Can Turnitin Detect ChatGPT? What Its AI Score Really Means
By Ashikur Rahman12 min readQuick answer: yes, it is built to, but only if your institution licenses and enables the feature, and it can miss text. Turnitin's FAQ lists GPT-4o, GPT-5 and versions up to GPT-5.4 among the models it detects in English. It scores prose sentence by sentence, needs at least 300 words, and hides any result below 20% behind an asterisk. Turnitin admits a report showing 50% "could contain as much as 65% AI writing". A score is a prompt for a conversation: Turnitin says it "does not make a determination of misconduct".
Which ChatGPT models does Turnitin say it can detect?
Turnitin publishes a list, and it runs to ChatGPT models released as late as March 2026. The table below comes from its AI writing detection FAQ, updated on 5 November 2026, for English submissions.
| Family | Models Turnitin lists | Newest listed release |
|---|---|---|
| OpenAI | GPT-4o, GPT-4o-mini, o1-mini, GPT-5, 5-mini, 5-nano, 5.1, 5.2, 5.2 Pro, 5.3, 5.4, 5.4-pro | March 2026 |
| Anthropic | Claude 3 Haiku, Sonnet 3.5, Sonnet 3.7, Haiku 4.5, Sonnet 4.5, Sonnet 4.6, Opus 4.6 | March 2026 |
| Gemini 1.0 Pro through Gemini 3.1 Pro Preview | February 2026 | |
| Others | Grok 4.1, DeepSeek V3.2, Llama 3.3 and 4 Maverick, Mistral Large 3, Amazon Nova 2 Lite | December 2025 |
Two gaps are worth knowing. The list as read does not name o3, GPT-5.5 or GPT-6, though other vendors now advertise detection for later models. GPTZero's site names GPT-6, and its 18 October 2026 post tests GPT-5.6. A missing name does not prove Turnitin cannot catch that output, and Turnitin says it is "continuously updating" its model. It only means Turnitin has not said so here.
The gap has shown up in research. In a 2026 study, 40 papers written by ChatGPT Deep Research in May 2025 all landed in Turnitin's lowest band. The authors say that mode was powered by an o3-based model, the name missing from the list. That test predates later Turnitin updates, so it is a snapshot and not a current rate. The evidence is in how accurate Turnitin's AI detector is.
How does Turnitin decide a sentence came from ChatGPT?
A trained model scores each sentence for how likely it is to be AI, in the context of the sentences around it. Turnitin describes no comparison against a stored copy of ChatGPT's answers. The FAQ says each sentence gets "a sentence score between 0 and 1", and "Sentences over a determined threshold are labeled AI, and the remainder are labeled Human."
The document score is plain arithmetic: "the percentage of sentences that are above the threshold." Turnitin updated the English model architecture in October 2026 to improve accuracy at "transition points" where text changes between human and likely AI writing. Turnitin also says the system is not programmed to check "burstiness" or "perplexity" as individual signals, despite how often those words appear online, and that "individual predictions may not always be explainable in simple feature-by-feature terms." So nobody, including your instructor, can read a highlight and say exactly why that sentence scored high.
Worked example: how the number appears
Assume an essay with 50 qualifying prose sentences. Lists, headings and references have already been removed.
| Sentences over the threshold | Sum | What the instructor sees |
|---|---|---|
| 0 | 0 / 50 = 0% | 0% |
| 8 | 8 / 50 = 16% | An asterisk, no number and no highlights |
| 12 | 12 / 50 = 24% | 24%, with highlights |
| 50 | 50 / 50 = 100% | 100%, with highlights |
The percentage is a share of evaluated sentences, not of your whole file. Turnitin says it "is not necessarily the percentage of the entire submission", which is why a report can look out of step with its highlights.
What does your instructor actually see?
Your instructor sees an indicator inside the Similarity Report, plus a separate AI report with highlighted text. The FAQ lists these states:
| Indicator | Meaning |
|---|---|
| Blue with a percentage | Processed. The share of qualifying text likely AI-generated or modified. |
| Low percentage (*%) | Detected below 20%. No score or highlights, because of false-positive risk. |
| Gray (- -) | Could not process: too short, too long, wrong language or file type, or submitted before the feature existed. |
| Error (!) | Processing failed. Try again later. |
Turnitin's pages are inconsistent about students. The FAQ says the indicator and report "are not visible to students". The newer AI Writing Report guide, updated on 6 November 2026, says "If you make the AI Writing Report available to students, they can open it". Its Canvas workflow guide also has a setting to let students view the report. So whether you see it depends on how the assignment was set up. Ask your instructor. How this appears inside a course is covered in what Canvas can and cannot detect.
Does Turnitin check every assignment for AI?
No. Three conditions have to be true, and this is the part most student guides skip.
