Direct answer
A similarity detection service compares your document against a large body of existing text — web pages, academic publications, and previously submitted student papers — and returns a report showing which passages match those sources and how much of your text overlaps with them [1]. The percentage it produces is a measure of matched text, not a verdict on intent: a high score can come from correctly cited quotations, and a low score does not by itself prove originality [2]. Understanding what the service compares, and how the report is filtered, is what turns a number into useful feedback.
What Does a Similarity Detection Service Actually Check?
A similarity detection service works by matching strings of text, not by judging ideas. When you upload a document, the system breaks it into comparable segments and searches those segments against three broad pools of content: the open web, a paid index of journals and books, and a repository of papers previously submitted by other students [1]. Turnitin's own documentation describes this as a comparison against its repository, and the output is a set of highlighted overlaps rather than a single pass/fail signal [1]. This is why the same essay can score differently at two providers — the size and composition of the comparison database is a core part of the product, not a detail.
The practical consequence is that the service is answering a narrow, mechanical question: where does this sentence also appear? It is not answering did the author intend to deceive? That second question is reserved for a human being, usually your instructor or an academic integrity officer, who reads the highlighted passages in context [2]. A service that claims to detect intent is overstating what pattern matching can do.
A second consequence is that your own prior work can appear as a match. If you submitted an earlier draft through an institutional account that stores papers, a later submission of overlapping text may be flagged against your own previous file [1]. This is not an error; it is the repository doing what it was configured to do. It is also the reason some institutions deliberately run draft checks in a non-repository mode.
Finally, similarity detection is a different technology from AI writing detection. Similarity compares your text to existing text; AI detection estimates whether text was generated by a model. A paper can be 0% similar and still carry an AI score, or 30% similar with no AI signal at all [2]. Treat the two reports as separate diagnostics, not as two views of one number.
What Does a Similarity Report Look Like, and How Do I Read the Similarity Score?
The report is a document with your text on the left and a column of numbered source cards on the right, each card representing one matched source and listing the passages that overlapped [3]. The headline percentage is the share of your paper's text that matched something in the database; the match overview ranks sources by how much text each one contributed [2]. Reading the cards in descending order of contribution is the fastest way to find the passages that actually need attention, because the top two or three sources usually account for most of the score.
The number is also unstable in a way that surprises people. Institutional accounts almost always let the reader exclude quotations, the bibliography, and matches below a small word threshold — and applying those filters changes the displayed percentage, sometimes substantially [3]. A paper with a long block-quoted appendix can drop many points once quotations are excluded. This is why comparing your score to a classmate's is meaningless unless you know which filters were applied to each report [3].
What the score cannot tell you is whether a match is a problem. A correctly cited paraphrase still matches; a common phrase like "the results of this study suggest" matches thousands of papers. The judgement you need to make is whether each highlighted passage is attributed, whether the paraphrase is too close to the original wording, and whether the source is one you actually used [2]. Turnitin frames the report explicitly as a tool for interpretation rather than an automated finding of misconduct, and the same framing applies to any comparable service [3].
One useful habit: read the report as a checklist of edits, not as a grade. For each source card, decide cited, needs a citation, or needs rewriting in my own words. That converts an anxiety-inducing percentage into a concrete set of revisions you can finish before the deadline.
Can I Check My Own Draft With a Similarity Detection Service Before I Submit It?
Yes — but the route matters more than the tool. Many institutions enable a draft-check submission route, sometimes called a draft assignment or a self-check, that lets you run your own paper before the graded submission; whether that draft is stored in the student paper repository depends on how the institution configured the assignment [4]. Where such a route exists, it is the most accurate option available to you, because it uses exactly the database and settings your instructor will use.
Where no draft route exists, students commonly turn to independent checking services. The key question to ask any provider is whether the check is non-repository — that is, whether your file is added to a student paper database or shared onward. A non-repository check compares your draft against existing content without depositing your work into the pool that future students will be measured against [4]. It is also worth confirming that you receive the actual report rather than a bare percentage, since a number without source cards gives you nothing to revise.
Interpretation still belongs to a human. Turnitin's guidance to institutions emphasises that reports should be discussed and contextualised rather than acted on mechanically, and that framing should guide how you use any pre-submission check [4]. If a draft check returns a high score, the correct next step is to open the source cards and fix the passages — not to assume the submission is doomed. If it returns a low score, the correct next step is to confirm your citations are still complete, because a low similarity percentage is not evidence of good referencing [2].
A pre-submission check is therefore best understood as a rehearsal of the feedback loop: you see what an instructor would see, you revise, and you submit with the matches already resolved. That is a materially different activity from shopping for a reassuring number.
If you want to see the report an instructor would actually open — the cover, the score band, the flagged passages and the similarity summary side by side — turnitin0 runs a non-repository check that returns both the similarity report and the AI writing report as downloadable PDFs, usually in under fifteen minutes, so you can revise against real source cards rather than a single unexplained percentage.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is a similarity detection service the same as a plagiarism checker?
The technology is the same; the framing differs. A similarity service reports where text matches existing sources, while "plagiarism checker" implies a judgement the software cannot make [1]. The report gives you evidence; a person decides whether the match is a citation problem.
What similarity percentage is acceptable?
There is no universal threshold, and institutions set their own expectations. Because the score shifts when quotations and the bibliography are excluded, a figure that looks high under one filter set can look ordinary under another [3].
Does a similarity check store my paper?
It depends entirely on the provider and the assignment settings. Institutional assignments may store submissions in a student paper repository, while non-repository checks compare your draft without adding it to the database [4]. Always confirm which mode you are using.
Can a similarity report detect AI-written text?
No. Similarity matching and AI writing detection are separate analyses with separate reports [2]. A document can score 0% similarity and still be flagged for AI writing, so check both if your institution reviews both.
Why did my score change between two checks?
Different comparison databases, different filter settings, or a repository that now contains a paper matching yours will all move the number [1][3]. Compare reports only when you know the database and filters behind them.