What is a good Turnitin similarity score?
A 23% similarity score is not, on its own, trouble, and Turnitin does not think it is either: the company states in writing that there is no fixed number to receive as a score in your Similarity Report. The percentage counts matched words, and it counts your correctly cited quotations and your entire reference list right alongside anything you actually copied, unless your instructor switched those exclusions on. What decides your case is what the highlighted text turns out to be, plus a local rule that varies wildly: Ulster University treats 16 to 30% as a moderate misconduct concern, while the London School of Economics has a formal policy forbidding its departments from using any threshold at all. A 23% made of forty small matches to common phrasing and correctly cited quotes is ordinary academic writing; a 4% made of one unattributed paragraph is the score that gets referred.
Most people searching this have a number in mind and want to know which side of a line it falls on. There is no line. Turnitin publishes no pass mark, most universities publicly refuse to name one, and the few that do disagree with each other by a factor of three. If the number worrying you is the AI indicator instead, that is a different measurement and we cover it in what your Turnitin AI score means.
There is no pass mark, and Turnitin says so
Start with the company that makes the number. Turnitin's help centre answers the question directly and negatively: "There is no single 'acceptable' or 'perfect' Similarity score. The percentage returned is a Similarity score, not a plagiarism score." In a 2022 post for educators it went further, writing that "there is no good or bad similarity score, no magic number", and that "0% similarity is inherently neither good nor bad, nor is 30%, or even 80%."
The clearest demonstration is Turnitin's own worked example. One student copies and pastes a chunk of text they did not understand and scores 20%. Another, who knows the material and gathers evidence from several sources with correct quotation and referencing, scores 22%. The careful writer has the higher number, and that comes from the vendor's instructor documentation.
None of which stops people being referred over numbers, because departments and markers set local rules Turnitin has no say in. The honest answer has two halves: there is no universal threshold, and there may well be a specific one that applies to you.
What the percentage actually counts
The arithmetic is published. Turnitin calculates overall similarity "by dividing the number of matching words by the total number of words in the document", and a submission needs at least 20 words before a report is generated at all.
What counts as a match is broader than students expect. Correctly quoted material matches. Your reference list matches, because other people cited the same works. Standard phrasing in your field matches, and a literature review on urban heat or grid-scale storage is full of standard phrasing. Turnitin's smallest-match filter cannot be set below eight words, which tells you how short a matching run can be.
Length moves the number too. Turnitin states that "longer assignments tend to have higher similarity scores", and Strathclyde's Senate-approved guidance makes the point from the other end: 10% of a 12,000-word dissertation is a different animal from 10% of a 300-word short answer.
The number under the number
Turnitin's newer report carries a far more useful figure than the headline. Match Groups splits your similarity by whether it detected quotation marks and an in-text citation around each match: Cited and Quoted, Missing Quotations, Missing Citation, and Not Cited or Quoted. That last group is what anyone reviewing your work looks at, and it is usually a small fraction of the total.
There is a catch worth knowing before you panic at it. Turnitin says the citation recognition behind Match Groups is "trained on English content" and on "APA, MLA, Turabian, and IEEE numbered citations and references". Write in Harvard, Vancouver or Chicago notes and a perfectly attributed passage can land in Not Cited or Quoted because the software missed the citation, not because you failed to make one.
Your instructor changed your number before you saw it
The biggest thing students are never told is that the percentage is a setting, not a fact. When your instructor built the assignment they chose which repositories to compare against and which of Turnitin's five exclusions to apply: bibliography, quoted text, cited text, small matches and small sources. Two identical papers submitted to two assignments in the same department can return very different numbers.
Heriot-Watt puts a figure on the swing. Its staff guidance warns that "percentages are easy to manipulate via assignment settings and individual submission settings", and its illustration notes the exclude-bibliography toggle moving the same paper by 60 points or more when Turnitin misreads where the reference list starts.
