Blog
How can teachers tell if you used ChatGPT?
Mostly by knowing your work, not by software. Exam board guidance points to a mismatch with your usual writing, references that cannot be found and a talk about the work. Detector scores are not proof on their own.
Short answer: the main check is a teacher who knows your work. A detector score is one piece of evidence at most. The exam boards’ guidance says this, and so do OpenAI and Turnitin, in their own words.
The signals teachers actually use
The most detailed UK source is the Joint Council for Qualifications (JCQ) guidance, AI Use in Assessments, dated 30 April 2025 and written for heads of centre, teachers and assessors. It says identifying AI misuse “requires the same skills and observation techniques teachers are already using” to check that work is a student’s own.
The first of those is comparing new work with your previous work, especially work done in class or under supervision. JCQ lists what to compare: spelling and punctuation, grammar, writing style and tone, vocabulary, complexity and coherency, general understanding and working level, and whether the work is handwritten or typed.
It then lists potential indicators of AI misuse. A selection:
- American spelling, currency or terms used by default
- vocabulary that does not fit the qualification level
- references that cannot be found or verified (JCQ notes some AI tools have produced false references to books or articles by real authors)
- a style that differs from how you write in class, or that shifts partway through a piece
- generic content where the task asked for something personal, local or specific
- AI warnings or provisos left in the text by accident
- confidently incorrect statements inside otherwise cohesive writing
The point about fake references fits the Department for Education2’s guidance on generative AI in education. It says these tools “can produce nonsensical, inaccurate or false information presented as fact, known as hallucination”.
None of these signals needs software. A teacher who has read your work all term will notice them. In JCQ’s words: “Teachers will know their students best”.
What AI detectors can and cannot show
JCQ names several detection tools, including Turnitin’s, and says they may be used as a check or to verify concerns. It also says they “vary in accuracy” depending on the AI tool, how much of the text is AI-written, the type of prompt and other factors, including a person’s English language competency. Its conclusion is that detection tools, where used, “should form part of a holistic approach” and that all available information must be considered.
OpenAI, which makes ChatGPT, is blunter. Its educator help page asks “Do AI detectors work?” and answers: “In short, not in our experience.” It says its research found detectors were not reliable enough, given that teachers could be making judgements with lasting consequences for students. When OpenAI tried to train a detector, it labelled human writing such as Shakespeare as AI-generated. The page also says there were indications that such tools could unfairly affect students learning English as a second language and students whose writing is formulaic or concise. ChatGPT itself has no knowledge of what content is AI-generated. If you ask it “did you write this?”, OpenAI says its answers “are random and have no basis in fact”.
Turnitin, which sells a detector, says it “does not make a determination of misconduct” and that its percentage “should not be used as the sole basis for action or a definitive grading measure”. It says it aims to keep false positives under 1% for documents with over 20% AI writing, which it explains as possibly flagging one in every 100 fully human-written documents. For results below 20% it shows an asterisk and no percentage. Those figures come from Turnitin’s own testing. Turnitin also says students cannot see the AI report unless an instructor downloads it and shares it.
The exam boards, the maker of ChatGPT and a detector vendor all reach the same conclusion: a score can prompt a closer look, but it does not settle the question.
Why “we can talk about this work” is the check that matters
JCQ lists steps schools and colleges should take to prevent misuse. One is to “consider whether it is helpful to engage students in a short verbal discussion about their work to ascertain they understand it and it reflects their own independent work.” Some private candidates have teachers who do not know their usual standard. For them, JCQ suggests supervised work, a review of their portfolio and a short discussion about the work.
A conversation tests something a detector cannot: whether you understand the piece. If you wrote it, you can say why you chose that argument, what you cut and where the evidence came from. OpenAI’s page comes at this from another angle. It suggests teachers ask students to share their ChatGPT conversations, so they can see how the student worked.
Version history is another record a teacher may look at. Google’s help page says Google Docs shows earlier versions of a file and who made changes, and that you need edit permission for the file to browse earlier versions. The sources do not say how often UK teachers check it.
If you are accused and you wrote it yourself
- Ask what the concern is based on. If it is a detector score, you can fairly point to the JCQ, OpenAI and Turnitin passages above. None of them treats a score as proof on its own.
- Bring your evidence. That means notes, plans, drafts, the sources you read and any edit history.
- Offer to talk through the work. Explaining your own choices is the most direct answer to a doubt.