- Your institution pays for it. "AI writing detection is only available to customers that license Turnitin Originality", or the add-on for iThenticate 2.0.
- Your institution left it on. Admins can disable it, and "we will only process submissions for AI writing detection if the institution has the feature enabled." Vanderbilt, Pittsburgh's Teaching Center and Waterloo have all switched it off or stopped supporting it.
- The file qualifies. The requirements are below.
| Requirement | Turnitin's rule | What it means |
|---|---|---|
| Minimum length | "at least 300 words of prose text in a long-form writing format" | A short reflection gets no AI report |
| Maximum length | "must not exceed 30,000 words" | Long theses may need splitting |
| File size | Less than 100 MB | Rarely an issue for prose |
| File types | .docx, .pdf, .txt, .rtf | PowerPoint is not on the list |
| Language | English, Spanish, Japanese, Arabic, German, Dutch, Turkish | Others get no report |
| Text type | Prose sentences in long-form writing | Not poetry, scripts, code, bullet points, tables or annotated bibliographies |
Those rules answer several common searches. A PDF is an accepted file type. A PowerPoint file is not listed for an AI report, and bullet points are not detected reliably either. Copy-pasted text is just text once submitted. Turnitin reads the prose in the file and does not describe watching how it got there, unless the assignment uses a process tool, covered below.
Transparency note: CriticNest has no affiliate or commercial relationship with Turnitin, OpenAI, GPTZero, Grammarly or QuillBot. Nothing here is a paid placement. No link in this article earns a commission. Every quotation was read from the linked pages on 6 November 2026, and no paper was submitted to Turnitin for this article. If that changes, the disclosure will move up.
Can Turnitin detect ChatGPT if you paraphrase it?
Turnitin says yes for paraphrasing tools and is silent on doing it by hand. The FAQ says its detection includes "likely AI-generated content that may have been modified using a word spinner/AI paraphrasing or bypassing tool". It adds that this runs "automatically for all submissions" from institutions with the indicator on, in English only. On 27 August 2025 it added bypasser tools, which it also calls humanizers.
| What happened to the ChatGPT text | What Turnitin says |
|---|---|
| Pasted unchanged | The core use case |
| Run through a word spinner or paraphraser | Detected as likely AI-generated, in English |
| Run through a humanizer or bypasser | Included since August 2025, in English |
| Translated from another language | Not addressed on the pages read |
| Rewritten by hand, sentence by sentence | Not addressed. No rate published. |
| Used only for ideas or an outline, with your own sentences | The detector scores sentences, so there is no output to score. Your rules may still apply. |
The last row is where policy and detection split. Turnitin's own guide on how students might use AI describes outlines and prompts as common uses. Whether those are allowed is for your course to decide. A low score does not make banned help allowed, and this article gives no method for lowering one. QuillBot, one of the tools Turnitin names, has its own breakdown in whether QuillBot shows up as AI.
Does Turnitin flag Grammarly and other editing tools?
Turnitin separates correction from generation. It says grammar fixes were "in most cases" not flagged, and that Grammarly's generative features "will likely be flagged". The full reasoning is in whether Grammarly shows up as AI, and the same line is the safest way to think about ChatGPT used as an editor. Asking it to polish a whole paragraph produces model-written sentences, even when the idea was yours.
What can Turnitin miss, and why?
Turnitin chooses to miss some AI text to avoid accusing people wrongly. The FAQ is direct: "there is a chance that we might miss some AI written text in a document. We're comfortable with that since we do not want to incorrectly highlight human-written text as AI-written."
| Situation | What Turnitin says or shows |
|---|---|
| A document that is partly AI | "if we identify that 50% of a document is likely written by an AI tool, it could contain as much as 65% AI writing" |
| A small AI share | Below 20% shows as an asterisk with no number |
| Very short work | Under 300 words gets no report; "a few hundred words" can be "all or nothing" |
| Non-prose | Lists, tables, code and references are left out |
| Older reports | New model versions do not update them unless the work is resubmitted |
None of this is a loophole worth relying on. A report can be regenerated later, since Turnitin says past work can be checked "if they're re-submitted", and other evidence exists beyond the score. Instructors can compare earlier work, ask you to explain your argument, check whether citations are real, and look at your drafts. The University of Pittsburgh notes that assignments with hallucinated sources can be an integrity violation on their own, without any detector.
Can a teacher see how you wrote it, not just what you wrote?
Sometimes, with a process tool rather than the AI detector. Standard AI reports look at the final file. Turnitin also sells Clarity, whose Writing Report includes "Writing process playback" and pasted text findings for work done in that tool. Google Docs and Microsoft Word keep version histories, and Grammarly Authorship and GPTZero's Writing Replay record typing and pasting in the same spirit.