Misreading it is easier than it sounds, because that filter is keyword driven. Turnitin publishes the trigger list, some sixty words including references, works cited, sources and footnotes. If one appears earlier in your document, say in a table of contents, Turnitin can treat it as the start of your reference list and "the entire remainder of the student's paper will be excluded from the Similarity Report".
If you can see your own report, you can switch off quotes and the bibliography and watch your number fall. It is worth knowing what that is worth. Turnitin's own student guidance says that "any exclusions you make to your Similarity Report are temporary, and will not reflect your official similarity score that your instructor sees". The score in your instructor's inbox is produced by the assignment's settings, not by yours. If the exclusions look wrong for the kind of writing you were asked to do, the only thing that changes anything is asking your instructor to change them.
Your score may not even be final when you first see it. On assignments that allow resubmission, Turnitin regenerates every report within an hour of the deadline so submissions can be compared against each other, and notes that "it is possible that the report score will change after a collusion check has occurred".
What real universities actually publish
Here the question gets answered properly, and uncomfortably. Twelve institutions, six countries, each quoted from its own published page.
| Institution | Written for | What it says | Source |
|---|---|---|---|
| London School of Economics United Kingdom |
Staff policy | "Principle Four: No similarity score (%) threshold should be used within Departments." | Turnitin policy, v1.1 |
| Ulster University United Kingdom |
Tutors | "When evaluating the level of concern for suspected plagiarism, the Turnitin similarity score is considered as follows: Minor: 10%-15% / Moderate: 16%-30% / Major: >30%" | A tutor's guide to plagiarism |
| University of Technology Sydney Australia |
Students | "there is no specific percentage that is considered acceptable or unacceptable at UTS." Any figure a tutor names is "their personal preference rather than an official UTS requirement." | Using Turnitin |
| Charles Sturt University Australia |
Students | "There is no ideal score! Generally, a score between below 25% could be considered trustworthy, but this still doesn't exclude the possibility of plagiarism" | Interpreting your similarity report |
| Charles Sturt University Australia |
Staff | "Under 5% can be a result of contract cheating being present." And: "Above 25% is required as this is a general threshold to which issues may be present." | Turnitin submission workflow |
| Aberystwyth University United Kingdom |
Students | "There is no ideal percentage to aim for... even a 1% score could potentially indicate plagiarism." | Referencing and plagiarism guide |
| UBC Sauder School of Business Canada |
Instructors | "A 'good' similarity score will be in the range of 20%-30%. If a score is above or below this, you should review the paper in the Feedback Studio." | Best practices for Turnitin |
| UBC Okanagan Canada |
Students | "Sometimes we might hear that anything below, say, 20% means you are in the clear or even that 0% is the goal, but that is not the case." | Academic integrity handout |
| National College of Ireland Ireland |
Students | "there is no perfect Turnitin similarity score... As a guide, somewhere between 15-20% might be considered a good score to aim for." | Library FAQ |
| Gazi University Turkey |
Graduate students | Senate decision 2019/103 set the Turnitin similarity criterion at 20%. Theses scoring above it cannot be assigned an examination jury. | Graduate school notice |
| The Ohio State University United States |
Instructors | "there is no Similarity Report outcome that unequivocally indicates plagiarism." | Turnitin best practices |
| Queen Mary University of London United Kingdom |
Staff | "It is quite common for a department to set a threshold percentage, above which an assignment will be examined more closely." | Turnitin staff guidance |
Three things fall out. Refusal is the majority position: no ideal percentage at Reading, no magic number at Durham, no rule at Western Sydney, no safe level at the Western Institute of Technology in New Zealand.
Where numbers do exist, they live on the staff side of the building. Ulster's bands hang off its misconduct penalty framework. Charles Sturt's 5% and 25% are a marker's triage rule. Gazi's 20% is a gate on forming a thesis jury. A number appears where somebody has to make an administrative decision, not where a student is being taught to write.
