What AI Percentage Is Acceptable on Turnitin?

There is no acceptable percentage published by Turnitin. No pass mark, and no figure anywhere in its documentation that separates acceptable from unacceptable. Turnitin does publish a display rule: results from 1% to 19% appear as an asterisk with no percentage and no highlights, and 20% is where a figure starts being shown. That boundary describes what appears on a screen. It gets quoted as though it were a pass mark, which it is not. The number that actually governs your submission is written by your institution, and Turnitin puts the final call on misconduct with the person marking, which is one reason two markers can answer this differently. If the figure you were given sits below twenty, the report cannot show anyone whether you met it.

So the answerable version of the question is not "what number is safe". It is "which document defines mine, and what happens when it is crossed". The rest of this covers where we went looking for a vendor figure, where yours is likely written down, and what to ask for before a deadline rather than after a flag.

A claim that something does not exist is worth exactly as much as the search behind it, so here is ours.

Turnitin's help centre has a section called AI writing. It contains nine articles, and on 18 August 2026 we opened all nine and searched the rendered text of each.

The nine articles run to roughly 58,000 characters. Every percentage in that text belongs to one of two conversations: the display rule (0, 19, 20) and the vendor's own error reporting (the under-1% false positive target, and the 50/65 illustration of how much AI writing it is willing to miss in exchange). Two of the nine pages also carry screenshots of the interface, and those do contain other figures, but they are sample reports: one shows a submission sitting at 56% with a similarity score of 23% next to it. Nothing in the text or the screenshots is presented as a limit for a student's paper.

We also used the help centre's own search. Queries like "acceptable percentage" and "what percentage of AI is acceptable" return dozens of results, because the search matches on individual words rather than the phrase. The top-ranked article for the second query is titled "What should I do if the AI writing percentage is high?", and its first instruction is: "First, refer to your institution's policy on plagiarism and the use of generative AI in the classroom."

The page aimed most directly at a high percentage answers by pointing somewhere else. That, rather than any absence of documentation, is the finding.

That page is short, and it is written for your marker rather than for you. We went through both of the educator-facing pages line by line in what your instructor is told to do when your AI percentage is high.

Somebody local. Which local somebody depends on where you study, so treat the following as a search order rather than a description of how universities are organised.

Two things follow. Turnitin's own wording puts the decision with the reviewer or instructor rather than with the tool, so a rule that lives at course level is not an oddity, and two markers in one department can answer differently without either of them being inconsistent. And "there is no number, we look at flagged work case by case" is a complete answer rather than an evasion. It tells you the trigger is the flag itself, which changes what you should prepare.

Numbers circulate between students without their origin attached. Fifteen. Twenty. Thirty. Before you treat one as your AI limit, read the sentence it came from and see which report it names, because Turnitin runs two separate analyses and says so plainly: "The Similarity score and the AI writing detection percentage are completely independent and do not influence each other."

Elsewhere it adds that AI writing highlights are not visible in the Similarity Report at all. Two reports, two sets of highlights, no arithmetic connecting them.

The tell is the wording around the figure. Similarity, matching text, sources, originality: those belong to the other report, and a limit written for them says nothing about the AI indicator. If the handbook does not specify, that is a question worth asking rather than a gap to fill by assumption. The two reports and the four score combinations they can produce are laid out in Turnitin AI report vs similarity report.

Now suppose your course does give you a figure, and it sits below twenty. Ten, say.

"To avoid potential incidence of false positives, no score or highlights are attributed for AI detection scores in the 1% to 19% range. When AI is detected below the 20% threshold in the report, it is now indicated with an asterisk (*%) and no percentage is attributed."

Read that against a ten percent rule. A submission scoring 8 and a submission scoring 18 produce the same thing on screen: an asterisk, no figure, no highlighted passages. There is no display state that distinguishes them. So nobody in the process, including the person marking, can tell from the report which side of ten a paper landed on.

Add the second constraint and the loop closes. Turnitin states that "only instructors and administrators are able to see the indicator" and that the indicator and report "are not visible to students", while noting that an instructor can download the report as a PDF and share it. You cannot check the figure yourself, and in that band there is no figure to check.

Which is why the productive question to put to a course leader is not "what percentage is fine". It is: what happens when the indicator comes back as an asterisk, and what happens when it comes back with a number on it? Those are the two states the report can actually deliver, so those are the two the answer has to cover.