That cuts both ways. A replay can expose a pasted essay, and it is also the best evidence an honest writer can offer against a wrong score, as the next section explains.
What should you do with all this?
Use it to work honestly and to protect yourself, whichever side of the desk you are on.
If you are a student:
- Read the AI rule for each assignment. It often changes course by course.
- Write in a tool with version history, and keep notes and drafts.
- Disclose permitted AI help the way your course asks.
- Be ready to explain your argument and sources in your own words.
- If flagged, ask to see the report and bring your process record. Turnitin says the score "should not be used as the sole basis for adverse actions against a student".
If you are an instructor:
- Treat the percentage as "a single data point rather than a definitive response", in the words of Turnitin's review guide.
- Ignore precision that is not there: a score of 24% is an estimate, and text near the threshold can go either way.
- Talk to the student and ask for drafts before making any finding.
- Remember that detector vendors disagree. GPTZero handles edited text differently, so a second tool can return a different reading.
How this was researched
Turnitin's FAQ, AI Writing Report guide, file requirements, release notes, Canvas workflow, review guide and Clarity Writing Report pages were read through its public help-centre API on 6 November 2026. GPTZero's site and the International Journal for Educational Integrity study were read the same day.
Not verified: how Turnitin handles translated or hand-rewritten AI text, because it publishes nothing on either; whether any model missing from its list goes undetected; whether a given school has the feature on, which only that school can tell you. No paper was submitted to Turnitin for this article, because access requires an institutional licence. The author's role is careful reader of the fine print, not tester.
The verdict: it can detect ChatGPT, but the score is not the finding
Turnitin is built to flag ChatGPT, including paraphrased output in English, and it says so in detail. It only works if your institution has it on and your file has at least 300 words of prose. It accepts missing some AI text in exchange for fewer false accusations, and it hides anything under 20%.
The useful way to read a report is the way Turnitin tells instructors to: as the start of a conversation. Students should plan on being able to show their work. Instructors should plan on asking for it.
Frequently asked questions
Can Turnitin detect ChatGPT 5?
Turnitin's FAQ lists GPT-5, GPT-5-mini, GPT-5-nano and versions through GPT-5.4 and GPT-5.4-pro among the models it can detect in English submissions. The list as read on 6 November 2026 stops there. Later models are not named, which does not prove they go undetected.
Can Turnitin detect ChatGPT if you copy and paste?
Pasted ChatGPT text is the core case the detector is built for, assuming the feature is enabled and the file qualifies. The standard AI report reads the submitted text rather than recording how it arrived. Process tools such as Turnitin Clarity can show paste events when the work is done inside them.
Can Turnitin detect ChatGPT if you paraphrase it yourself?
Turnitin publishes no answer or rate for manual rewriting. Its stated detection covers AI paraphrasing and bypasser tools. Its FAQ also warns that text "paraphrased without developing new ideas" can produce false positives. Rewording banned AI output is still banned, whatever the score shows.
Can Turnitin detect AI in a PDF?
Yes. Turnitin accepts .pdf, .docx, .txt and .rtf files for an AI report. The file needs at least 300 words of long-form prose. Turnitin advises having all students submit in one format so submissions are processed uniformly.
Can Turnitin detect AI in PowerPoint?
PowerPoint is not among the accepted file types for an AI writing report. Turnitin also says its model does not reliably detect short-form writing such as bullet points. Your instructor can still read the slides and ask about them, and the Similarity Report is a separate check.
Can Turnitin detect Claude and Gemini too?
Turnitin lists Claude models through Opus 4.6 and Sonnet 4.6 and Gemini models through 3.1 Pro Preview, plus Grok, DeepSeek, Llama, Mistral and Amazon Nova. Those lists apply to English submissions. Other languages run on different models with their own FAQs.
What percentage of AI is acceptable on Turnitin?
Turnitin sets no acceptable percentage. It shows nothing numeric below 20%, because false positives are more common there, but your institution decides what any score means. Any AI use may be fine or banned depending on the assignment. Ask for the policy rather than assuming a safe number.
Can students see their Turnitin AI score?
Only if it is made available. Turnitin's FAQ says the report is not visible to students, while its newer report guide says instructors can make it available, and its Canvas guide has a setting for that. Instructors can also download the report as a PDF and share it with you.
Can you check your paper for AI on Turnitin yourself?
Not directly. AI writing detection is sold to institutions that license Turnitin Originality, and processing runs only where the feature is enabled. Sites offering a personal Turnitin check are not described on Turnitin's pages. Be careful uploading unpublished work to them.
Does a high Turnitin score prove you used ChatGPT?
No. Turnitin says it "does not make a determination of misconduct" and that its model "may misidentify human-written, AI-generated, and AI-paraphrased text". The score should not be the sole basis for action. The final decision rests with the reviewer and your institution's policy.
