Turnitin AI detector
Vendor Turnitin, LLC Official site turnitin.com/solutions/topics/ai-writing How this card is sourced
Updated · AI Detector Index editors
Turnitin AI writing detection, launched in 2023, scores English submissions of 300 to 30,000 words and covers 7 languages according to the vendor's capabilities FAQ read on 8 November 2026 (Turnitin FAQ). Turnitin is the only detector in the index with no public tool: the AI indicator ships inside the Similarity Report for licensed institutions. How this card is sourced explains the rules, and the full detector index holds the other twenty-one cards.
What Turnitin detects
| Field | Value | Source |
|---|---|---|
| Content types detected | Text | guides.turnitin.com/hc/en-us/articles/284... |
| Generators covered | Amazon Nova 2 Lite, Claude-3-Haiku, Claude Sonnet-3.5, Claude Sonnet-3.7, Claude-Haiku-4.5, Claude Opus 4.6, Claude Sonnet 4.5, Claude Sonnet 4.6, DeepSeek V3.2, Gemini-1.0-pro, Gemini-1.5-pro, Gemini-2.0-flash, Gemini-2.5-flash, Gemini 2.5 Pro, Gemini-3-flash-preview, Gemini-3-pro-preview, Gemini-3.1-pro-preview, GPT-4o, GPT-4o-mini, GPT-5, GPT-5-mini, GPT-5-nano, GPT-5.1, GPT-5.2, GPT-5.2 Pro, GPT-5.3, GPT-5.4, GPT-5.4-pro, o1-mini, Grok-4.1, Llama-3.3-70b, Llama-4-Maverick-17b, Mistral-large-3, Spanish model: GPT-3.5 and GPT-4, Japanese model: GPT-4 (0613), GPT-4o, GPT-4o-mini | guides.turnitin.com/hc/en-us/articles/284... |
| Languages | English, Spanish, Japanese, Modern Arabic, Dutch, German, Turkish | guides.turnitin.com/hc/en-us/articles/284... |
| Input limits | English: file size under 100 MB; at least 300 words of prose text in a long-form writing format; must not exceed 30,000 words; accepted file types .docx, .pdf, .txt, .rtf. Lists, headers, references, tables, formulas and code are excluded from scoring. Unsupported-language submissions are not processed. Capabilities FAQ (edited 3 November 2026, read 10 November 2026): "We do not currently have plans to add these capabilities to Gradescope, since the primary use case for Gradescope is handwritten text while for AI detection we're focusing on typed text"; no OCR of scanned pages is stated. | guides.turnitin.com/hc/en-us/articles/284... |
| Free tier | None. AI writing detection is only available to customers that license Turnitin Originality (or the AI writing add-on for iThenticate 2.0). | guides.turnitin.com/hc/en-us/articles/284... |
| Integrations | LMS (Moodle, Blackboard, Canvas, etc.) via the Similarity Report; no AI indicator embedded in the LMS user interface, Microsoft Teams (Feedback Studio integration only) | guides.turnitin.com/hc/en-us/articles/284... |
| Data retention | There is no separate repository for AI writing detection, and customers retain the ability to choose whether to add their student papers into the repository or not. Results regarding the percentage AI writing identified by the detector, along with the segments identified highly likely written by AI, are retained as part of the similarity report. We will only process submissions for AI writing detection if the institution has the feature enabled. Customers can request a full deletion of their submissions; we cannot support partial data deletion requests to delete only the AI writing component. | guides.turnitin.com/hc/en-us/articles/284... |
| Launch year | 2023 | turnitin.com/blog/understanding-false-pos... |
| Ownership | Turnitin, LLC. No parent company stated on pages read. | turnitin.com/solutions/topics/ai-writing |
| Independent studies | 5 records | Studies index |
Turnitin AI writing detection scans text only, and only prose in a long-form format; lists, headers, references, tables, formulas and code are excluded from scoring (guides.turnitin.com FAQ, read 8 November 2026). The product page is at turnitin.com.
