Your supervisor mentions that the AI indicator on your chapter reads 34%. Does that mean a third of your thesis is considered machine-written? Not in the way the number suggests. The percentage refers only to the prose sentences the model could analyze, it comes from a probabilistic classifier that Turnitin itself says can be wrong, and Turnitin tells instructors not to treat it as the sole basis for any action.
This article is part of our guide How to Write a Master’s Thesis by the Mastermind PhD team and relates to our Paraphrasing & Plagiarism Check service. Sources last checked: November 2026.
This guide explains Turnitin AI detection from Turnitin’s own documentation: what the indicator measures, how it is calculated, why low scores are hidden behind an asterisk, where false positives come from, and what you should keep as evidence of your own writing.
What is Turnitin’s AI writing detection? A feature of the Similarity Report, available to institutions that license Turnitin Originality, that estimates how much of the qualifying text in a submission was likely generated by a large language model, or generated and then modified by an AI paraphrasing or bypasser tool. It produces an overall percentage, highlights the sentences concerned, and is visible to instructors and administrators but not to students. Turnitin states that the model “may not always be accurate” and “should not be used as the sole basis for adverse actions against a student”.
What does the percentage actually measure?
Only “qualifying text”: prose sentences inside a long-form piece of writing. Bullet points, tables, lists, code, poetry and annotated bibliographies are not analyzed, so the percentage is not a share of the whole document, and Turnitin notes that a file with mixed writing types shows a mismatch between the percentage and the highlighted area. The Turnitin AI detection score is also independent of the similarity score: the two do not influence each other, and AI highlights never appear in the similarity view.
How does Turnitin AI detection calculate the score?
Sentences are extracted and segmented into overlapping sections; each segment is classified and given a value between 0 and 1 for the probability of being AI-generated; each sentence inherits the scores of its segments, which are pooled, and the sentence scores are aggregated into a document score.
The classifier is a transformer-based deep learning model trained on academic writing and AI output, and Turnitin says it is not programmed around single metrics such as “perplexity” or “burstiness”, so individual predictions cannot always be explained feature by feature. English, Spanish, Japanese, Arabic, Dutch, German and Turkish long-form text are supported; paraphrase and bypasser detection exists only for English.
| What the indicator shows | Turnitin’s explanation |
|---|---|
| 0% | No qualifying text identified as likely AI-generated |
| *% | A score between 1% and 19%: not displayed because false positives are more frequent in this range (for reports generated since 8 July 2024) |
| 20% or more | Share of qualifying text likely AI-generated, or AI-generated and then AI-paraphrased, with highlights |
| Grey, no percentage | Not processed: the file falls outside the requirements or was submitted before the feature was enabled |
| Error | Processing failed; resubmission may be needed |
For Turnitin AI detection, a file needs at least 300 words of prose and at most 30,000 words, under 100 MB, in .docx, .pdf, .txt or .rtf, so a long thesis may be checked chapter by chapter.
📱 Check your similarity with a Turnitin report — Every file comes with a Turnitin report and an AI check; similar passages are reworked without changing the meaning.
What does Turnitin say about accuracy?
Turnitin states that it aims to keep the false positive rate, meaning fully human-written documents flagged as AI-written, under 1% for documents with more than 20% detected AI writing, and that before each model update it tests on more than 700,000 academic papers written before the release of ChatGPT. It also accepts the trade-off: to keep false positives low, Turnitin AI detection may miss some AI text.
These are the company’s figures about its own product, not an independent evaluation, and a rate under 1% is not zero. Vanderbilt University, when it disabled the detector in August 2023, noted that at that rate around 750 of the 75,000 papers it had submitted the previous year could have been wrongly labeled.
Why does Turnitin AI detection produce false positives?
Turnitin’s FAQ lists the typical cases: text with little structural variation, text that repeats itself, and text that has been paraphrased without developing new ideas. Short documents of a few hundred words are predicted almost “all or nothing”, so a mix of human and AI text can be flagged entirely. For Turnitin AI detection, formulaic academic prose, standard methods descriptions and templated writing therefore sit close to the model’s picture of machine output.
