How Turnitin Works: What Your Similarity Score Actually Means

17 October 2026 · 8 min read

Students usually meet Turnitin as a single percentage, and panic or celebrate based on the number alone. Both reactions misread what the tool does. Turnitin is a text-matching system: it compares your submission against its databases and flags passages that match. The score is the proportion of your text that matched something. Whether any of those matches is plagiarism is a judgement a human still has to make — which is why markers open the report instead of reading the headline figure.

What Turnitin actually compares against

Three databases matter. First, a repository of previously submitted student papers — the largest and the one no free online checker can see. Second, current and archived public web pages. Third, academic journals, periodicals and publications from partnering publishers. Your draft is broken into overlapping text strings and matched against all three; each match is recorded with its source and its share of your word count.

Two practical consequences follow. A free web-based checker that shows 0% does not predict a low Turnitin score, because it cannot search the student-paper repository. And text copied from a classmate's old assignment — which many students assume is invisible — is precisely what the repository exists to catch.

Why honest assignments still score 15-25%

The reference list: every entry matches the same source published elsewhere. Most institutions exclude the bibliography from the score, but not all do.
Standard phrases: methodological boilerplate ("participants were informed that their responses would remain confidential") and discipline-specific terminology match across thousands of papers.
Properly quoted material: a quoted passage is a match even when quoted correctly. Quotation settings vary by institution.
Your own earlier submissions: drafts and related papers you submitted before are in the repository and will match.
The assignment brief itself, if you reproduced the question or marking criteria in your document.

This is why there is no universal safe percentage. A 20% score built entirely from the reference list, quotes and standard phrasing can be clean, while a 6% score that is one copied paragraph can be misconduct. Markers read the match breakdown, not the number.

Reading the report the way your marker does

If you get to see your report (many institutions allow it), work through the matched sources in order of size. The top matches tell the story. Ask of each: is this my reference list, a legitimate quote, standard phrasing, or a passage that is genuinely too close to the source? Only the last category is a problem, and it is the one to fix. Filters matter here: the report can exclude quotes, bibliography and small matches, and the score changes with the settings — another reason the raw number alone means little.

Fixing a genuinely high score

For each problematic match, the fix is not synonym-swapping — Turnitin matches structure and phrase sequences, so lightly edited copying still flags. Instead: return to the source, close it, and write the point from your own understanding in your own structure. Then cite the source for the idea. If a passage must stay close to the original, make it a proper quotation with quotation marks and a page number, and keep such passages short and rare. A paragraph that cannot be paraphrased without losing its meaning is usually a paragraph that should be quoted and then analysed.

Checking yourself before submission

Run your draft through any self-check your institution offers — many allow a draft submission that is not added to the repository.
Free web checkers catch public-web matches only; useful for finding pasted notes, useless for predicting the repository match.
AI-writing indicators in the report are even less decisive than similarity — they are flags for a human review, not evidence. Careful, formal human writing is regularly misread.
The strongest protection is process: draft from your own notes, paraphrase with the source closed, and cite as you write rather than retrofitting references at the end.

If a report has flagged your draft and you are not sure how to read it, Assignment Byte's editors can walk the matched sources with you, identify which ones actually need rewriting, and show you how to paraphrase and cite them properly — the fixes stay yours, the second pair of eyes is ours.