The generators the FAQ names run from GPT-4o and the GPT-5 family through GPT-5.4-pro, Claude-3-Haiku through Claude Opus 4.6, Gemini-1.0-pro through Gemini-3.1-pro-preview, plus Amazon Nova 2 Lite, DeepSeek V3.2, Grok-4.1, Llama-3.3-70b, Llama-4-Maverick-17b and Mistral-large-3. The Spanish model covers GPT-3.5 and GPT-4; the Japanese model covers GPT-4 (0613), GPT-4o and GPT-4o-mini (guides.turnitin.com FAQ, read 8 November 2026).
The 7 languages are English, Spanish, Japanese, Modern Arabic, Dutch, German and Turkish. Submissions in other languages are not processed (guides.turnitin.com FAQ, read 8 November 2026). The English detector launched on 4 April 2023 (Turnitin launch post, read 9 November 2026); the Spanish model was released on 12 September 2024 and updated on 5 May 2026, and the Japanese model on 2 April 2025, according to the model release notes (guides.turnitin.com release notes, read 9 November 2026); the Arabic, Dutch, German and Turkish models carry no release note, and the dated list is on the Turnitin languages question.
On paraphrased text, Turnitin says the detector "can also identify instances where AI-generated text may have been modified by AI paraphraser or bypasser (also called humanizers) tools to evade detection", with word-spinner detection added December 6, 2023 and bypasser detection August 27, 2025, both English-only; the vendor states: "We're unable to disclose the names of these tools" (model page, read 8 November 2026).
Turnitin, LLC owns the product. The AI indicator reaches Moodle, Blackboard, Canvas and other LMS platforms through the Similarity Report, with no indicator embedded in the LMS interface itself; a Microsoft Teams integration covers Feedback Studio only (guides.turnitin.com FAQ, read 8 November 2026).
Limits and plans
An English submission must be under 100 MB, contain at least 300 words of prose and not exceed 30,000 words, in .docx, .pdf, .txt or .rtf (guides.turnitin.com FAQ, read 8 November 2026). The minimum was raised from 150 words in May 2023 (Turnitin claims record, read 8 November 2026). The same FAQ, edited by Turnitin on 3 November 2026 and read 10 November 2026, states that "for AI detection we're focusing on typed text" and that Gradescope, whose primary use case is handwritten text, will not gain the capability, so a scanned page gets no AI score (guides.turnitin.com, read 10 November 2026); the scanned PDF question and the handwriting and OCR view rest on that sentence.
There is no free tier. The FAQ states that AI writing detection "is only available to customers that license Turnitin Originality (or the AI writing add-on for iThenticate 2.0)" (guides.turnitin.com FAQ, read 8 November 2026). Turnitin publishes no plan table or price list for the feature, so this card carries none; the index records the free allowance as 0 words per month (detectors.json, read 8 November 2026).
On retention, Turnitin says there is no separate repository for AI writing detection; the percentage and highlighted segments are retained as part of the similarity report, and customers can request full deletion of submissions but not deletion of the AI component alone (guides.turnitin.com FAQ, read 8 November 2026).
What the vendor claims
Turnitin publishes no overall accuracy or recall figure for the detector. On false positives, Turnitin, LLC says: "We strive to maximize the effectiveness of our detector while keeping our false positive rate, incorrectly identifying fully human-written text as AI-generated, under 1% for documents with over 20% of AI writing. Before every update or new model release, we perform tests on over 700,000 additional academic papers that were written before the release of ChatGPT to further validate our less than 1% false positive rate." The sentences are from the capabilities FAQ, read 8 November 2026. The May 2023 blog post gave 800,000 papers; the current FAQ gives over 700,000; this card quotes the FAQ.
Turnitin, LLC also says: "ELL writers received a 0.014 false positive rate, and native English writers received a 0.013" (guides.turnitin.com FAQ, read 8 November 2026).
Turnitin, LLC adds: "the percentage on the AI writing indicator should not be used as the sole basis for action or a definitive grading measure by instructors", and that scores of 1% to 19% show as an asterisk with no percentage (guides.turnitin.com FAQ, read 8 November 2026). An asterisk on a Turnitin report means the model saw something below its reporting line, not that a student used AI.
