Turnitin Similarity Report: What Each Number Means
The percentage counts words, not misconduct
A similarity report is a matching tool. The headline number is the share of your document that was found to match something in a comparison database, counted by word. Nothing about that number, on its own, distinguishes a correctly cited quotation from a copied paragraph, because the matching algorithm does not know what a citation is — it matches strings.
Most institutions configure their report to exclude the bibliography and the quote marks. If yours does not, your reference page alone can push the number up by ten or fifteen points. Before you panic, check which exclusions are switched on.
What the parts of the report are
The similarity percentage. Matched words against total words, after exclusions. This is the number everyone quotes and the least informative part of the report.
The sources list. Every matched source with its own contribution. This is where the real information lives. Ten matches of 1% each is a very different situation from one match of 10%.
The highlighted text. Colour-coded overlap in the body. Multiple sources can land on the same sentence, and the colours stack.
The match breakdown. Whether the matched text was found in a publication, another student's paper, or a web page. Institutional repositories behave differently from the open web.
Reading common numbers without panicking
- 1–10% is the normal background noise of academic writing. Common phrases, standard method descriptions and well-known definitions all match something.
- 10–25% is usually citation density. Look at whether the overlaps are quotations and paraphrases with attribution.
- 25%+ with a few large blocks is worth reading carefully. It can still be legitimate — a long appendix, a data table, a quoted statute — but it is the level where a human will open the report.
- One 40% source is a different animal to forty 1% sources, even at the same total. The first means one place you did not paraphrase; the second usually means your phrasing is generic.
What instructors actually look at
The number is a triage signal, not a verdict. The questions that get asked in a conversation are usually these:
- Is the highlighted passage attributed, in the text, in the way your field expects?
- Does the paraphrase actually paraphrase, or does it swap two words and keep the sentence?
- Is the matched source in the bibliography?
- If it is your own earlier work, is it cited as such?
None of those can be answered by reading a percentage. Which is why the useful move before submitting is to open the report yourself, source by source, and fix the one or two places where the highlight sits on a sentence you wrote from memory.
The difference between similarity and AI detection
They are reported on different tabs and measure unrelated things. Similarity compares your text to a corpus of existing text. AI detection reads style. You can have 0% similarity and a high AI score — an original essay written by a model. You can have 35% similarity and no AI score — a well-sourced literature review. Confusing the two is the single most common misreading, and it is why the same report gets described as "I got flagged for plagiarism" when nobody said that.
Does a similarity score follow you?
In most institutions, submissions go into a repository and later submissions get matched against them. That is what makes the matched to another student's paper category possible. It also means resubmitting your own earlier essay without citation is a real match, not a glitch. If you are reusing your own work, cite it the way your department requires.
Before you submit
- Download the report and read the sources list, not the headline.
- Check the exclusion settings your institution uses.
- Fix the single largest block first; it is almost always a paragraph you wrote with the source open.
- Rewrite from your notes rather than from the sentence you can see. Paraphrasing while looking at the original is how the matched phrasing survives.
- Keep the drafts. They are the answer to every later question.
Frequently asked
Is a 20% similarity score bad? Not by itself. It means a fifth of your words matched something after exclusions. Whether that is a problem depends entirely on whether the matches are attributed and whether the paraphrasing is real.
Can I get a high score from my own previous essay? Yes. Institutional repositories keep submissions, so your own work matches.
Does the report tell teachers I used AI? No. That is the other tab, a separate model, and a separate score.
Why did the number change between drafts? Because you changed the words, and possibly the exclusion settings. Small edits can move it several points either way.