Short answer first: sometimes, and less often than students assume. Turnitin's AI writing indicator reports the share of qualifying prose that its model thinks may have been generated, it needs at least 300 words, and at Leeds Beckett only staff ever see the number. Whether it runs at all is your university's switch. Newcastle has disabled the AI detection features of its platforms. Bristol has the feature but tells staff not to use it as the primary or sole indicator in a suspected case of cheating. A percentage is a reason to look, not a finding.
What follows is about evidence: what the tool measures, what it cannot measure, what a UK investigation actually examines, and what you should have kept. What you are allowed to use in the first place is a different question, answered in the rules on ChatGPT at UK universities.
What is Turnitin's AI writing indicator?
It is a separate feature from the similarity report, running a different model on the same file. Leeds Beckett's guide gives the clearest definition published by a UK university: the report shows the overall percentage of prose sentences contained in a long-form writing format within the submitted document, that Turnitin's AI writing investigation model indicates may have been generated by AI.
Three things in that sentence do real work.
- Prose sentences in a long-form format. Bullet lists, tables, code, equations and reference lists are not what the model is scoring. An essay is; a lab report full of figures is partly.
- Indicates may have been generated. It is a model output about the text, not a record of what happened on your laptop.
- At least 300 words. Leeds Beckett states that a file needs at least 300 words of prose text in a long-form writing format to be processed at all. Short reflective pieces and discussion posts often fall under the threshold.
Turnitin's own description is narrower than the rumour too. Its AI writing pages say the detector helps educators identify when AI writing tools such as ChatGPT may have been used in students' submissions, and place detection as one piece in a broader puzzle rather than as the answer. The company says it has processed more than 130 million papers for AI detection since launch, with 12.5 million carrying over 20 percent AI writing and 3.5 million at 80 percent or more.
Is the percentage proof of anything?
No, and the universities that run it say so in their own guidance. Bristol's advice to staff is direct: do not use Turnitin AI writing detection as your primary or sole indicator in a suspected case of cheating, and be extremely cautious about the number values and the risk of false positives. Bristol also says the use of the feature is under review with faculties and schools. Leeds Beckett tells staff the AI writing report should be used with caution and not as a final judgement on a piece of work, and to gather context first, including comparison with the student's other work and a conversation with them.
| What the indicator does show | What it does not show |
|---|---|
| A model's estimate of how much long-form prose resembles generated text | Whether you used a generative tool |
| A figure calculated from at least 300 words of qualifying prose | Anything about lists, tables, code, equations or references |
| Highlighted passages for staff to read | Which tool was used, or when |
| A reason to open a conversation | A finding of academic misconduct |
| Nothing at all if the feature is switched off | Whether the use, if any, was permitted for that assessment |
The last row matters more than students expect. The indicator has no view on permission. An assessment where AI was allowed and acknowledged can still return a high number, and that is not a case.
Is it even switched on at your university?
This is the first thing to establish, because a great deal of student anxiety is about a feature that is not running.
Newcastle has turned it off and explains why. Its guidance for staff states that the AI detection features of our current platforms have been disabled, that text and digital media generated by AI cannot be reliably detected, and that detection tools provide insufficient detail about how scores are generated and what they mean. It goes further and says that expecting markers to identify AI-generated text is difficult in most scenarios and is not an expectation the university should place on colleagues.
Bristol sits on the other side, with the feature available and a warning attached to every use of it. Leeds Beckett documents the report in its staff guides, with student visibility switched off.
So the honest answer to "will my essay be scanned" is that it depends on your institution, and that your institution has probably published its position somewhere in staff guidance rather than in a student handbook. If you cannot find it, ask your school office directly. You are entitled to know what is run against your work.
Why do detectors flag people whose first language is not English?
This is the best-evidenced problem with the whole category of tool, and it lands hardest on international students.
Liang, Yuksekgonul, Mao, Wu and Zou tested seven GPT detectors and published the result in Patterns in 2023 as GPT detectors are biased against non-native English writers. On a set of 91 TOEFL essays written by non-native English speakers, the detectors flagged an average of 61.22 percent as AI-generated, and 97.80 percent of the essays were flagged by at least one detector. The same detectors were close to accurate on essays by US school students, misclassifying 5.19 percent. When the TOEFL essays were rewritten with richer word choices, the average false positive rate fell from 61.22 percent to 11.77 percent.
The mechanism is the uncomfortable part. What moved the score was vocabulary range, not authorship. A writer with a smaller active vocabulary and more predictable sentence patterns reads to these models like a generator, which is exactly the profile of a competent student writing in a second or third language. If that is you, the wider picture of studying in English is in studying in English as an international student in the UK.
Turnitin says this does not apply to its own detector. Its pages report a false positive rate of 0.014 for English language learner writers against 0.013 for native English writers, and conclude there is no statistically significant bias against non-native English speakers. Both claims can be on the table at once, and a UK panel should weigh them rather than pick one. What you need is not a winner in that argument. It is evidence of your own process.
What does a UK investigation actually look at?