What independent studies found
- AIDI-S-0001Weber-Wulff 20232023
Turnitin: highest score of 14 tools, 41 of 54 documents correct (76% binary accuracy, rank 1); 81% logarithmic accuracy; 0 false accusations in 18 human-written documents; the only tool to classify all unmodified ChatGPT documents (03-AI and 04-AI) correctly; 10 of 36 AI documents undetected (27.8%), all among manually edited and Quillbot-paraphrased texts. Tested April 14 to 20, 2023; Turnitin knew the documents were part of a test suite.
- AIDI-S-0003Walters 20232023
TurnItIn: 100% of 126 documents correct, 0 incorrect, 0 uncertain (0 false positives in 42 human papers; 100% on both GPT-3.5 and GPT-4 papers). The author notes the student papers could potentially have been used to train the TurnItIn detector. Tested June 25 to July 12, 2023.
- AIDI-S-0005Perkins 20242024
Turnitin: baseline mean accuracy 61% (rank 2 of 7, Table 7); 50% on non-manipulated AI text falling to 7.9% after adversarial techniques (the largest drop, 42.1%, Table 8); 0 false accusations in 10 human samples; 88 of 104 AI samples undetected (UDC 84%).
- AIDI-S-0015Van Vlasselaer 20262026
Scored all 40 fully AI-generated papers from ChatGPT Deep Research mode under 20%, a 100% false negative rate for that category, reached 60.0% accuracy on 40 hybrid and 50.0% on 40 humanised papers, and classified all 40 human papers correctly (Tables 4 and 5).
- AIDI-S-0008Turnitin claims2026
Vendor statements, 2023 to 2026, not an independent study: document false positive rate under 1% for documents with over 20% AI writing, validated on 800,000 (May 2023) and over 700,000 (current FAQ) pre-ChatGPT academic papers; sentence-level false positive rate about 4% (2023); scores of 1% to 19% shown as an asterisk; minimum raised from 150 to 300 words (May 2023). No true positive rate, recall or overall accuracy figure, and no third-party audit, is published.
- AIDI-S-0031Vanderbilt 20232023, statement
Institutional statement, not a test: Vanderbilt University disabled the Turnitin AI indicator on 16 August 2023, reasoning that Turnitin's claimed 1% document false positive rate applied to the 75,000 papers it submitted in 2022 implied about 750 wrongly flagged papers a year, and citing no insight into how the tool works and student-data privacy (Vanderbilt Brightspace post; Turnitin's own figure is 'less than 1%' for documents scored over 20% AI).
Weber-Wulff 2023 gave Turnitin the highest score of 14 tools: 41 of 54 documents correct (76%, rank 1), 0 false accusations in 18 human-written documents, and 10 of 36 AI documents undetected, all among manually edited and Quillbot-paraphrased texts. Testing ran April 14 to 20, 2023, and Turnitin knew the documents were part of a test suite (arXiv 2306.15666, read 8 November 2026).
Walters 2023 found TurnItIn 100% correct on 126 documents between June 25 and July 12, 2023, with 0 false positives in 42 human papers. The author notes the student papers could potentially have been used to train the TurnItIn detector (De Gruyter PDF, read 8 November 2026).
Perkins 2024 reports a Turnitin baseline mean accuracy of 61% (rank 2 of 7), falling from 50% on non-manipulated AI text to 7.9% after adversarial techniques, the largest drop of the seven tools at 42.1%, with 0 false accusations in 10 human samples and 88 of 104 AI samples undetected (arXiv 2403.19148, read 8 November 2026).
The Turnitin claims record (2026) collects vendor statements from 2023 to 2026, not a study: document false positive rate under 1% for documents with over 20% AI writing, validated on 800,000 papers in May 2023 and over 700,000 in the current FAQ; a sentence-level false positive rate of about 4% in 2023; and no published recall, overall accuracy or third-party audit (Turnitin FAQ, read 8 November 2026).
