Is 36% Similarity on Turnitin Bad? What the Score Actually Means

Reviewed and updated August 2026 against Turnitin’s own similarity-score guides.

If you have just opened a Similarity Report and seen a number you did not expect, the useful next step is not to compare it against a figure you read on a forum. It is to open the report and look at what is highlighted, because two submissions with identical scores can mean completely opposite things.

What the similarity score actually measures

Turnitin is unusually direct about this, and it is worth quoting because it contradicts how most people talk about the tool: “Turnitin does not check for plagiarism in writing. We do check all submissions against our database.”

The database is large — Turnitin describes billions of web pages, both current and archived, a repository of works students have submitted previously, and thousands of periodicals, journals and publications. Anywhere your text resembles something in there, the passage is highlighted. The score is the proportion of your submission made up of those highlights.

Turnitin then hands the judgement to a human, in almost the same words the other vendors in this category use: use the score “as a tool within your review process to make your own determination if any academic misconduct is present.” There is no automated verdict anywhere in that sentence.

So is 36% bad?

It depends entirely on composition, and this is the part a single percentage cannot tell you.

36% that is probably fine: a well-referenced essay with several block quotes, a long bibliography, a standard methods description and some unavoidable technical phrasing. Each highlight is short, attributed and spread across many different sources.

36% that is a genuine problem: a single continuous passage of several hundred words matching one uncited source. The percentage is identical; the meaning is not.

Why a legitimate paper still scores high

Turnitin says outright that “it is perfectly natural for a submission to match against some of our database,” and that quoted text still shows as a match even when quotation marks and references are used correctly. These are the usual contributors:

Short assignments inflate all of this. On a 500-word piece, one quoted paragraph and a reference list can be a third of the document. The same content inside a 5,000-word dissertation barely registers, which is why comparing your score to a friend’s on a different assignment tells you very little.

Discipline matters too. Law, medicine and the sciences carry more fixed terminology and standard formulations than an English essay does, so identical honesty produces a higher number in some subjects than others. A score that would raise eyebrows on a creative-writing assignment can be entirely routine on a clinical case report, and a marker who teaches that subject will already know it and will read your report with that expectation in mind.

Is 10% bad? Is 30%? Is 50%?

These are the three numbers people search for most, so they are worth answering individually — with the same caveat attached to all of them, which is that Turnitin sets no threshold and your institution’s guidance always outranks a number from the internet.

Around 10% is what a normally-referenced essay often looks like. It usually reflects citations, a reference list and ordinary shared phrasing. It is rarely queried, though a 10% made up of one continuous uncited passage would still be a problem — composition beats magnitude every time.

Around 30% is the range where students start to worry, and where the answer genuinely depends. On a quote-heavy literature essay with a long bibliography and a modest word count, 30% can be entirely legitimate. On a short reflective piece with two references, 30% suggests a large block of text came from somewhere and is worth examining.

Around 50% and above warrants a proper look, because at that point half the document matches existing sources. It still is not automatic misconduct — heavily templated assignments, legal or clinical documents, and papers built around extended primary-source quotation can legitimately reach it — but it is high enough that you should be able to explain where it comes from.

If you take one thing from this section: none of these are rules. They are descriptions of what those ranges typically contain. Your department may publish its own guideline, and if it does, that is the number that applies to you.

What instructors actually look at

Experienced markers spend very little time on the headline percentage. The report itself carries the information that matters, and knowing what they check tells you how to read your own report.

The longest single match. One unbroken 400-word match is a different conversation from four hundred one-word matches, even where the percentage is identical.

How many sources are involved. Matches spread thinly across dozens of sources look like normal academic writing. Concentration in one source does not.

Whether matches are attributed. A highlighted passage sitting next to a citation reads completely differently from the same passage with nothing around it.

Where the match sits. Similarity concentrated in a reference list or methods section is expected. The same volume in an analysis or discussion section is the part that is supposed to be yours.

Whether the source is another student. Turnitin’s repository includes work other students have submitted, and its guides address collusion directly — which is why a match to a classmate’s paper is treated differently from a match to a journal article.

Similarity is not AI detection

These two get conflated constantly, and confusing them causes real distress. They are separate measurements answering separate questions.

Similarity asks: does this text match something already in the database? It is a comparison against existing documents, and everything on this page is about that number. Turnitin’s similarity guides describe matching against web content, publications and student submissions — they do not describe authorship analysis.

AI-writing indicators ask something entirely different: does this text look like it was generated rather than written? That is a prediction about how the text was produced, not a match against a source. Text can be wholly original — 0% similarity — and still be flagged by an AI indicator, because nothing it says exists anywhere else. The reverse is equally true.

