Direct answer
When a paper is submitted to Turnitin, the system runs two simultaneous scans: one for similarity matching and one for AI writing detection. The AI detection scan breaks the document into segments of roughly a few hundred words (about five to ten sentences), overlaps those segments to capture each sentence in context, and runs them against a deep-learning classification model trained on millions of academic and AI-generated documents [1]. Each sentence receives a score between 0 and 1—0 indicates human-written, and 1 indicates fully AI-generated—and the average across all segments produces the overall percentage you see in the AI Writing Report. Turnitin's model is specifically trained to detect outputs from GPT-3, GPT-3.5, GPT-4, GPT-4o, GPT-5, Gemini, Claude, LLaMA, and tools built on these large-language models [1]. The scan is designed for long-form English, Spanish, and Japanese prose, and files must contain at least 300 words of qualifying text (paragraph-based prose, not bullet points, poetry, or code) to generate a report [2].
How Does Turnitin Scan For AI Detection?
Turnitin's AI detection process begins the moment a submission meets the file requirements: the document must be under 100 MB, contain 300–30,000 words of prose text, and be in.docx,.pdf,.txt, or.rtf format [2]. Once these criteria are satisfied, the system splits the document into overlapping segments of text, each spanning roughly five to ten sentences. This overlapping technique ensures that every sentence is evaluated within its surrounding context rather than in isolation [1].
Each segment is then scored by Turnitin's classifier, which was trained on a representative sample of both AI-generated and authentic academic writing across multiple geographies and subject areas. The model specifically accounts for statistically under-represented groups—such as second-language learners and students from non-English-speaking countries—to minimize bias [1]. The classifier examines word-probability patterns: AI-generated text tends to select highly probable next words with consistent frequency, whereas human writing is more idiosyncratic and unpredictable [1]. Sentences that match AI-generation patterns receive a score closer to 1, while human-written sentences score nearer to 0.
The final overall percentage is computed from the average of all segment scores. Importantly, Turnitin's AI detection scan is completely independent of the Similarity Report—a document can have a 0% similarity score but a high AI detection score, or vice versa [2]. The scan also includes two additional detection layers for English submissions: AI-paraphrased text detection (identifying content run through tools like Quillbot) and AI-bypasser detection (identifying text passed through humanizer tools), each highlighted in different colors within the report [2].
What Does a Turnitin AI Detection Report Show?
The AI Writing Report presents an overall percentage at the top of the interface, indicating how much of the qualifying text in the submission is likely AI-generated. Below this percentage, the Submission Breakdown categorizes flagged text into two detection categories: "AI-generated only" (highlighted in cyan) and "AI-generated text that was AI-paraphrased" (highlighted in purple) [2]. An interactive submission breakdown bar runs across the top of the document pages, allowing instructors to click on any color segment to jump directly to the corresponding flagged passage.
The report also handles scores below 20% with a specific display rule: any AI detection score between 0% and 19% is shown as an asterisk (*%) rather than a numeric value. This design choice reduces the risk of misinterpretation, as Turnitin's own testing found a higher incidence of false positives in this range [2]. Only a confirmed 0% or scores of 20% and above display as exact numbers. The indicator can also show states such as "Loading" (report processing), "Processing error" (resubmission needed), or "File didn't meet the requirements" (document below the 300-word threshold) [2].
Instructors can access the AI Writing Report directly from the Similarity Report sidebar in Turnitin Feedback Studio. The flagged highlights are visible only in the AI Writing Report view, not overlaid on the Similarity Report, keeping the two detection systems visually separate [2]. Sentence-level granularity means instructors can see exactly which passages were flagged, making it possible to have informed conversations with students about specific sections of their work. Turnitin emphasizes that the AI percentage should not be used as the sole basis for academic misconduct decisions—it provides data for educators to make informed judgments within their institution's academic policies [1].
Is It Possible To Preview a Turnitin AI Detection Scan Before Submitting?
In most institutional setups, students cannot independently run a Turnitin AI detection scan before submitting to an official assignment. Turnitin's AI detection is a feature enabled by institutions and is accessible to instructors through the Similarity Report workflow—students do not have a self-service button to trigger an AI scan on their draft [1]. If a student's institution provides Turnitin Draft Coach, they can check similarity within Google Docs or Microsoft Word, but Draft Coach does not include the AI writing detection feature in the same way the full Feedback Studio report does.
This creates a practical challenge: students who want to know how their draft would score on Turnitin's AI detector before the final submission have limited options within the official system. Some instructors may set up practice assignments with resubmissions enabled, allowing students to upload drafts and receive reports (including AI detection if enabled on that assignment). However, even in these cases, Classic Standard Assignments limit rapid resubmissions—after the third attempt, a 24-hour waiting period applies before the next report generates [2].
Because the official Turnitin system does not offer a direct student-facing preview for AI detection, many students turn to third-party services that replicate the same detection model. These services allow users to upload a document and receive an AI detection report with the same percentage scoring and highlighting that instructors would see, enabling students to understand their AI score before the assignment deadline. The key advantage is transparency: rather than submitting blind and hoping the score is low, students can verify whether their writing—whether human-written, AI-assisted, or a mix—falls within acceptable parameters.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Can Turnitin detect AI-written content in any language?
Turnitin's AI detection supports long-form English, Spanish, and Japanese submissions. Documents submitted in other languages will not be processed, and the indicator will display an empty or error state with guidance about supported languages [1]. The AI-paraphrased and AI-bypasser detection layers are currently only available for English submissions.
How accurate is Turnitin's AI detection scan?
Turnitin reports that its model maintains a false positive rate of less than 1% for full documents under controlled testing conditions. However, the model may misidentify text at the sentence level, particularly in the 0–19% range where scores are displayed as asterisks (*%) rather than exact numbers [2]. Turnitin explicitly states the indicator should not be used as the sole basis for academic misconduct decisions.
Does a high similarity score mean high AI detection too?
No. The similarity score and the AI detection percentage are completely independent metrics. The similarity score measures text matches against Turnitin's database of web pages, publications, and student papers, while the AI detection score measures whether text was generated by a large-language model [1]. A paper can have 0% similarity but 100% AI detection, and vice versa.
What file types are accepted for an AI detection scan?
Turnitin accepts.docx,.pdf,.txt, and.rtf files for AI detection processing. The file must be less than 100 MB, contain at least 300 words of prose text in a long-form writing format, and not exceed 30,000 words [2]. Files with bullet points, tables, poetry, scripts, or code may not produce reliable results.
Can I check my own paper for AI detection before my instructor runs the scan?
In most institutional setups, students cannot self-initiate a Turnitin AI detection scan. The feature is instructor-facing and integrated into the assignment submission workflow. Students who want to preview their AI score before the official submission need either an instructor-created practice assignment with resubmissions enabled or a third-party service that provides Turnitin-compatible AI detection reports.