A fifth record was added on 9 November 2026. Van Vlasselaer 2026 reports that Turnitin scored all 40 fully AI-generated papers from ChatGPT Deep Research mode under 20%, a 100% false negative rate for that category, reached 60.0% accuracy on 40 hybrid and 50.0% on 40 humanised papers, and classified all 40 human papers correctly; the authors write that Turnitin, GPTZero and Copyleaks "completely failed to detect the AI-generated content" in the fully AI category (DOI 10.1007/s40979-026-00226-w, read 9 November 2026).
The Vanderbilt 2023 statement, added 9 November 2026, is listed after the study records as a statement and counts toward none of the five: Vanderbilt disabled the Turnitin AI indicator on 16 August 2023, reasoning that the claimed 1% false positive rate applied to the 75,000 papers it submitted in 2022 implied about 750 wrongly flagged papers a year, while Turnitin's own figure of 23 May 2023 is "less than 1%" for documents scored over 20% AI.
Where Turnitin appears in the index
Rankings
- AIDI-R-01AI detectors ranked by independent testing Two records added in the fifth wave on 9 November 2026, Pegoraro 2023 and Fleckenstein 2024, changed nine counts and none: Pegoraro 2023 tested ZeroGPT, GPTZero, Copyleaks, Originality.ai, Writer, Content at Scale (BrandWell), DetectGPT, the GPT-2 Output Detector and GLTR, lifting GPTZero to 12, the GPT-2 Output Detector to 10, Copyleaks to 7 and Originality.ai and ZeroGPT to 6; Fleckenstein 2024 tested 289 teachers and no software, so it counts for no card ([Fleckenstein 2024 record](/studies/fleckenstein-2024-teachers/)). Illuminarty, one of four cards added in the same wave, joined with its one image record from Ha and Passananti 2024, so 18 of 72 cards now qualify.
- AIDI-R-02AI detector false positive rates, ranked
- AIDI-R-03AI detector free tiers, compared
- AIDI-R-04AI detectors ranked by languages supported
- AIDI-R-05AI detectors ranked by maximum input per check
- AIDI-R-08AI detectors ranked by published robustness to paraphrased and edited text
- AIDI-R-09AI detector false negative rates, ranked
- AIDI-R-11AI detectors ranked by named integration surfaces
- AIDI-R-12AI detectors ranked by the upload formats the vendor states
- AIDI-R-13AI detectors ranked by minimum text length per check
- AIDI-R-15AI detectors ranked by the LMS platforms the vendor names
- AIDI-R-17AI detectors ranked by the generator models the vendor names
- AIDI-R-18AI detectors ranked by study records on non-English and non-native text
- AIDI-R-20AI detectors ranked by launch year, with every change of hands and withdrawal the cards record
Fits these use cases, questions and comparisons
- AIDI-U-02AI detectors for school districts
- AIDI-U-05AI code detectors in the index
- AIDI-U-06AI detectors for admissions offices
- AIDI-U-08AI detectors for researchers checking their own manuscripts
- AIDI-U-10AI detectors for grant and proposal reviewers
- AIDI-Q-04Is JustDone's AI detector legit or accurate? What is published
- AIDI-Q-11Is an AI detector itself an AI? The four detection methods behind the cards, each sourced
- AIDI-Q-14What AI detector does College Board use? What the AP and SAT policy pages state
- AIDI-Q-15Which AI detectors support Spanish? The cards that name it and the one study that tested it
- AIDI-Q-16Does Gradescope have AI detection? What Gradescope's and Turnitin's own pages state
- AIDI-Q-31Which AI detectors were in the Weber-Wulff 2023 study? The 14 tools, their result on 54 documents, and what each one is today
- AIDI-Q-34What input languages does Turnitin's AI detection cover? Seven, added from April 2023 to 2026, with paraphrase detection in English only
- AIDI-Q-40Which AI detectors are free for nonprofits or education? The 12 cards that sell an institutional programme, the one university platform that charges nothing, and the two nonprofit tools that closed
- AIDI-C-02Pangram vs Turnitin: the 14 fields and the study results side by side
- AIDI-C-03ZeroGPT vs Turnitin: the 14 fields and the three shared studies side by side
- AIDI-C-05Winston AI vs Turnitin: the 14 fields and the study results side by side
- AIDI-C-10Scribbr vs Turnitin AI detection: the 14 fields and the study results side by side
- AIDI-C-14QuillBot vs Turnitin AI detector: a free consumer check against an institutional licence, on the same 14 fields
Turnitin qualifies for four of the five launch rankings and two of the five added on 9 November 2026: independently tested on 5 study records, false positive rate on the Walters 2023 figure of 0 in 42, free tiers with a row of 0 words because access is by institutional licence, and languages on a count of 7 (detectors.json, read 8 November 2026).
