Direct answer
A Turnitin percentage is a measurement of overlap or classification, not a verdict. The Similarity Score tells you what share of your document's text matched sources in Turnitin's comparison database, while the AI Writing Score tells you what share of your qualifying prose the detector classified as AI-generated [1]. Neither number is a plagiarism finding or a misconduct ruling — both are signals that a human reader interprets in context, and a figure that looks alarming on its own often turns out to be a properly cited quotation or a reference list [2].
What Does the Similarity Percentage in Turnitin Actually Mean?
The similarity percentage is the proportion of your submission's text that Turnitin matched against its comparison database, which includes internet pages, academic journals, and previously submitted student papers [1]. It is calculated from the total matched text, so a 24% score means roughly a quarter of your document's word count was found somewhere else. Crucially, Turnitin itself describes this as a similarity index rather than a plagiarism score: matching text may be correctly attributed, and the report exists to show where the match occurs so an instructor can judge whether attribution is adequate [1].
The number is also an aggregate, which is why it can be misleading in isolation. A single long block quotation with a full citation, or an intact reference list, can push the total into double digits without any integrity problem at all [2]. This is why the match overview matters more than the headline figure: it breaks the total into individual sources, each with its own percentage, so you can see whether 18% comes from one over-cited paragraph or from fifteen scattered three-word phrases [2].
Exclusions change the reported figure, and this is the single most common source of confusion. When an instructor or a student applies settings that exclude quotations, bibliography, or small matches below a word threshold, the score recalculates and can drop substantially [1]. Two people looking at the same submission can therefore see two different percentages and both be correct — they are reading reports generated under different exclusion settings [2].
Finally, a low similarity score is not evidence of original work. Text that has been paraphrased heavily, translated, or generated by a language model may match very little in the database simply because it does not exist anywhere else yet [1]. Reading a 2% score as proof of originality mistakes the absence of matches for the presence of authorship.
What Does the Turnitin AI Writing Percentage Mean, and Why Does It Sometimes Show an Asterisk?
The AI writing percentage answers a different question entirely: of the prose segments Turnitin's detector evaluated, what share did it classify as AI-generated [3]. It is not a measure of quality, effort, or intent, and it is calculated only over qualifying prose — reference lists, headings, short fragments, and quoted material are typically excluded from the evaluated word count, so the denominator is smaller than your total document [3]. A 60% AI score therefore does not mean 60% of your essay was written by ChatGPT; it means 60% of the detected, qualifying text was flagged.
The asterisk is the detail most students misread. When Turnitin's detector returns a result below its 20% confidence threshold, the report displays an asterisk in place of a precise figure, signalling a low-confidence signal rather than a confirmed finding [3]. Seeing *% on a report is not the same as seeing 0%, and it is not an accusation — it means the system's evidence was not strong enough to commit to a number, and the underlying segments should be reviewed individually rather than judged by the headline [3].
Segment-level detail is where the AI score becomes genuinely useful. The report highlights the specific sentences the detector flagged, allowing you to compare flagged passages against your own drafting history and determine whether the classification reflects reality [3]. Because detectors work on statistical patterns in word choice and sentence structure, formulaic academic phrasing, heavily templated methodology sections, and discipline-specific boilerplate can all raise a score even in writing produced entirely by hand.
The two percentages are independent and often diverge, which is exactly what confuses readers. A paper can show 30% similarity and *% AI, or 4% similarity and 85% AI, because one measures textual overlap with existing sources and the other measures statistical patterning [1][3]. Treating them as a single "Turnitin score" leads to the wrong remedy: rewriting to lower similarity will do nothing for an AI flag, and humanizing prose will not change a match against a journal article.
What Percentage Is Considered Acceptable on a Turnitin Report?
There is no universal acceptable percentage, because Turnitin does not set thresholds — institutions and individual instructors do [4]. Turnitin's own guidance is explicit that the similarity score should be interpreted by a human in the context of the specific assignment, and that a high score is not automatically a problem nor a low score automatically a clean bill of health [4]. Any article quoting a fixed "safe" number is describing a local policy, not a global standard.
In practice, thresholds vary widely by discipline and task. A literature review built from extensive quotation may legitimately sit far higher than a lab report, and a reflective portfolio may sit lower than a source-heavy essay, so the same number can be unremarkable in one module and worth a conversation in another [4]. What matters is whether each match is attributed: an instructor reading the match overview is checking whether cited material is cited correctly, not whether the total is under an arbitrary line [2][4].
The AI percentage carries a similar caveat but a different risk profile. Because detection is probabilistic and threshold-dependent, a flagged score is best treated as a prompt to review the highlighted segments and be ready to explain your drafting process, rather than as a fixed pass/fail gate [3][4]. Many institutions require staff to review AI indicators alongside a conversation with the student precisely because the number alone is not dispositive [4].
The practical takeaway is to read the report, not the headline. Open the match overview to see which sources contribute, check whether exclusions were applied, review the AI-flagged segments, and compare the result against your module handbook or ask your instructor what threshold applies [2][4]. A percentage becomes meaningful only once you know what it is measuring and what rule is being applied to it.
Reading a percentage is one thing; seeing your own draft's real numbers — the match overview, the flagged segments, and the AI figure exactly as an instructor's system renders it — is what actually settles the question before you submit. Turnitin0 generates that same pair of reports on your own document, so you can inspect every contributing source and every highlighted passage while there is still time to act on it.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does a high similarity percentage mean I plagiarised?
No. The similarity score measures matched text, including correctly cited quotations and reference lists, so a high figure can be entirely legitimate [1]. Whether there is a problem depends on whether each match is properly attributed, which is what the match overview shows [2].
Why does my Turnitin AI score show an asterisk instead of a number?
An asterisk appears when the detector's result falls below its 20% confidence threshold, meaning the signal was too weak to report a precise percentage [3]. It is a low-confidence indicator, not a confirmed finding, and the flagged segments should be reviewed individually.
Can two people see different percentages for the same essay?
Yes. Exclusion settings for quotations, bibliography, and small matches recalculate the score, so reports generated under different settings legitimately show different totals [1][2]. Always check which exclusions were applied before comparing figures.
Is there a percentage I should aim to stay under?
There is no universal threshold, since institutions and instructors set their own expectations and interpret scores against the assignment type [4]. Check your module handbook or ask your instructor rather than relying on a number quoted online.
Do the similarity and AI percentages measure the same thing?
No. Similarity measures textual overlap with sources in the comparison database, while the AI score classifies qualifying prose as AI-generated based on statistical patterns [1][3]. They are independent, which is why one can be high while the other is low.