Does Turnitin Detect AI? What UK Students Need to Know in 2026–27
Key Takeaways Yes, Turnitin detects AI. Its model flags text likely written by ChatGPT and other large language models, plus AI-paraphrased text and text...
Key Takeaways
- Yes, Turnitin detects AI. Its model flags text likely written by ChatGPT and other large language models, plus AI-paraphrased text and text run through "humaniser" tools.
- Your university decides whether it is switched on. Glasgow, Dundee, Newcastle and Liverpool say they do not use it. Bristol says it does.
- Low scores are hidden. Anything from 1% to 19% shows only as an asterisk (*%). A number appears from 20% upwards.
- You cannot see the AI report by default. Only your tutor can, unless they choose to share it.
- A flag is not a finding. Turnitin says the score should never be the sole basis for action against you.
Your best protection is a process trail: drafts, version history, notes, sources and any AI declarations your brief asks for.
The Short Answer
Yes, Turnitin can detect AI-generated writing. Its AI writing detection model estimates how much of the qualifying prose in your work was produced or paraphrased by tools such as ChatGPT. The result is a probability, not proof, and your university decides whether the feature is switched on at all.
That matters more every term. The HEPI Student Generative AI Survey 2026 found that 94% of UK undergraduates now use generative AI for assessed work, and 12% put AI-generated text straight into their submissions, up from 8% a year earlier. Universities are responding. A Guardian investigation reported almost 7,000 proven cases of AI cheating in 2023–24, about 5.1 for every 1,000 students.
This guide from The Academic Papers UK explains how Turnitin AI detection works, how to read your score, which UK universities use it, and exactly what to do if your work is flagged. Every claim has been checked against Turnitin's own documentation and published university policies.
Need Someone Who Knows Your Subject?
Your request is matched to a subject team, with a writer usually assigned within an hour.
How Does Turnitin AI Detection Work?
Turnitin splits the qualifying prose in your submission into segments, scores each one for patterns typical of large language model output, and reports the share it judges likely to be AI-written. Turnitin does not publish its full method, but the model is trained to recognise the statistical habits of machine-written text.
Crucially, the AI score is not your similarity score. The similarity report compares your work with Turnitin's database of published material, web pages and past submissions. The AI report looks for machine-writing patterns. A paper can flag for AI with almost no matching text, and a paper full of copied text can show no AI flag at all.
| Similarity score | AI writing score | |
| What it checks | Matches with published sources, websites and earlier submissions | Patterns typical of AI-generated or AI-paraphrased prose |
| What it can suggest | Copying, poor referencing or self-plagiarism | Possible unauthorised AI use |
| Can you see it? | Often, if your tutor enables it | No, unless your tutor shares it |
| Is it proof? | No. A tutor still judges each match | No. Turnitin says it must not be the sole basis for action |
What Does Turnitin Actually Scan?
The detector only reads "qualifying text", which Turnitin defines as prose sentences in long-form writing. According to the Turnitin AI writing detection FAQ (updated 21 September 2026), these limits apply:
- Length: at least 300 words of prose and no more than 30,000 words.
- Excluded content: bullet points, lists, tables, code and other non-sentence text are not scored.
- Languages: long-form English, Spanish, Japanese and Modern Arabic. Arabic support was added in August 2026.
- Files: .docx, .pdf, .txt and .rtf, up to 100 MB.
Two practical points follow. First, your percentage describes your prose only, not the whole document, so a methods table or appendix does not dilute it. Second, a dissertation longer than 30,000 words of prose may not be read in full, so ask your department how it handles long submissions before you hand in.
Can Turnitin Detect Chatgpt, Paraphrasers And AI Humanisers?
Yes, it is designed to. Turnitin's report now separates three kinds of text:
- AI-generated only: text that looks written by a large language model such as ChatGPT, Gemini or Claude.
- AI-generated and AI-paraphrased: AI text that was then reworded by a paraphrasing tool. Turnitin added this category in July 2024.
- AI bypasser detection: text altered by "humaniser" tools sold to beat detectors. Turnitin announced this in August 2025.
Paraphrase and bypasser detection currently work in English only. Independent research shows that disguised AI text is harder to catch. Weber-Wulff et al. (2023) found that about half of manipulated AI text slipped past the 14 tools they tested. However, detection keeps changing, and many universities treat using a humaniser as misconduct in itself. Trying to beat the detector is a gamble with your degree, not a strategy.
What Does Your Turnitin AI Score Mean?