- Ask for the written policy and follow its process. For exam board work, JCQ says that if misuse is suspected after you have signed the declaration of authenticity, the case must be reported to the awarding organisation under its published malpractice procedures. At university, the Russell Group3 principles say policies will vary by discipline. They also encourage an environment where students can ask about specific cases “openly and without fear of penalisation”.
- Mention it if it applies. If English is not your first language, or your writing is plain and concise, you can say that OpenAI’s own page names these students as possibly affected by detectors.
If you did use AI where it was not allowed
This post will not help anyone get past a check. It can set out the stakes. JCQ says students who misuse AI so that the work they submit is not their own have committed malpractice. The sanctions for “making a false declaration of authenticity” and for plagiarism include disqualification and debarment from taking qualifications for a number of years.
The declaration is the turning point. JCQ says any sections reproduced from AI must be identified by the student and will not earn credit. Once misuse is suspected and the declaration has been signed, the case must be reported. If you have not signed yet, tell your teacher now and acknowledge the AI use properly. If you have already signed, talk to your teacher or tutor before it goes further. No source here promises a particular outcome. That depends on the rules that apply to your work.
What the sources do not settle
- Real-world false positive rates in UK classrooms. Turnitin publishes its own target and testing. No source here measures how often UK students are wrongly flagged.
- How often teachers use detectors or check version history. None of the sources reports this.
- Whether any detector can prove AI authorship. None of the sources claims one can, and OpenAI says detectors are not reliable enough.
If you see a confident accuracy figure online, check who measured it, and on what writing.
How to do it well
If your course allows AI help:
- Check the rule for this piece of work. The DfE suggests schools review their homework policies with generative AI in mind, and university rules vary by department. Can you use AI to do your homework? sets out where the lines usually sit.
- Use AI to understand, then write yourself. Using AI to help you study covers ways to keep the work yours, and summarising a PDF with AI covers help with reading.
- Verify every reference before it goes in. JCQ says sources from an AI tool must be checked by the student and referenced in the normal way.
- Acknowledge AI use properly. JCQ says writing just “AI” or “ChatGPT” is not enough. Students should keep a non-editable copy, such as a screenshot, of the questions and the AI output, with a brief explanation of how it was used.
- Keep your drafts as you go. They are your record if anyone asks.
University applications have their own considerations. See using AI on a personal statement.
Rules and risks
For exam board coursework, the JCQ guidance is the benchmark: teachers “must only accept work for qualification assessments which is the students’ own”. For ordinary homework, your school’s own policy applies. For university work, your institution’s and department’s policies apply.
The serious sanctions in these sources attach to submitting work that is not your own and to falsely declaring that it is. Both are within your control. This is general information, not legal advice.
Sources
6- JCQ: AI Use in Assessments: Your role in protecting the integrity of qualifications (30 April 2025)
- Department for Education: Generative artificial intelligence (AI) in education
- Russell Group: Principles on the use of generative AI tools in education
- OpenAI Help Center: How can educators respond to students presenting AI-generated content as their own?
- Turnitin Guides: AI writing detection capabilities FAQs
- Google Docs Editors Help: Find what's changed in a file
FAQ
Questions people ask
Can teachers actually tell if you used ChatGPT?
Often, but mainly by knowing your work. JCQ guidance tells teachers to compare new work with what you produce in class and lists warning signs such as references that cannot be found, a change in style and generic content. Detector scores can prompt a closer look, but JCQ, OpenAI and Turnitin all say they should not decide the matter on their own.
Can your teacher see if you copy and paste into Google Docs?
Google's help page says Docs keeps earlier versions of a file and shows who made changes. Browsing those earlier versions needs edit permission for the file. The help page does not describe a separate paste marker, and no source here says how often teachers check version history.
Can AI detectors be wrong about my own writing?
Yes. OpenAI says its own detector research labelled human writing, including Shakespeare, as AI-generated. Turnitin says false positives are possible and that its score should not be the sole basis for action. No source measures how often this happens in UK classrooms.
Can ChatGPT tell my teacher whether it wrote something?
No. OpenAI says ChatGPT has no knowledge of what content is AI-generated. When asked questions like "did you write this?", its answers are random and have no basis in fact.
Can I see my Turnitin AI score?
Not directly. Turnitin says the AI writing indicator and report are not visible to students. Instructors can download the report as a PDF and share it. If you are accused, it is reasonable to ask to see it.
All posts · This post as markdown. General information for UK readers, not legal, financial or medical advice.