Since 9 November 2026 Turnitin also ranks third in false negative rate on 0 of 84 AI papers missed in Walters 2023, with the 40 of 40 missed in Van Vlasselaer 2026 in the same row, and sixth in robustness to paraphrase on 20 of 40 humanised papers in the same study.
Turnitin fits school districts through Moodle, Blackboard and Canvas delivery inside the Similarity Report, and recruiters only in the narrow sense of a text-only scope, since no public or API access exists (guides.turnitin.com FAQ, read 8 November 2026). The cards that share the Walters and Perkins studies with Turnitin are GPTZero and Copyleaks. The home page carries the full table.
Two comparisons added 9 November 2026 read this card against other tools: Winston AI vs Turnitin on the one shared record, Weber-Wulff 2023, and Scribbr vs Turnitin on Walters 2023. A third, QuillBot vs Turnitin, added 10 November 2026, sets the free consumer check a student runs beside this licence with no shared record, and a cost table that says the vendor publishes no price. One Turnitin product carries no detector: Gradescope, whose pages list AI-assisted grading and code similarity and about which Turnitin's FAQ says it has no plans to add AI writing detection (does Gradescope have AI detection?, read 9 November 2026).
Questions about Turnitin AI detector
Can anyone use Turnitin's AI detector?
No. The capabilities FAQ read 8 November 2026 states that AI writing detection is only available to customers licensing Turnitin Originality or the AI writing add-on for iThenticate 2.0. There is no free check, no price list and no API.
What is Turnitin's published false positive rate?
Under 1% for documents with over 20% AI writing, validated on over 700,000 pre-ChatGPT academic papers, per the FAQ read 8 November 2026. Turnitin's ELL research gives 0.014 for ELL writers and 0.013 for native English writers. Walters 2023 and Perkins 2024 each recorded 0 false positives.
How many words does Turnitin need?
At least 300 words of prose, and no more than 30,000, in a file under 100 MB, per the FAQ read 8 November 2026. The minimum was raised from 150 words in May 2023. Short documents produce mostly all-or-nothing predictions, the vendor says.
Does Turnitin detect humanized AI text?
Turnitin says yes for English: word-spinner detection arrived December 6, 2023 and bypasser detection August 27, 2025, per the model page read 8 November 2026. Perkins 2024 recorded accuracy falling from 50% to 7.9% after adversarial edits, the largest drop among seven tools.
Sources and revisions
Sources
- turnitin.com/solutions/topics/ai-writingRead 8 November 2026
- guides.turnitin.com/hc/en-us/articles/284...Read 8 November 2026
- guides.turnitin.com/hc/en-us/articles/282...Read 8 November 2026
- turnitin.com/blog/understanding-false-pos...Read 8 November 2026
- arxiv.org/pdf/2306.15666Read 8 November 2026
- degruyterbrill.com/document/doi/10.1515/o...Read 8 November 2026
- arxiv.org/pdf/2403.19148Read 8 November 2026
Revision log
- Card created
- Van vlasselaer 2026 record added, 5 study records
- Two comparisons and the gradescope question linked (fifth wave)2026-11-09: vanderbilt 2023 statement added as a note in the sixth wave
- Launch date and the spanish and japanese model release dates added from the launch post and release notes
- Quillbot vs turnitin comparison linked (tenth wave)2026-11-10: the capabilities faq's typed-text sentence added to the limits section from the 10 november re-read (twelfth wave)
). Illuminarty, one of four cards added in the same wave, joined with its one image record from Ha and Passananti 2024, so 18 of 72 cards now qualify.: register plate](/img/independently-tested-480.webp)