At 20% or above, the same figure on two documents does not describe the same thing, because the denominator moves. Turnitin scores what it calls qualifying text, and defines it narrowly: "This qualifying text includes only prose sentences, meaning that we only analyze blocks of text that are written in standard grammatical sentences and do not include other types of writing such as lists, bullet points (short non-sentence structures), or other non-sentence structures." The sentence immediately after is the one to keep: "This percentage is not necessarily the percentage of the entire submission."

A departmental figure applied to a lab report that is half tables, and applied to a coursework essay that is prose from the first line to the last, is measuring two different fractions of two very different documents. What each percentage actually covers, and why the highlight count rarely matches it, is the subject of how to read a Turnitin AI writing report.

There is a further reason a low local threshold is a strange object. Turnitin's own accuracy claim is scoped: it says it works to keep its false positive rate "under 1% for documents with over 20% of AI writing". That wording excludes exactly 20%, even though the report starts showing a percentage at 20%. A rule below 20% sits outside the stated scope of that accuracy claim and in the range where the report stops printing numbers.

One short message to the module leader or supervisor, before the deadline, beats any amount of guessing afterwards. Five things worth having in writing:

• Which document governs this assessment, and can you link it? If two documents disagree, which one wins. This single question replaces most of what people try to infer from other students' scores.
• Is the figure in it about the similarity score or the AI writing indicator? They are separate analyses, and handbooks are not always explicit about which one a number belongs to.
• What happens at that figure: a conversation, a referral, a mark penalty, or nothing automatic? The consequence tells you far more than the number does about how seriously to treat it.
• What happens when the report comes back with an asterisk rather than a figure? Below 20% that is all anyone sees, so it is worth knowing whether it triggers anything on your course.
• Does permitted use have to be declared, and in what form? Where declaration is required, an undeclared use is the problem regardless of what the detector reported.

If you are asking after a flag rather than before one, the first hour matters more than the wording, and we set out what to do with it in appeal a false AI detection flag.

At 20% or above, the report stops being a symbol and starts being a document: a figure, plus highlighted passages you can open and read. That is the only state that gives you something specific to work on, and it is the state HumanPen takes as its input. Upload the DOCX together with the Turnitin or iThenticate AI report, and the highlighted passages define what gets touched. Nothing outside them is edited, so the paragraphs nobody marked keep the wording you already wrote, and what you pay for is the words that actually changed rather than the length of the document you handed over.

HumanPen's public page also avoids promising a fixed score. It says detectors keep changing and tells you to import a fresh report with the rewritten document for a second targeted pass if residual passages are still flagged. The new report, rather than a predicted score, defines the next scope.

That does not turn 20% into a standard. It remains Turnitin's display boundary, with nothing to say about what your department will accept. We will not tell you what a future report will say, either. What the public product page does state in advance is which passages get touched and that billing counts the words actually rewritten.

Does Turnitin publish an acceptable AI percentage? No. We read all nine articles in the AI writing section of its help centre on 18 August 2026 and the word "acceptable" does not appear in any of them. The documentation publishes figures about display behavior and model performance, including the 20% display boundary, but none is presented as an acceptable limit. The page aimed at a high percentage tells instructors to consult their institution's policy.

Is 20% the pass mark, then? It is the point at which the report begins showing a number. Results from 1% to 19% are shown as an asterisk with no percentage and no highlights. Whether 20% means anything on your course is set by your course, not by that display rule.

My handbook says 15%. Is that my AI limit? Check which report the sentence is about. Turnitin states that the similarity score and the AI writing percentage are completely independent and do not influence each other, so a limit written for one says nothing about the other. If the handbook does not specify, ask.

Can I see my own AI percentage? Not through the indicator. Turnitin says only instructors and administrators can see it and that it is not visible to students, though an instructor can download the report as a PDF and share it.

If my course allows AI use, does a high percentage still matter? That depends on what your course asked you to declare. Where permitted use has to be declared, the declaration is what the rule turns on, and Turnitin says its own indicator does not determine misconduct in any case.

Who decides in the end? Turnitin's guidance puts it with the person marking: the final decision on whether misconduct has occurred rests with the reviewer or instructor, and the tool provides data rather than a determination.