Direct answer
Turnitin's AI paper detector is a machine-learning classifier that estimates how much of a submitted paper's prose was likely generated by an AI model, and it reports that estimate as a percentage of qualifying text rather than as a verdict about the student. It analyses long-form academic prose, excludes reference lists and short quoted passages from the qualifying word count, and flags text at the sentence level so an instructor can inspect the passages behind the number [1]. Crucially, Turnitin itself presents the score as a signal for human judgement, not as proof of misconduct [1].
How Does the Turnitin AI Paper Detector Actually Decide Whether a Paper Is AI-Generated?
The detector does not search for a database of known AI sentences. It is a trained classifier that scores statistical and structural properties of your writing — patterns in word choice, sentence rhythm, and the predictability of phrasing — and estimates the probability that a passage was machine-generated rather than composed by a person [2]. Because it works on probability rather than matching, the output is always a percentage with an attached confidence level, never a binary "AI" or "human" stamp [2].
The percentage you see is calculated against qualifying text only. Turnitin's report defines the denominator as the prose long enough for the model to analyse, which means a paper with extensive reference lists, block quotations, tables, and headings can produce a score based on noticeably fewer words than the total word count [2]. A 40% AI score on a 3,000-word essay therefore does not mean 1,200 words were copied from ChatGPT; it means 40% of the analysable prose was flagged [2].
The report is also designed to be read at the sentence level. Flagged passages are highlighted so the reader can compare the flagged sentence with the surrounding unflagged text and judge whether the flag is plausible [2]. This is why Turnitin's own documentation instructs instructors not to treat the overall figure as a standalone finding and to consider the highlighted evidence alongside drafts, version history, and conversation with the student [2].
Finally, the model is validated continuously against new generative systems, and Turnitin publishes a low false-positive rate for its classifier while acknowledging that no detector is perfect [1]. Practically, that means the detector is strongest on clearly machine-drafted prose and least reliable on heavily edited, mixed-authorship text — exactly the category most students produce when they use AI for outlining and then rewrite by hand [1].
What Does a Turnitin AI Detection Report Show, and How Should You Read the Percentage?
A Turnitin AI detection report contains an overall AI score, a breakdown of qualifying versus total words, sentence-level highlighting, and a standing disclaimer that the result should not be used as the sole basis for an academic integrity decision [3]. Turnitin describes the feature as one input among several and positions it inside the instructor's existing assignment workflow rather than as a public self-service tool [3]. That framing matters: the report is evidence to be interpreted, not a certificate.
Reading the percentage correctly depends on three things. First, check the qualifying word count — if only 60% of your paper was analysable, the score describes that 60%, not the whole document [3]. Second, look at where the highlights fall; a concentrated block of flagged text in one section reads very differently from scattered single sentences across a literature review [3]. Third, note that when the detector's confidence falls below its reporting threshold, Turnitin displays an asterisked score such as *% instead of an exact figure, which signals a low-confidence result rather than a clean bill of health [3].
Turnitin also states that its detection covers output from the major generative models, including ChatGPT, Claude, and Gemini, and that the classifier is retrained as those models change [3]. This is why a paper that scored low six months ago can score differently today: the model, not just the text, has moved [3]. Students who assume a single past result is permanent are reading the report as a static document when it is actually a snapshot of one model version at one moment [3].
Can You Check a Paper With Turnitin's AI Detector Before You Submit It?
Not through Turnitin directly, in most cases. Student access to Turnitin is normally provisioned by the instructor through the institution's learning management system, which means the AI Writing Report is generated when you submit to the assignment — not on demand beforehand [4]. Turnitin's student-facing guidance describes access as something granted by the course, not something a student can self-serve [4]. There is no official public upload page where any student can run an AI check on a draft.
That gap is why independent pre-submission checking services exist. A third-party service such as turnitin0.com accepts a .docx, .pdf, or .txt file, runs it through a Turnitin check without adding it to the student paper repository, and returns the AI detection report together with the similarity report as downloadable PDFs — the same two outputs an instructor sees in the LMS [4]. Because the file is not deposited in the institutional database, the preview does not create a match against your own later submission, which is the main technical risk students worry about when they consider self-checking [4].
Two practical caveats are worth stating plainly. A preview reflects the detector's behaviour at the time you run it, so if you make substantial edits afterwards, the earlier score no longer describes your current draft [4]. And a preview is diagnostic, not protective: seeing a high score tells you which passages to revise, but the underlying cause — machine-like phrasing, uniform sentence length, generic transitions — has to be fixed in the text itself before the next check [4].
If you would rather see the actual report than guess at your risk, turnitin0.com gives you the same AI score, sentence-level flags, and similarity summary your instructor will open — so you know exactly which paragraphs to rewrite before the deadline, not after it.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is a Turnitin AI score the same as proof that I used AI?
No. Turnitin presents the score as an indicator that requires interpretation, and its report carries a disclaimer against using it as the sole basis for a misconduct finding [3]. The sentence-level highlights are the part an instructor is expected to examine.
Why does my score show an asterisk instead of a number?
Turnitin displays *% when the detector's confidence falls below its reporting threshold, which indicates a low-confidence signal rather than a confirmed result [2]. It is not equivalent to zero AI text.
Does Turnitin check the whole paper or only part of it?
Only qualifying prose is analysed. Reference lists, short quotations, and other non-prose elements are excluded from the denominator, so the percentage describes the analysable portion of your document [1].
Can I run my own paper through Turnitin before submitting?
Turnitin access is normally granted by your instructor through the course, so there is no official self-service check [4]. Independent services can produce a pre-submission preview without adding your file to the student paper repository [4].
Do detectors keep changing after I graduate?
Yes. The classifier is retrained as generative models evolve, so the same text can score differently across model versions [1]. Treat any single score as a snapshot rather than a permanent property of your writing.