The AI Writing Report gives your tutor one headline figure: the percentage of qualifying text the model judges likely to be AI-generated or AI-paraphrased. Flagged passages are highlighted in the document so the tutor can read them in context.
| Score shown | What it means | What you should do |
| 0% | No qualifying prose was judged AI-written | Nothing, but keep your drafts in case questions arise on other grounds |
| *% | A low score between 1% and 19%. The number is hidden because false positives are more likely at this level | Treat it as inconclusive. It is not evidence of AI use |
| 20% to 100% | That share of qualifying prose looks AI-written or AI-paraphrased to the model | Check your brief's AI rules and gather your process evidence now |
What Does the Asterisk (*%) Mean on Turnitin?
Since 16 July 2024, Turnitin no longer displays a number for AI scores between 1% and 19%. The report shows "*%" instead. Turnitin made the change because low scores carry a higher risk of false positives. Older guides that say you will see a low percentage, such as 7% or 12%, are now out of date.
Consult with one of our 650+ PhD Experts.
Ready to elevate your essay game? Let our experts do the heavy lifting!
AI lacks the depth your degree requires. Our human experts deliver tailored research and authentic academic insights.
Consult with an Expert, It's freeCan You See Your Own Turnitin AI Score?
Not by default. Turnitin states that the AI writing indicator and report are not visible to students. Your tutor can download the report as a PDF and share it with you, and you are entitled to ask for it if your work is questioned.
Some modules now use Turnitin Clarity, a paid add-on released in July 2025. You write inside a Turnitin workspace, and your tutor can see a revision timeline, which text was pasted rather than typed, a summary of any approved AI chat, and a playback of the piece taking shape. If your module uses Clarity, your writing process becomes the evidence, which strongly favours students who draft their own work.
How Accurate Is Turnitin AI Detection?
Turnitin reports a document-level false positive rate of less than 1% for submissions where it detects 20% or more AI writing. At sentence level the rate is higher, at about 4%, and Turnitin found that 54% of wrongly flagged sentences sat next to genuinely AI-written text (Chechitelli, 2023). The company says it tests each model update against more than 700,000 academic papers written before ChatGPT existed.
Independent studies paint a more cautious picture. Here is what the research actually tested, which many online guides get wrong:
| Study | What was tested | Key finding |
| Weber-Wulff et al. (2023), International Journal for Educational Integrity | 14 detection tools, including Turnitin | No tool reached 80% accuracy. Turnitin scored highest, but the authors judged detectors neither accurate nor reliable |
| Perkins et al. (2024), International Journal of Educational Technology in Higher Education | Seven detectors, including Turnitin, on plain and manipulated AI text | Turnitin caught 61% of unmodified AI text, falling to about 19% once simple manipulation techniques were applied |
| Liang et al. (2023), Patterns | Seven public GPT detectors. Turnitin was not tested | An average of 61.2% of non-native English (TOEFL) essays were wrongly flagged as AI |
| Turnitin (Adamson, 2023), in-house | Turnitin's own detector on essays by English learners | No statistically significant bias for documents of 300+ words. A significant gap appeared for 150 to 300 words |
Two of these studies used older versions of the model, so treat the exact figures as historical. The pattern is consistent, though. Turnitin is careful about false accusations on longer documents, it misses a good deal of disguised AI text, and short or formulaic writing is the hardest to judge. Whatever the score, a human has to weigh it against other evidence.
Who Is Most At Risk Of A False Positive?
- Students writing in a second language, whose phrasing can be more predictable. This is the concern behind most fairness debates.
- Short or highly formulaic pieces, such as reflective templates or rigid lab write-ups, close to the 300-word minimum.
- Students who let editing tools rewrite whole sentences. Text reworded by an AI-powered tool may be read as AI-paraphrased, even if the ideas were yours.
Does Your UK University Use Turnitin AI Detection?
There is no single UK rule. Each institution chooses which Turnitin features to switch on. A university can run similarity checks on every essay while leaving the AI detector off entirely.
The sector body Jisc advises that no AI detection software can conclusively prove a text was written by AI, and the QAA has warned against relying on detection tools. As a result, many universities have published a position. These are verified against each institution's own page:
| University | Position | What it says (checked October 2026) |
| University of Glasgow | Opted out since March 2023 | Cited concerns about false positives and a release made without full testing |
| University of Dundee | Opted out since April 2023 | Could not test reliability. Staff should not use unauthorised AI detection tools |
| Newcastle University | Does not use it | Calls detectors highly unreliable. A tool may be used only within a misconduct investigation |
| University of Liverpool | Switched off from 1 August 2024 | Cited reliability, false positives and bias against non-native speakers |
| University of Cambridge | Does not encourage detectors | Says detection tools offer no evidence for investigations |
| University of Oxford | No endorsed detector | Each summative assessment must state whether and how AI may be used |
| University of Southampton | Leaving Turnitin altogether | Will not renew its contract beyond 2026–27, over licence changes on student data. Turnitin denies training its AI tools on student work |
| University of Bristol | Uses it | Applied Turnitin's bypasser detection update from August 2025 |
Policies change, so treat this table as a snapshot dated October 2026. An independent tracker reviewing 95 UK universities in August 2026 found that most institutions taking a public position have chosen not to run the AI indicator, while around half have published nothing at all.