Independent research adds an equity concern: a 2023 study in the journal Patterns by Liang and colleagues found that several widely used GPT detectors consistently misclassified writing by non-native English speakers as AI-generated while classifying native speakers’ writing accurately, and attributed this to the more limited linguistic variety of non-native prose.
That study did not test Turnitin’s model, and Turnitin says its training data deliberately includes second-language writers, but the mechanism it describes is why a careful instructor treats any flag as a question rather than a finding.
What should you do before and after a flag?
Turnitin AI detection is easier to answer when the evidence already exists. Before: keep dated drafts and the version history of your Word or Google document, keep your reading notes and reference-manager library, write in your own voice with specific links to your data and sources rather than generic templated sentences, and disclose any permitted AI use in the form your university prescribes.
After a Turnitin AI detection flag: ask to see the report and the highlighted passages (students cannot open them, but instructors can download the PDF and share it), and explain how the passages were written, with your drafts as evidence.
Turnitin advises instructors to use the indicator as a starting point for a conversation about the student’s writing process and to refer to their institution’s policy, and states that “the final decision on whether any misconduct has occurred rests with the reviewer”. Do not respond by running your text through a paraphrasing tool: the detector treats AI-paraphrased text as AI-generated.
📱 Ask on WhatsApp now — 5 years of experience across all disciplines and 14 academic services. Two-stage payment (80/20); revisions are free until supervisor approval, within 30% of the chapter.
Frequently asked questions
Can students see the AI writing indicator?
No. Turnitin states that the indicator and report are not visible to students; instructors can download the report as a PDF and choose to share it. Ask your supervisor for the highlighted passages rather than discussing a number you have not seen.
Does Grammarly trigger Turnitin’s AI detection?
Turnitin says its detector is not tuned to flag Grammarly’s spelling, grammar and punctuation corrections, and that in its tests such changes were mostly not flagged. Content produced by Grammarly’s generative features, such as draft generation, paraphrasing or summarizing, is likely to be flagged.
Why does the report show an asterisk instead of a percentage?
Because the estimated share is between 1% and 19%. Since 8 July 2024 Turnitin does not display scores or highlights in that range, where it found a higher incidence of false positives; an asterisk is not evidence of anything.
Is the AI percentage part of the similarity score?
No. The two are independent and do not influence each other. The similarity score measures matching text against Turnitin’s databases; the AI indicator estimates the probability that prose sentences were machine-generated.
Will the same file always get the same AI score?
Not necessarily. Turnitin updates its model, and a document is re-scored only if it is resubmitted, so two submissions of identical text at different times can differ. Reports generated before July 2024 may still show numbers below 20%.
Sources
- Turnitin. Using the AI Writing Report. https://guides.turnitin.com/hc/en-us/articles/22774058814093 (accessed October 2026)
- Turnitin. Turnitin’s AI writing detection capabilities FAQs. https://guides.turnitin.com/hc/en-us/articles/28477544839821 (accessed October 2026)
- Turnitin. File requirements for an AI Writing Report. https://guides.turnitin.com/hc/en-us/articles/28234943089933 (accessed October 2026)
- Turnitin. What should I do if the AI writing percentage is high? https://guides.turnitin.com/hc/en-us/articles/27139113024269 (accessed October 2026)
- Chechitelli, A. (2023). Understanding false positives within our AI writing detection capabilities. Turnitin blog. https://www.turnitin.com/blog/understanding-false-positives-within-our-ai-writing-detection-capabilities (accessed October 2026)
- Vanderbilt University, Brightspace Support. Guidance on AI Detection and Why We’re Disabling Turnitin’s AI Detector (16 August 2023). https://www.vanderbilt.edu/brightspace/2023/08/16/guidance-on-ai-detection-and-why-were-disabling-turnitins-ai-detector/ (accessed October 2026)
- Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7), 100779. https://doi.org/10.1016/j.patter.2023.100779
Sources last checked: November 2026. This article is general educational guidance and does not replace your university regulations or your supervisor’s instructions. How we prepare our content.