And almost everyone who publishes a number hedges it in the same breath, while two contradict themselves outright. Charles Sturt's student guide opens with "There is no ideal score!" and then hands over 25%. De Montfort's examiner guidance says "There is no threshold figure for the similarity score which reliably indicates evidence of plagiarism" and, next sentence, that "a figure below 20% is unlikely to indicate widespread plagiarism". UBC is two documents at one university pointing opposite ways. If you have seen a confident number online, this is the ground it was standing on.
High can be innocent, low can still be plagiarism
A percentage cannot carry a verdict because both tails are ambiguous. A high score is what you would expect from a paper that quotes heavily and carries a long reference list; Turnitin's instructor guide describes a qualitative study scoring 53% on quotations and bibliography alone. Reading's staff guidance handles the low tail: "While a Green score might suggest the document is OK, it is simply an indication of the amount of matched text, so potentially, up to 24% of the document could still have been copied without referencing."
Aberystwyth puts it more sharply for students: "even a 1% score could potentially indicate plagiarism." Oxford Brookes explains the mechanism in the most useful sentence any university has published on this: "Low scores do not necessarily indicate absence of plagiarism, as Turnitin does not find all plagiarism, and essay writing companies are known to produce texts with 0% match."
Which is why some markers read a very low number as a warning rather than a relief. Charles Sturt trains staff to treat anything under 5% as a possible sign of contract cheating, and UBC Sauder tells instructors that "a zero or very low score can indicate issues such as falsified references, the use of paraphrasing tools". If you write honestly about transport emissions or attention and social media, a near-zero score is not the win it looks like. Shape beats size, and National College of Ireland says so plainly: a high score made of small matches to many sources is just good academic writing, while a lower score built from one or two big matches is the one worth worrying about.
The 100% self-match, and how it gets fixed
A score of 100%, or anything close, usually means Turnitin is matching your paper against a copy of your own paper already in its database. This is common, it is not misconduct, and it has a specific fix.
Turnitin excludes self-matches automatically in one situation only. Submit the same work to different assignments inside the same class and a default self-match exclusion hides it. It does not apply across different classes if either assignment was set to save to the standard repository. Resubmitting to the same assignment is safe, because your earlier version is excluded automatically, though the fresh report can take up to 24 hours.
Institutions guard against this in advance. Gazi University tells thesis supervisors to upload with the repository option set to no repository, because otherwise the thesis is stored and, in the school's own words, the similarity ratio comes out higher at the next revision. Turnitin's Draft Coach does not create the problem either: checking a draft there "will not add your document to our repository".
The resolution ladder is short. Your instructor excludes the matching source inside the report, which takes seconds and requires nobody else, and Turnitin explicitly recommends this over deletion. If the paper genuinely has to come out of the repository, only your institution's Turnitin administrator can do it, on a request raised by your instructor with your submission ID, because "Turnitin cannot process deletion requests directly from students".
What to do with your report
Open the report rather than staring at the headline. What is in the Not Cited or Quoted group, and is it genuinely uncited or just cited in a style Turnitin did not recognise? What is your largest single source, and is that match a quotation you attributed? How much of the total is your reference list? And which exclusions did this assignment use, which is a fair thing to ask.
If something in there is genuinely wrong, fix the writing: attribute what you did not attribute, quote what you paraphrased too closely, cite what you leaned on. Oxford Brookes says the rest better than we can: "do not set out with the aim of keeping Turnitin scores low, set out with the aim to write effectively with sources." Our guide on how to avoid plagiarism covers the habits that make this a non-question.
Here is the part that does not flatter us. Writing with a research tool, ours included, will not lower your similarity score, and it should not. Properly cited source material matches by design, and a paper that engages seriously with published work on grid storage or remote work will match more, not less. What good tooling changes is whether your sources are real, whether every claim traces to one, and whether you can point at the passage behind each sentence when somebody asks.
So a good similarity score is one you can explain, line by line, in a five minute conversation. If that is true of your 23%, you are fine. If it is not true of your 6%, the number was never going to save you.