How Can You Find Your University's Position?
Search for "[your university] Turnitin AI detection", then check these documents in order:
- Your module or assignment brief, which overrides general guidance.
- Your university's generative AI guidance for students.
- The academic integrity or academic misconduct policy.
- Your assessment handbook or Turnitin student guidance.
If you are still unsure, email your module leader and ask two questions: is Turnitin's AI detector switched on for this assessment, and what AI use is allowed? A written answer protects you far better than a remembered conversation.
What UK Universities Actually Count As AI Misconduct
AI detection and AI misconduct are not the same thing. A detector may flag text as probably machine-written, but it cannot know whether AI was allowed for that assignment.
Rules differ widely. UCL sorts every assessment into one of three categories: GenAI tools cannot be used, GenAI can be used in an assistive role, or GenAI has an integral role. Oxford requires each summative assessment to state whether and how AI may be used, and breaking that statement counts as cheating under its disciplinary rules. In every case, the instructions for your specific task come first.
So the real question is never just "did Turnitin detect AI?" It is "what did my assessment allow, and can I show how I produced this work?" If you are weighing up whether an AI tool is worth the risk, read our breakdown of the disadvantages of using a ChatGPT essay writer.
What Happens If Turnitin Flags Your Essay As AI?
A Turnitin flag is not an automatic finding of academic misconduct. It usually leads to a conversation or a formal integrity meeting. Follow these seven steps.
1. Do Not Panic Or Delete Anything
Keep every draft, note, source and file, including any AI chat history if AI use was permitted. These are your strongest evidence, and deleting them can look like concealment.
2. Ask For The Evidence In Writing
You cannot answer an allegation you have not seen. Ask for the AI report, the exact allegation and the policy clause being applied. You can adapt this short email:
Subject: Request for information about the academic integrity concern on [module code / assignment]
Dear [name],
Thank you for letting me know about the concern raised on my [assignment title]. So that I can respond properly, please could you send me:
- the Turnitin AI writing report or any other evidence being relied on;
- the specific allegation, including which sections are in question;
- the policy and regulation clauses that apply; and
- the date, format and attendees of any meeting, and whether I may bring a Students' Union adviser.
I wrote this work myself and I will bring my drafts, version history and notes to show how it was produced.
Kind regards, [name, student number]
3. Check exactly what you are accused of
Is the claim that you used AI at all, that you used it without permission, or that parts do not read like your own work? These are different issues with different answers. Re-read your assessment brief to confirm which category of AI use applied.
4. Gather evidence of your writing process
- Word or Google Docs version history
- Earlier drafts and outlines
- Handwritten or digital notes and research plans
- Downloaded and annotated sources
- Reference manager records, such as Zotero or EndNote
- Supervisor emails and written feedback
- Data analysis files or lab notebooks
- Any AI-use declaration your brief required
Checking The Cost First?
Undergraduate support starts from £13.95 per page, with free revisions included.
5. Contact your Students' Union
Your Students' Union advice centre is free and independent of your department. Advisers can explain the procedure, help you prepare a written statement and, at many universities, attend the meeting with you.
6. Walk them through your process at the meeting
Expect questions about your argument, not just your software. Be ready to explain why you chose your main sources, how your structure developed, and what key terms mean in your own words. Show an early draft next to the final version. Calm, specific answers backed by files are far more persuasive than a general denial.
Never invent proof after the fact. Do not edit version histories or write backdated notes. Genuine working materials matter precisely because you created them while the work was happening.
7. Appeal, then use the right external route
If you believe the university broke its own procedures, ignored relevant evidence or made another procedural error, use its internal appeal process first and keep to its deadlines. Once that is finished, you can take a complaint to an independent body:
- England and Wales: the Office of the Independent Adjudicator for Higher Education (OIA). You normally need a Completion of Procedures letter and must complain within 12 months of the university's final decision.
- Scotland: the Scottish Public Services Ombudsman (SPSO). Complaints are usually due within 12 months, and it cannot overturn academic judgement.
- Northern Ireland: the Northern Ireland Public Services Ombudsman (NIPSO), once the university's process is complete.
