Direct answer
Turnitin's AI content detector is a classifier built into the Turnitin platform that estimates how much of a submitted document was likely generated by an AI writing model, and it reports that estimate as a percentage of qualifying prose [1]. It analyses sentence-level text rather than whole documents, excludes reference lists and bibliographies from the calculation, and highlights the specific segments it believes were AI-generated [1][2]. The score is a probabilistic signal for instructors, not a verdict, and Turnitin itself acknowledges that false positives are possible [1].
How Does Turnitin's AI Content Detector Actually Decide What Counts as AI-Generated?
Turnitin's detector does not look for "AI words" or a plagiarism match. It runs a classification model over your prose and returns a likelihood that each analysed segment was machine-generated, which is why the output is expressed as a proportion of your document rather than as a single yes-or-no flag [1]. Because the model works on qualifying text, only sentences long enough to be statistically meaningful are assessed — short headings, bullet fragments, and table cells typically fall outside the calculation [2]. That structural detail matters: a paper padded with short bullet points can show a lower AI percentage than the same argument written in full paragraphs.
The second thing to understand is that the detector is deliberately narrow in scope. Turnitin states that its AI writing detection capabilities are designed to identify text likely produced by AI models and to surface it for human review, not to adjudicate misconduct [1]. Reference lists, bibliographies, and quoted material are excluded from the qualifying-text count, so a long Harvard or APA reference section will not inflate your score [1]. What remains — your introduction, analysis, and conclusion — is where the percentage is actually earned.
Third, the model is probabilistic and Turnitin is explicit about the limits. Its own FAQ documentation notes that detection can produce false positives and that results should be interpreted alongside a conversation with the student rather than as standalone proof [1]. In practice this means two students can submit genuinely human writing and receive different scores, and a heavily edited AI draft can score lower than an unedited one. The report tells you where the model's attention landed; the interpretation is still a human judgement call, which is exactly how the AI Writing Report is designed to be read [2].
What Does the AI Detection Percentage Mean, and Why Does Turnitin Sometimes Show an Asterisk Instead of a Number?
The headline percentage in the AI Writing Report represents the share of qualifying prose the model classified as likely AI-generated, and the report pairs that number with sentence-level highlighting so you can see precisely which passages drove it [2]. The highlighting is the useful part: it distinguishes a document where one paragraph is flagged from a document where the entire argument is flagged, even when the two carry a similar overall figure. Turnitin builds the report so that flagged segments can be reviewed in context rather than treated as isolated verdicts [2].
The asterisk is the detail most students misread. When the AI score falls below Turnitin's 20% confidence threshold, the report displays an asterisk (*%) in place of an exact percentage, because the signal is too weak to report as a confident figure [3]. An asterisk is therefore a low-confidence result, not a zero and not a clean bill of health — the underlying segments may still be visible for instructor review [3]. Treating *% as "safe" is the single most common misreading of the report.
A high percentage deserves the same care in the opposite direction. Because the model assesses qualifying prose only and works probabilistically, a high score on a genuinely human draft usually points to writing that reads as generic, formulaic, or unusually uniform in sentence structure [1][2]. That is a stylistic diagnosis you can act on. What the percentage cannot tell you is intent, and neither Turnitin nor your instructor should be asked to infer intent from a number alone [3].
Can You See Your Turnitin AI Score Before You Submit, and What Should You Do If It Comes Back Flagged?
In most institutions, students cannot run an official Turnitin AI report on their own draft. Access to the AI Writing Report is controlled by the instructor and the institution's settings, so whether you can see a score before the deadline depends entirely on how your course is configured [4]. Turnitin's own student-facing guidance is built around this gap: it advises students to keep drafts and revision history, because a documented writing process is the strongest available evidence of authorship if a report is ever questioned [4].
If a report does come back flagged, the productive response is revision, not denial. Turnitin's guidance for students emphasises rewriting flagged passages in your own voice — restating the argument in your own sentence structures, adding specific evidence and citations, and removing the smoothed-over phrasing that machine-generated prose tends to produce [4]. Keep the earlier draft alongside the new one; the diff between them is exactly the kind of authorship evidence that supports a conversation with your instructor [4].
The practical problem is timing. Students who only discover a flagged score after the official submission have no room left to revise, which is why pre-submission checking has become standard practice in the UK, US, Canada, Australia, New Zealand, and Ireland. A preview report gives you the same two documents your instructor sees — the AI detection report and the similarity report — while you can still rewrite the flagged paragraphs and resubmit the improved version. That is a different activity from guessing at your score, and it is the only version of the process where the result is still actionable [4].
If you want to see where Turnitin's detector actually lands on your draft — the same report cover, AI percentage, sentence flags, and similarity summary your instructor opens — turnitin0 runs that check for you before the deadline, so a flagged paragraph becomes a revision task rather than a post-submission problem.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is Turnitin's AI detector accurate enough to accuse someone of using AI?
No — Turnitin positions its AI writing detection as a signal for human review, and its own documentation acknowledges that false positives occur [1]. A score should start a conversation or prompt a revision, never function as standalone proof of misconduct [1][3].
Why does my AI score change between two versions of the same essay?
The percentage is calculated over qualifying prose only, so edits that change paragraph length, sentence structure, or the balance of bullet points to full sentences shift the analysed text and therefore the score [2]. Rewriting flagged passages in your own voice is the reliable way to move it [4].
Does a low AI percentage or an asterisk mean my work is cleared?
Not automatically. An asterisk appears when the score sits below Turnitin's 20% confidence threshold, which signals low confidence rather than a confirmed low score [3]. Flagged segments can still be visible to your instructor for review [3].
Does Turnitin check my reference list and bibliography for AI content?
No. Reference lists and bibliographies are excluded from the qualifying-text count, so citations do not inflate your AI percentage [1]. The prose you actually wrote — introduction, analysis, conclusion — is what the model assesses [1][2].
Can I check my own draft with Turnitin before submitting it?
Official report access is controlled by your instructor and institution, so student self-service is not guaranteed [4]. Many students use an independent pre-submission check to see the AI and similarity reports while the draft can still be revised [4].