Direct answer
Turnitin's AI detector checks long-form, prose-style academic writing for statistical patterns that resemble machine-generated text, then reports the share of the document it considers likely AI-generated as a percentage. It does not check who wrote the text, and it does not produce a verdict — it produces a signal your instructor interprets [1]. It only works on English documents of roughly 300 words or more, and it analyses sentence-level predictability rather than individual words [1][2].
What Exactly Does Turnitin's AI Detector Analyze in a Document?
Turnitin's AI writing detection is built for long-form, prose-style writing such as essays, reports, and dissertations, and it is deliberately not designed for short answers, bullet lists, or code [2]. Within that scope, the model looks at how predictable and uniform the writing is across the whole submission, reading the text at sentence level rather than scanning for individual "AI words" [2]. This matters because there is no single keyword that triggers a flag — a passage is judged by how its structure and word-choice variation compare with patterns the model associates with generated text [2].
The report itself is granular: it returns an overall AI percentage for the document plus sentence-level highlighting of the segments the model considers likely AI-generated [2]. Segments that fall below Turnitin's confidence threshold are excluded from the calculation rather than counted, which is why two drafts with similar wording can produce different percentages [2]. The detector also requires a minimum length — around 300 words — and supports English only, so very short or non-English submissions simply will not return a meaningful AI score [1].
One practical consequence is that heavy editing does not automatically erase a signal. If you drafted with an AI assistant and then reworded a few sentences, the underlying predictability patterns can survive, and the flagged segments in the report will show you exactly where [2]. That is also why Turnitin frames the output as a set of highlighted segments to review, not a single pass/fail number [2].
What Does the Turnitin AI Score and Percentage Actually Mean?
The headline percentage is the share of the submission that the model flags as likely AI-generated — it is not a probability that the student cheated, and it is not proof of misconduct [3]. Turnitin is explicit that the score is one indicator among many and should be interpreted alongside the highlighted text, the assignment context, and any conversation with the student [3]. Reading the number as a cheating probability is the single most common misunderstanding of the report [3].
Turnitin also applies a confidence threshold: when the AI score falls below its 20% confidence level, the report displays an asterisk (*%) instead of an exact figure, signalling a low-confidence result [3]. That asterisk is not a hidden high score — it means the detector does not have enough signal to report a reliable percentage [3]. Institutions are advised to treat both the percentage and the asterisk as starting points for review rather than conclusions [3].
False positives are a real possibility, and they tend to cluster around formulaic or heavily templated prose — the kind of writing that is genuinely predictable even when a human wrote every word [3]. Conversely, a low or asterisked score does not certify that text was human-written; it only means the model did not find strong enough patterns to flag [3]. Understanding this asymmetry is what lets you use the report sensibly instead of panicking over a single number [3].
How Can I See What Turnitin Would Report for My Own Draft?
The most reliable way to know what Turnitin would say about your draft is to see the same style of report before you submit it. Pre-submission checking services return an AI detection report alongside a similarity report, showing an overall AI percentage and the specific passages flagged as likely AI-generated [4]. Seeing the highlighted segments turns an abstract score into something actionable: you can revise the flagged passages rather than guessing which sentences caused the flag [4].
When you preview a draft this way, look for two things. First, the overall AI percentage and whether it sits above or below the confidence threshold, since a sub-threshold result is reported differently [4]. Second, the sentence-level highlights, because those show precisely where the writing reads as machine-like — often the most formulaic transitions and summary sentences rather than the analytical core [4].
It also helps to choose a non-repository check. A non-repository submission is compared without being added to Turnitin's student paper database, so your draft does not become a future match source for other students and your report is not shared with third-party databases [4]. That keeps a pre-submission preview a private revision step rather than a permanent record [4].
If you want to see your own flagged passages before your deadline rather than after, turnitin0 gives you the same AI and similarity reports your instructor would see — so you can revise with evidence instead of guesswork.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does Turnitin check for AI in every assignment?
No. AI writing detection only runs when the instructor enables it for that assignment, so a missing AI score usually means the feature was not turned on rather than that your text was clean [2].
Is there a minimum word count for Turnitin's AI detection?
Yes. The detector is designed for long-form prose and needs roughly 300 words of English text to return a meaningful result, which is why short answers rarely produce a usable AI percentage [1].
What does an asterisk (*%) in the Turnitin AI report mean?
It means the AI score fell below Turnitin's 20% confidence threshold, so the report shows *% instead of an exact figure — a low-confidence signal, not a concealed high score [3].
Can a fully human-written essay still be flagged by Turnitin?
It can. Formulaic or heavily templated prose is genuinely predictable, and predictable writing is what the model looks for, so false positives are possible and the score should be read as one indicator among many [3].
Can I see my Turnitin AI report before I submit?
Yes. A pre-submission check returns an AI detection report and a similarity report for your draft, and a non-repository check keeps it out of Turnitin's student paper database [4].