The Office of the Independent Adjudicator for Higher Education, which reviews student complaints about providers in England and Wales, has published a casework note on complaints relating to AI and academic misconduct. It is the closest thing the UK has to a national description of a fair process, and it is worth knowing before you are ever in one.
- The burden sits with the university. The note states that the responsibility is on the provider to prove that the student has done what they are accused of doing, not on the student to disprove it.
- Detection evidence has to be weighed. Decision-makers should understand the strengths and limitations of detection software and weigh this evidence carefully against other available information.
- Drafts and notes are asked for. The note says it is helpful to explicitly ask students to supply copies of any notes or drafts.
- Metadata has to be explained. Where a provider relies on electronic file metadata, it should explain what it thinks the metadata shows and give the student an opportunity to respond.
- A viva tests understanding, not nerves. Providers may hold a viva to test the student's understanding of the work, focusing on the content and on how the submission was prepared.
- Your earlier work is a comparator. Providers may compare the work under scrutiny to other assessed work you previously completed.
On outcome, the note says a decision-maker should explain its choice of penalty and why lesser penalties were not suitable, and describes an educative rather than punitive approach to minor or first instances as good practice.
What should you be keeping while you work?
Everything in that list is something you either have or you do not, and you cannot manufacture it afterwards. Build the trail while the work is happening.
- Version history. Write in a tool that keeps it. Google Docs and Word with autosave both retain a revision timeline that shows a document growing over weeks rather than appearing in one evening.
- Dated drafts. Save a separate file at each real milestone: outline, first full draft, post-feedback draft, final. Four files across a month are more convincing than a hundred autosaves.
- Your reading. Notes with page numbers, the PDFs themselves, your library loan history. This is what lets you talk about content in a meeting.
- Supervisor and tutor contact. Emails, feedback, tutorial notes. A supervision trail is the strongest evidence a long piece of work has, which is one reason the dissertation process is built the way it is, as set out in writing a dissertation at a UK university.
- Your AI acknowledgement, if you used it. Prompts and outputs saved as you go, not reconstructed.
Five minutes at the end of each work session
- Save a copy of the document with the date in the filename if you passed a milestone.
- Paste any new sources into your reference manager with a one-line note on why they are there.
- Write two sentences to yourself about what you decided and why.
- If you used a generative tool, add the prompt and the output to the same running file.
That last file is also a working document. Four weeks later it is the answer to "talk me through how you wrote this".
What do you do if you are accused?
Slowly, and in writing.
- Ask what the allegation is and what it rests on. You are entitled to know the specific assessment, the specific claim and the evidence. If a detection score is part of it, ask for the report and for the university's own guidance on how that report may be used.
- Get your students' union involved. Students' unions run advice services for exactly this, they know the local procedure, and an adviser can usually attend the meeting with you.
- Assemble the trail, do not argue the score. Drafts, version history, notes, reading, emails. The OIA note tells universities to ask for exactly this, so it is the evidence the process is built to accept.
- Prepare to talk about content. Why this structure, what you read, what you cut and why, what the weakest part of your argument is. A viva is a conversation about your work, and it is the part students underprepare for.
- Answer the metadata question if it comes up. Writing on a borrowed laptop, in a library account or across two devices produces odd file properties for entirely ordinary reasons. Say so, with specifics.
If the outcome is wrong
Use the internal appeal first, on the grounds the procedure allows, usually procedural irregularity, new evidence or a penalty out of proportion. When the internal route is exhausted you receive a Completion of Procedures Letter, and only then can you go to the OIA, which describes itself as the independent, free student complaints body for higher education providers in England and Wales.
One more thing you can do while a case is open: the evidence the university holds about you is your personal data, and you can ask to see it. How a subject access request works in practice is in your data rights as a UK student.
Does editing software push your score up?
It can, and that is a different problem from cheating. The indicator responds to the statistical shape of the finished text, so a paragraph you wrote and then ran through a rewriting tool several times has moved towards the patterns the model associates with generated prose, even though every idea in it is yours. This is why the sensible defence is a trail of drafts rather than restraint about tools you were allowed to use.
The question of which of those tools you may use on work you wrote yourself, and what a human proofreader may change, is a policy question rather than a detection one. It is answered in proofreading and Grammarly rules at UK universities.
Where Notibo fits
Notibo does not write assignments and it has nothing to do with detection. It records a lecture, or takes an audio file you upload, and returns a transcript, structured notes and flashcards with spaced repetition, and it turns PDF, PPTX and DOCX files into the same. What that gives you in this context is a dated record of the teaching your work is built on, in 89 languages, with the transcript in the language that was spoken.
The data answer, plainly. Files are stored in the EU, on Supabase in Frankfurt, while transcription and the AI notes run on US processors under standard contractual clauses. Uploads are capped at 50 MB, and long recordings are saved in smaller parts as they run and joined back into one transcript. You can try Pro free for 14 days, with 240 minutes of recording and no card. After that the free plan covers 30 minutes of recording a month. Pro is EUR 9.99 a month or EUR 88.99 a year. Ask permission before you record a lecture.