So a low similarity score is not evidence about AI use, and an AI flag is not evidence of copying. If a concern has been raised with you, establishing which of the two it is should be the first question, because the appropriate response is completely different.

What the report colours mean

The icon colour in your assignment inbox reflects how much matching text was found, on a range from 0% to 100%. Turnitin documents blue for no matching text and green for one word up to 24% matching text, with further colour bands above that.

It is worth being clear about what the colours are not: they are a visual shorthand for the amount of matched text, not a severity rating. A green report can still contain one unattributed passage that matters, and a report in a higher band can be entirely quotations. The colour is a prompt to open the report, not a conclusion.

How to lower a similarity score legitimately

The honest version of this advice is short, because the goal is not a smaller number — it is writing that is properly your own with properly credited sources. Those two things usually produce a lower score as a side effect.

Quote less and explain more. Where a quotation is not doing specific work, replace it with your own analysis of the point. This reduces matched text and almost always improves the writing.

Cite everything you use. Citation does not remove a match, but it converts an unattributed match — the kind that actually causes problems — into a legitimate one.

Ask about exclusion settings. Instructors can exclude quotes and bibliographies from the calculation. Whether that is enabled is an assignment-level decision, and it is a perfectly reasonable thing to ask about.

⛔ What we will not tell you is how to defeat the check. Rewording tools that exist to disguise copied text do not make work original — they leave you submitting someone else’s ideas with the words shuffled, which is the thing academic-integrity policies are actually about.

Drafts, resubmissions and matching your own work

One of the most alarming things that happens to students is a near-total similarity score on a paper they wrote themselves. The explanation is almost always mundane.

Turnitin’s database includes a repository of works students have submitted to Turnitin in the past. If an earlier draft of your assignment was submitted — to a draft inbox, to a previous attempt, or to a different course — that copy can sit in the repository. Your new submission then matches it, sometimes almost completely, and the report shows a very high percentage against a source that is you.

This is usually easy to resolve, because the report names the matched source. If it points at your own prior submission, that is visible in the report and your instructor can see it too. Many courses deliberately set up a separate draft inbox that does not store submissions in the repository, precisely to avoid this.

A genuinely different question is reusing your own work across assignments. Some institutions treat submitting the same work twice for separate credit as an integrity issue in its own right, sometimes called self-plagiarism. That is a policy question rather than a Turnitin question, and the answer varies — if you intend to build on something you wrote before, ask first.

If your instructor raises your score

Open the report together and look at the matches rather than the number. Ask how many distinct sources are involved and how long the longest single match is — those two facts explain most of any score. If your matches are quotations and references, they are visible as such in the report, and that is a straightforward thing to demonstrate.

This mirrors what every other tool in this category says about its own signals. Instructure states Canvas quiz logs “are not intended to validate academic integrity”; Proctorio says its technology “informs human judgment rather than replacing it”; Respondus sends flagged events to the instructor for review. Across the board, the software surfaces things and people decide. See the full guide to online cheating detection, or what Canvas can detect.

Frequently asked questions

Is 36% similarity on Turnitin bad?

Not automatically. Turnitin publishes no pass/fail threshold, because the similarity score is not a plagiarism score. Turnitin states the score is “simply the percentage of text in a submission that matches other sources” and asks instructors to use it “as a tool within your review process to make your own determination if any academic misconduct is present.” A 36% made up of quotes, references and standard phrasing is very different from 36% lifted from one uncited source.

What is a good Turnitin percentage?

There is no official number. Some departments publish an internal guideline, but that is a local policy choice, not a Turnitin standard. Your course handbook or instructor is the only authoritative answer for your work.

Does a 0% score mean my work is perfect?

It means no matching text was found. Turnitin’s colour scale shows blue for no matching text. On a paper with quotes and a reference list, a 0% is unusual enough to be worth checking that the file uploaded correctly.

Does Turnitin detect plagiarism?

Turnitin is explicit that it does not: “Turnitin does not check for plagiarism in writing. We do check all submissions against our database.” It finds text matches. A person decides whether a match is misconduct.

Does Turnitin check work previously submitted?

Its database includes a repository of works students have submitted to Turnitin in the past, alongside web content and published material. That is why a resubmitted draft can match your own earlier upload.

Can I see my similarity score before the deadline?

Only if your instructor enabled it. Turnitin notes that a score may not appear because your instructor turned off student access to the Similarity Report.

Why does my score include the reference list?

Because it is text, and it matches other reference lists. Instructors can choose to exclude bibliographies and quotes from the calculation, but whether that setting is on is decided at the assignment level.

Does Turnitin detect collusion between students?

Turnitin’s own guides address student collusion directly, and the mechanism is the same one used for everything else: your submission is compared against the repository, which contains other students’ submitted work.