Direct answer
Plagiarism detection works by breaking a document into word strings and matching those strings against indexed databases of student papers, web pages, and scholarly publications, then highlighting every passage that overlaps with a source [1]. The output is a similarity report: a percentage plus a source-by-source list of what matched and where [1]. Crucially, a match is evidence of overlap, not proof of misconduct — quoted, cited, and bibliographic text can match legitimately, which is why a human still has to read each flagged passage [1].
How does plagiarism detection software actually detect plagiarism?
Detection is text comparison at scale, not mind-reading. The software converts your document into searchable sequences of words, then queries an index built from previously submitted papers, crawled web pages, and licensed journal content; when a sequence of sufficient length appears in that index, the passage is recorded as a match and attributed to the source it came from [2]. Because the unit of comparison is a string of words rather than an idea, the system can only find textual overlap — it cannot see that you read an article and reproduced its argument in your own vocabulary.
Matches are then grouped by source rather than listed as isolated fragments, so one paper might return a single source contributing 18% of the text or nine sources contributing 2% each [2]. That grouping matters for interpretation: a large single-source match usually points to a copied block, whereas many small matches often reflect common phrasing, definitions, or reference lists. The number you see is also settings-dependent — exclusions for quotations, the bibliography, and matches below a word threshold can move the index substantially, so two reports on the same file are not always comparable [2].
What the tool deliberately leaves out is intent. It reports overlap; it does not know whether you paraphrased carelessly, forgot a citation, or quoted properly with attribution [2]. That judgement belongs to the reader of the report, which is why every serious detection workflow ends with a human opening the flagged passages and checking the surrounding citation.
What does a plagiarism report show, and how do you read a similarity score without panicking?
A similarity report has two layers: a headline percentage and the underlying match list. The percentage represents the proportion of your document's text that matched sources in the database — it is a measure of overlap, not a verdict on your integrity [3]. The match list is the part that actually tells you something, because it names each source, shows the matched wording in context, and lets you decide whether that overlap is properly attributed [3].
Institutional reports often colour-code the score into bands, with cooler colours indicating lower overlap and warmer colours indicating higher overlap, and each institution sets its own threshold for what triggers a conversation [3]. A red band is therefore a signal to look closer, not a finding of plagiarism; conversely, a green band does not certify originality, because a short copied passage inside a long document can sit in a low band while still being a real problem. Reviewers are trained to open every match rather than react to the headline number [3].
The practical reading routine is to sort matches by size, then work down. For each match, ask three questions: is this wording quoted and marked as a quotation, is the source cited at the point of use, and does the sentence still need the borrowed phrasing at all? If the answer to the third question is no, rewrite it in your own structure — that removes the match permanently instead of hiding it behind an exclusion setting [3].
Can you check a draft for plagiarism before you submit it?
Yes, and doing so is normal academic practice rather than an admission of guilt. Students routinely self-check drafts to catch missing citations and accidental copy-paste before a deadline, and the useful part of that exercise is the revision it triggers, not the number it produces [4]. A pre-submission check is most valuable when you read every flagged match and fix the underlying attribution problem.
There is one structural catch worth understanding. When a draft is submitted through an institutional assignment, it is typically retained in the student paper repository, which means your own earlier version can later match your final version — and a draft shared with a classmate can match theirs [4]. A non-repository check avoids that side effect: the file is compared against the databases without being added to the student paper collection, so previewing a draft does not create a future originality problem for you or anyone else.
A workable routine looks like this: finish the draft, check it, open each match in order of size, add the citation or rewrite the passage, then re-check only if the changes were substantial [4]. Treat the score as a diagnostic. Chasing a specific number by swapping synonyms produces text that reads worse and still matches; fixing attribution and sentence structure is what actually clears a match.
Reading a similarity report is only useful if the report is the real thing — the same cover page, score, highlighted flags, and source summary your institution's system generates. turnitin0 produces exactly that pair of PDFs from a non-repository check, so the draft you inspect is never added to the student paper database, and most reports come back in under 15 minutes.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does a high similarity score mean I plagiarised?
No. The score measures how much of your text overlaps with sources in the database, and it includes properly quoted and cited material [3]. What determines plagiarism is whether the borrowed wording is attributed, which is a judgement a human makes by opening each match.
Can plagiarism detection software find paraphrased text?
It can find paraphrasing that preserves distinctive phrasing or sentence structure, because matching happens at the string level [2]. A paraphrase that genuinely restructures the sentence and adds a citation usually will not match, and that is the version worth writing.
Why do I get different similarity scores for the same paper?
Because the index, the exclusion settings, and the word-count threshold differ between reports [2]. Quotations and bibliography excluded in one run but included in another can shift the index noticeably.
Will checking my draft add it to the student paper database?
It depends on the system. Institutional assignment submissions are typically retained in the student paper repository, whereas a non-repository check compares the file without adding it to that collection [4]. Confirm which mode a tool uses before uploading.
What is the fastest way to lower a similarity score honestly?
Fix the attribution and rewrite the flagged sentences in your own structure, working from the largest match down [4]. Removing a match by rewriting is permanent; excluding it through a setting only changes the number you see.