How to Protect Yourself Before You Submit
Trying to outsmart a detector is the wrong goal. The strongest protection is a process trail: a record, built as you work, that shows how your ideas became a finished piece. Use this checklist on every assessed piece:
- Draft where versions are saved. Word with OneDrive or Google Docs keeps a version history automatically.
- Keep your thinking. Save outlines, mind maps, reading notes and calculations, even rough ones.
- Keep your sources. Store the exact PDFs and pages you used, ideally in a reference manager.
- Confirm the AI category for this task. Never assume last term's rules still apply.
- Declare permitted AI use using your university's template, if one is required.
- Log supervisor contact on dissertations, including chapter drafts and written feedback.
- Avoid humanisers and AI paraphrasers. They are built to disguise authorship, and Turnitin now looks for them.
Dissertations need extra care because they run over months. Our guide to dissertation writing shows how to plan each chapter so your drafts build a clear trail. For referencing and source handling, work through the guide to avoiding plagiarism before you submit.
International students and AI detection
International students face a distinct risk. In July 2026, a HEPI article by Brendal Aformeziem of the University of Strathclyde argued that AI detection falls hardest on non-native English writers. It noted that international students make up 24% of UK enrolment and 51% of postgraduates, and called on universities to stop treating detection as the main evidence of misconduct.
If English is your second language, three habits help. Learn your university's rules on proofreading and language support, because some limit what a third party may change. Keep records of any language help you receive. And save your drafts in your first language if you plan in it, since they show your ideas developing.
Getting human support that stays within the rules
Many students who worry about AI flags want a second pair of human eyes before they submit. Choose support that leaves a clean trail: feedback on your own draft, model answers you learn from, and editing that respects your university's proofreading policy.
The Academic Papers UK provides essay help in the UK from human subject specialists, and a dissertation editing service for students who want structure, clarity and referencing checked on work they have written. Every piece is written or edited by people, never generated by AI.
Conclusion: what UK students should remember
Turnitin does detect AI, and it now looks for paraphrased and "humanised" text too. Its score is still an estimate, a low result appears only as an asterisk, and Turnitin itself says the number must never decide a case alone. Whether it is used at all depends on your university, so check your policy and your brief before every submission. Above all, keep a process trail, because evidence of how you worked is what settles any doubt.
Frequently asked questions
Can Turnitin Detect ChatGPT?
Yes. Turnitin's AI writing detection model is built to flag text likely produced by large language models, including ChatGPT. It reports a probability rather than proof, and Turnitin says the score should never be the only basis for action against a student.
Can Turnitin Detect AI If I Paraphrase It Or Use A Humaniser?
Often, yes. Since July 2024 the report separates AI-paraphrased text, and since August 2025 Turnitin has looked for text altered by AI bypasser or "humaniser" tools, in English. Detection of disguised text is less reliable, but many universities treat using these tools as misconduct.
What Does An Asterisk (*%) AI Score Mean On Turnitin?
It means Turnitin detected a low level of AI-like writing, between 1% and 19%. Since July 2024 Turnitin hides these numbers because false positives are more likely at that level. An asterisk is not evidence that you used AI.
Can Turnitin's AI Detector Be Wrong?
Yes. Turnitin reports a document-level false positive rate below 1% for scores of 20% or more, but about 4% at sentence level. It openly says human writing can be misclassified, so a human must review every result.
Can I See My Own Turnitin AI Score?
Not by default. Turnitin states the AI indicator and report are not visible to students. Your tutor can download and share the report, and you should ask for it if your work is questioned.
Does Turnitin Detect AI In Dissertations Over 30,000 Words?
Turnitin's AI detector processes up to 30,000 words of qualifying prose. If your dissertation is longer, ask your department how it handles the AI report, as it may not cover every chapter.
Do All UK Universities Use Turnitin AI Detection?
No. Each university decides. Glasgow, Dundee, Newcastle and Liverpool have said they do not use it, Cambridge discourages detectors, and Southampton is leaving Turnitin after 2026–27. Bristol is among those that do use it. Check your own university's guidance.
Is Using Grammarly Or A Proofreader Treated As AI Use?
It depends on your university and your brief. Basic spelling and grammar checks are widely accepted, but tools that rewrite whole sentences may breach some policies and may be read as AI-paraphrased. Check your proofreading policy first.
What Should I Do If I Am Falsely Accused Of Using AI?
Stay calm, keep every file, and ask in writing for the AI report and the exact allegation. Gather your drafts and version history, contact your Students' Union, and explain your process at the meeting. If procedures were not followed, appeal internally and then contact the OIA, SPSO or NIPSO.
