Direct answer
Plagiarism detection software checks plagiarism by providing a database of source material — previously submitted student papers, archived and live web pages, and subscription publications — that each new submission is compared against. The software does not judge originality from nothing; it searches for overlapping strings of text inside that stored collection and reports where matches occur [1]. The database is therefore the core of the system: no comparison set, no detection.
What Kind of Database Does Plagiarism Detection Software Check a Paper Against?
The short answer is that it is not one database but three comparison sets running in parallel. Turnitin, the system most universities use, compares a submission against its student paper repository (papers previously submitted to institutions using the service), its internet index of current and archived pages, and a body of subscription publications such as journals, periodicals, and books [2]. Each of these is maintained separately, and each contributes its own matches to the final report.
The repository deserves particular attention because students often misunderstand it. A paper enters that database only if it was actually submitted through an institutional account — it is a record of real coursework, not a scraped copy of the open web [1]. This is why two students at the same university can match each other, and why a paper recycled from a different institution may still surface if that institution also uses the service.
The internet index is the largest and most volatile set. It includes live pages, cached versions, and archived snapshots, which is why a passage copied from a page that has since been deleted can still be detected [2]. Publications add the paywalled layer — journal articles, book chapters, and periodicals that a normal search engine cannot reach.
What the database does not contain matters just as much. Unpublished work, offline lecture notes, private messages, and paywalled sources outside the licensed index are invisible to the comparison [1]. A paper can be genuinely unoriginal and still score low if its source sits outside the comparison sets.
How Does a Database Comparison Actually Produce a Similarity Score and a Match Report?
The mechanics are string matching, not semantic judgement. The software indexes the submission, breaks it into overlapping sequences of words, and looks for those sequences inside the comparison database; matched passages are then highlighted and attributed to the source they came from [3]. Everything you see in a report — the highlighted text, the source list, the percentage — is built from those matches.
The headline percentage is a ratio, not a verdict. Turnitin calculates the similarity score as the volume of matched text divided by the total volume of text in the document, and institutional settings can exclude quotations, bibliography, and small matches below a set word count or percentage threshold [3]. Two papers with identical scores can therefore have completely different underlying problems: one may have copied a single long paragraph, the other may have scattered short matches across a dozen sources.
The match report is where the useful detail lives. The All Sources view ranks every match by the amount of text involved and names the source type, so a reader can see at a glance whether a hit came from a journal, a website, or another student's paper [3]. Reviewers are expected to open each match and judge it — a correctly cited quotation still appears as a match, and a legitimate paraphrase may appear as a low-percentage overlap.
This is also where the database model shows its limits. Because matching is textual, a thorough paraphrase that preserves meaning while changing wording can pass a database comparison untouched [3]. Detection measures textual overlap with a known corpus; it does not measure whether an idea was borrowed.
Can a Student See the Same Database-Match Results Before Submitting?
Usually yes — but only if the assignment is configured that way. Whether a student can view a similarity report on a draft depends on the settings the instructor chose for that assignment, including whether the report is released to the student at all and whether resubmissions are permitted [4]. Many institutions deliberately enable draft or formative submissions so students can see matches and revise before the graded version is due.
There is a practical catch worth knowing. Repeated submissions of the same file can interact with the repository depending on how the assignment is set up, and a paper submitted to a draft folder may be stored and matched against later [4]. Students who check the same essay multiple times should confirm the repository behaviour for their specific assignment rather than assume each attempt is isolated.
The more reliable route is a non-repository pre-check: a comparison run that returns the same kind of match data without adding your file to a student paper database. That gives you the diagnostic value of the database comparison — which passages match, from which source type, at what percentage — without the side effects [4]. For a student whose work is entirely hand-written, this is mainly reassurance; for anyone who has used AI assistance, it is the difference between discovering a problem now and discovering it after the deadline.
Reading about how the database works only gets you so far — the number that matters is the one attached to your own draft. turnitin0 runs your document against the same comparison sets and returns the actual Turnitin AI and similarity reports, match highlights included, in most cases within 5–15 minutes, without adding your file to the student paper repository.
※ Turnitin0.com - Actual [Turnitin AI](https://www.turnitin0.com/) Report Cover, Score, Flag And Similarity Summary
FAQ
Does plagiarism detection software store every paper it checks?
Not necessarily. Whether a submission is added to the student paper repository depends on the service and the assignment settings, and non-repository checks exist specifically so a file is compared without being stored [4].
Is a high similarity score proof of plagiarism?
No. The score is a ratio of matched text to total text, and it includes correctly cited quotations and reference lists unless those filters are applied [3]. A human has to open the matches and judge each one.
Can plagiarism software detect text copied from a book or paywalled journal?
Only if that publication is inside its licensed index [2]. Sources outside the comparison sets — including many books and unlicensed databases — will not produce a match even when the text is copied verbatim.
Why did my paper match a source I never saw?
Common explanations are a shared quotation, a standard phrase, a template, or an archived copy of a page you did read [1]. The match report names the source type, which usually makes the origin obvious.
Does the database catch paraphrased or AI-generated text?
A database comparison catches textual overlap, so a thorough paraphrase can pass it unnoticed [3]. AI writing is a separate detection problem handled by a different model, which is why Turnitin reports AI writing and similarity separately.