarism check asks where else has this text appeared; an AI check asks how was this text produced. The similarity engine matches strings of your submission against the repository, web pages, and journals, then returns an overall similarity index plus a match overview listing each source with the percentage of your document it accounts for [1]. The AI detector ignores source matching entirely and instead analyses sentence-level patterns — perplexity and burstiness — to estimate what share of your qualifying prose looks machine-written [2]. Nothing in the similarity engine feeds the AI score, and nothing in the AI score changes your similarity index.
The two reports also cover different amounts of your document. Similarity is calculated across the whole submission, so a reference list, a long block quote, or a standard methods template can all push the index upward without any misconduct [1]. AI detection, by contrast, only evaluates long-form English prose in qualifying segments, and Turnitin has been explicit that short answers, maths, tables, and bullets are outside its scope [2]. That asymmetry explains a very common student experience: a 38% similarity score next to a 4% AI score, or the reverse, on a file that has not changed at all.
Finally, the two indicators carry different evidential weight. Turnitin positions the similarity index as an interpretive starting point for an instructor, not a verdict, because correctly cited quotations still register as matches [1]. Its AI guidance follows the same logic, warning institutions that the AI score is an indicator to be weighed alongside drafting history, version files, and conversation with the student rather than treated as proof on its own [2]. Reading both reports as questions to ask about your draft — rather than as grades — is the single most useful habit to build.
Why Do Plagiarism and AI Scores Not Always Match What Students Expect?
The most frequent surprise is the asterisk. When the AI detector's result falls below its 20% confidence threshold, Turnitin displays *% instead of a number, and that asterisk means low-confidence signal, not zero and not a clean bill of health [3]. Students who read *% as "0% AI" sometimes submit prose that still reads heavily machine-polished, while students who read it as a failure rewrite text that was never flagged in the first place. The asterisk is a statement about the detector's certainty, not about your writing.
The second mismatch comes from what each score is measuring per unit of text. The AI percentage is calculated over qualifying prose only, so a document with a large appendix of tables and code can show a high AI figure driven by a small flagged section, while a document that is 100% prose spreads the same flagged sentences across a much larger denominator [3]. Similarity works the other way: a single 400-word passage copied from one source can dominate the index because one source accounts for a large share of the total word count [1]. Two files with identical problems can therefore produce very different-looking numbers.
The third mismatch is disciplinary and stylistic. Dense academic register — passive constructions, formulaic transitions, repeated terminology — is exactly the pattern that AI detectors find statistically predictable, so carefully written literature reviews and methodology sections attract false positives more often than reflective or argumentative writing [2]. Meanwhile, similarity indexes run high in subjects where standard phrasing is unavoidable, such as legal citations or lab protocols [1]. Neither pattern is evidence of misconduct, and both are precisely why a human reviewer is meant to interpret the report rather than automate a penalty [2][3].
Can You Check a Draft for Plagiarism and AI Before You Submit It?
Yes — but the route matters, because the two obvious options behave very differently. The first is an institutional draft assignment: many courses enable a pre-submission dropbox whose report you can view, and that report is generated inside the same system your final submission will use [4]. The catch is that a repository-enabled draft check adds your paper to the student paper database, so if you later submit a revised version, your own earlier draft can appear as a match against you [4]. Always confirm whether a draft dropbox is repository or non-repository before you use it.
The second route is an independent pre-submission check that runs without adding your file to the student paper database, which avoids the self-match problem entirely and keeps your report out of third-party databases [4]. This is the practical option when your course has no draft dropbox, when your deadline is close, or when you want to iterate — check, revise, check again — without leaving a trail of indexed drafts behind you. Whichever route you choose, the workflow that works is the same: read the match overview first and fix uncited borrowing, then read the AI report and rewrite flagged passages in your own voice, then re-check the revised file rather than assuming the fix landed [1][4].
One caution applies to both routes: a clean report is not a licence to submit unread. Reports describe a file at a moment in time, and the AI score in particular can shift after even modest edits, because rewriting a flagged paragraph changes the sentence-level statistics the detector is reading [2][3]. Treat the report as a diagnostic on a specific draft, not as a permanent property of your work.
If you would rather see both reports on your own draft before the deadline — the AI detection PDF and the similarity PDF, delivered together, generated without adding your file to the student paper database — turnitin0 runs that combined check in under 15 minutes in 98% of cases, with 100,000+ reports delivered to 20,000+ students in the US, UK, Canada, Australia, New Zealand, and Ireland.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Do I need two separate tools to check plagiarism and AI?
No. A single submission can return both a similarity report and an AI-writing report, because the two analyses run on the same uploaded file but use independent models [1][2]. What you should avoid is assuming one report implies anything about the other.
What does an asterisk instead of an AI percentage mean?
It means the detected AI share fell below Turnitin's 20% confidence threshold, so the system reports *% rather than a precise figure [3]. It is a low-confidence signal, not proof of zero AI use — read it together with the flagged passages in the body of the report.
Can a high similarity score appear on completely original work?
Yes. Correctly attributed quotations, standard methodology phrasing, and reference lists all register as matches, which is why Turnitin describes the similarity index as an interpretive tool rather than a plagiarism verdict [1]. Check the match overview before drawing conclusions.
Will checking my draft first cause my final submission to match itself?
Only if the draft check writes to the student paper database. A non-repository check does not add your file to that database, so your later submission will not be matched against your own earlier draft [4].
Does rewriting flagged text reliably lower the AI score?
Usually, but not predictably. The detector reads sentence-level statistical patterns, so genuine restructuring in your own voice changes those patterns far more than swapping synonyms does [2]. Re-check the revised file rather than assuming the edit worked.