What Your Turnitin Similarity Score Actually Means (2026 Student Guide)
By Daniel Kruger 9 min read
You submit your assignment, the report comes back, and there it is: 27%. Is that bad? Is 15% fine? Is anything above zero a problem? Every semester I watch students panic over a number that, on its own, means far less than they think. Here is what the similarity score actually measures, what markers do with it, and how to deal with a high one honestly.
What the Similarity Score Actually Measures
The Turnitin similarity score is the percentage of your text that matches other sources in Turnitin’s database: published articles, websites, and previously submitted student papers. That is the whole definition. It is a text-matching tool, not a plagiarism verdict.
This distinction matters more than anything else in this article. Turnitin does not know whether a match is a properly quoted passage with a citation, a reference list entry that obviously matches the original source, a common phrase like “statistically significant difference between the groups”, or copied text passed off as your own. It highlights the match. A human decides what the match means.
So a 30% score made up of quoted, cited material and a reference list can be completely fine. A 9% score where the entire 9% is one unattributed paragraph lifted from a journal article is a serious problem. The number tells you how much matched. It says nothing about why.
There Is No Universal Safe Percentage
I know that is not what you came here for, so let me give you the honest version of the numbers anyway. Most universities treat scores under roughly 15% as unremarkable, and many flag papers above 20 to 25% for a closer manual look. Some departments set their own thresholds. A law essay built on statutes and case citations will naturally match more than a personal reflection piece. Your institution’s threshold, if it publishes one at all, is in your course handbook or assignment brief, and that document beats anything a blog tells you.
Turnitin itself has been pushing in this direction. Its updated Similarity Report positions the score as one input into an integrity conversation rather than a pass or fail line, and the guidance it gives instructors says the same thing markers have said for years: read the matches, not the number.
One more thing worth knowing. Two students can submit equally honest work and get very different scores. If your methodology section describes a standard survey design, chunks of it will resemble hundreds of other methodology sections, because there are only so many ways to say “participants completed the questionnaire online”. That is not misconduct. It is the nature of academic writing in a database of millions of papers.
The Similarity Score Is Not the AI Score
Turnitin now shows two separate indicators, and students mix them up constantly. The similarity score measures matching text, as above. The AI writing indicator is a different tool that estimates how much of the document reads like it was machine generated.
They can move independently. Fully AI-generated text often has a low similarity score, because the model produces new sentences that do not match the database. Turnitin’s own figures from earlier this year suggest around 15% of submissions now contain a large majority of AI-generated writing, up from about 3% when its detector launched in 2023, which tells you why universities are paying attention to process rather than just output. AI detection is also famously unreliable at the level of an individual accusation, which is why a growing number of institutions have moved away from it, something we covered in AI detection is dead, universities are switching to process-based evaluation.
For this article, put the AI score aside. Your similarity score is about matching text, and matching text is something you can inspect and fix yourself before you submit.
How to Read the Report Like a Marker Does
When a marker opens a flagged report, they do not stare at the percentage. They look at three things.
- Where the matches sit. A score spread thinly across forty sources, a phrase here and a phrase there, reads very differently from 12% matching a single essay-mill website. Concentration is what raises eyebrows, not volume.
- What kind of text is matching. Quotes, references, headings, and standard terminology are expected matches. Turnitin can be set to exclude quoted material and bibliographies, and many instructors do exactly that. If yours did not, your score may include your reference list, which inflates the number without meaning anything. This is also why your score sometimes drops when a marker applies filters you cannot see.
- Whether matched passages are attributed. A highlighted paragraph next to a citation and quotation marks is a non-event. The same paragraph with no citation is the beginning of a difficult conversation.
If you get access to your own report before final submission, read it exactly this way. Click into each significant match and ask: is this quoted and cited, is it a reference entry, is it standard phrasing, or is it something I actually need to rewrite or attribute?
Fixing a High Score Honestly
Say your draft comes back at 38% and the matches are real. Here is what actually works, and none of it is a trick.
- Rewrite proper paraphrases. Most inflated scores I have seen come from patchwriting: taking a source’s sentence and swapping a few words. Turnitin catches it because the sentence skeleton survives, and it catches it correctly, because patchwriting is a form of plagiarism even when you cite. The fix is to close the PDF, write the idea from memory in your own structure, then cite it. Our guide to paraphrasing without plagiarising walks through the method with before and after examples.
- Quote what deserves quoting. If a sentence is doing precise work, a legal definition, a famous claim, an author’s exact framing, quote it with quotation marks and a citation. Quoted material is honest matching. Trying to paraphrase a definition that only has one correct wording just produces bad paraphrase.
- Add the missing citations. Sometimes a match is your own summary of a source that you forgot to cite. The text is yours; the idea is not. Add the citation. If your citations themselves are shaky, check them against the format guides for APA or whichever style your department uses.
- Cut the filler you absorbed. Long stretches of scene-setting (“The rapid growth of technology has changed many industries”) match because thousands of students have written the same empty sentence. Deleting it improves the essay and the score at the same time.
What I would not lose sleep over: your reference list matching, your assignment cover sheet matching, technical terms matching, or your own previously submitted draft matching if your university uses draft submissions. Mention that last one to your marker if it happens; self-matching from an earlier draft of the same assignment is a known quirk.
What Not to Do
The internet is full of tricks for lowering a similarity score: paraphrasing spinners, character substitution, inserting invisible symbols, translating text through two languages and back. Do not do any of this.
Partly because it is misconduct, and misconduct aimed at defeating an integrity check is treated more harshly than the original problem would have been. Partly because it does not work; markers read the actual essay, and spun text reads like spun text. And partly because it misses the point. A high similarity score caused by real problems is information. It is telling you the draft leans too hard on its sources, and that is fixable by writing, not by laundering text.
The same logic applies to buying “guaranteed low similarity” essays. Contract cheating services submit recycled work constantly, which is exactly how previous customers end up with 60% matches against another student’s submission.
If You Get Flagged Anyway
Sometimes honest work gets questioned. If that happens, the strongest thing you can bring to the meeting is evidence of your process: your notes, your source PDFs with highlights, your version history, your drafts. Markers are generally reasonable when a student can walk them through how the essay was built. We wrote a full guide on this in how to prove you wrote your own work, and it pairs well with keeping an AI-use disclosure statement if your course allows AI tools in any part of your workflow.
And check your references before you submit, every time. A fabricated or wrong citation does more damage to your credibility than any similarity percentage, and it takes five minutes to catch, as we showed in how to spot fake AI citations.
Turn Your Reading Into a First Draft
Most similarity problems start in the same place: a student with twenty open PDFs, a deadline, and no time left to put the reading into their own words, so the sources’ words creep in. The fix is starting the writing earlier, from your own sources, with citations attached from the first sentence.
Litrevu is an AI literature review assistant that turns the papers a researcher has already gathered into a cited first draft, with every citation traceable to the uploaded source.
Litrevu is an AI literature review assistant that produces a cited first draft from a researcher’s uploaded papers, and it is designed to be reviewed and revised rather than submitted as it stands. Every citation traces back to a source PDF, so quoted material can be checked.
That is the part Litrevu helps with. You upload the papers you have already gathered, and it drafts a cited first version of your literature review from them, with every claim traceable to a source you can open and check. You read every line, verify every citation against the original, rewrite what needs rewriting, and own every word. The first 800 words are free, no credit card required.
Try it free