Yes, but only as a support tool and within the limits your university sets. Leading universities do not ban ChatGPT: guidance at Cambridge, UCL and Edinburgh allows it for brainstorming, outlining or grammar checking, yet treats submitting unacknowledged AI-generated text as academic misconduct. You must acknowledge the use. What decides in the end is the policy of your own university, your department and the brief for each individual assessment.
Is using ChatGPT plagiarism?
Not automatically. Three different things get mixed up here, although each means something else:
- Plagiarism is presenting someone else's text or ideas as your own without crediting the source. If AI rephrases a paragraph from a book and you paste it into your thesis without citing the original author, that is plagiarism, but not because you used AI.
- Unauthorised assistance means that work which should be the result of your independent effort was produced by someone or something else. UK universities usually handle it under the broader label of academic misconduct, alongside contract cheating. This is where most problems with generative AI sit.
- A breach of academic integrity is the widest term. It also covers concealing how the work was created, which includes undeclared use of an AI tool your university expects you to acknowledge.
University rules reflect this distinction. Cambridge puts it in one sentence: "A student using any unacknowledged content generated by artificial intelligence within a summative assessment as though it is their own work constitutes academic misconduct, unless explicitly stated otherwise in the assessment brief." So this is not about one narrow plagiarism rule, but about a whole framework of academic integrity. How the text-matching side of it works is covered in our article on plagiarism checks and originality.
What do university policies actually say?
One important warning first: there is no single national rule. Every university publishes its own guidance, departments and programmes refine it, and for a specific module or dissertation the assessment brief and your supervisor's instructions decide. Below are three universities whose documents are publicly available.
| University | Document | Core rule for students |
|---|---|---|
| University of Cambridge | AI guidance within its plagiarism and academic misconduct pages | Unacknowledged AI content in a summative assessment counts as academic misconduct unless the brief says otherwise |
| UCL (University College London) | Student guide "Engaging with Generative AI in your education and assessment" and three categories of GenAI use | Every assessment falls into one of three categories; acknowledge GenAI wherever it assisted your work |
| University of Edinburgh | "Guidance for working with Generative AI ('GenAI') in your studies" (October 2024) | Learn, don't copy: never paste AI output into assessed work; credit the tools you used |
Cambridge: fine for personal study, misconduct in assessment
Cambridge's guidance on artificial intelligence draws the line between learning and assessment. Using AI tools appropriately for personal study and formative work is acceptable, though the university recommends discussing it with your supervisor first. For summative work the sentence quoted above applies: unacknowledged AI content presented as your own work constitutes academic misconduct, unless the assessment brief explicitly states otherwise.
Note the two safety valves in that rule. It only bites where the content is unacknowledged, and an assessment brief can expressly allow more. Staff are expected to communicate their expectations clearly, so if the brief is silent, ask before you submit, not after.
UCL: three categories for every assessment
UCL's student guide works with a simple system: departments assign each assessment to one of three categories of GenAI use. In Category 1, "The purpose and format of these assessments makes it inappropriate or impractical for GenAI tools to be used." In Category 2, "Students are permitted to use GenAI tools in a limited, critical and responsible way. The student should still be the author of their own work." In Category 3, "GenAI is used as a primary tool during the assessment process."
Two duties sit on top of the categories. First, transparency: "You should acknowledge GenAI use where it has been used to assist in the process of creating your work." Second, authorship: "It is not acceptable to use GenAI tools to write your entire assessment and present this as your own work." UCL also reminds students why checking matters, because "GenAI tools can provide misleading or incorrect information."
Edinburgh: golden rules and a model acknowledgement
Edinburgh's guidance for students is the most hands-on of the three. Its first golden rule reads: "Use GenAI to aid your learning, but never copy-paste any GenAI outputs into your own assessed work. Doing so constitutes academic misconduct." Acceptable uses include brainstorming, explaining difficult concepts, planning and structuring your writing, summarising notes and improving grammar and spelling.
The guidance is also reassuring: "You will not be penalised for using GenAI, as long as you are working within the acceptable uses of GenAI outlined, and you are adhering to any other restrictions that have been specified for a particular assessment." The hard boundary stays in place: "It is not acceptable to present AI-generated content as your own work. If you do, this will be regarded as academic misconduct."
Which AI uses do universities usually accept and which are risky?
Comparing the three documents above gives a fairly consistent picture.
Low risk (usually accepted, often without a formal declaration):
- brainstorming topics, questions and possible angles,
- getting an unfamiliar concept or method explained in plain words,
- drafting an outline and chapter structure,
- checking spelling, grammar and style of your own text,
- an orientation search, after which you find and read the real sources yourself.
High risk (either must be acknowledged or is prohibited):
- generating whole chapters, the introduction, discussion or conclusions,
- taking over a reference list that AI produced for you,
- translating a foreign-language source and passing it off as your own writing,
- using AI during an exam or test without permission,
- letting AI interpret your own data instead of you.
The difference between the two groups is not how much AI you use, but whether a human remains the author of the ideas and conclusions. Background research, summaries and consultations are a legitimate form of help precisely because you stay the author of the text; if you need that kind of support, get in touch with us.
How to acknowledge and reference AI use
Universities differ in the format, not the principle. An overview of the options:
- Check the category or brief first. At UCL every assessment has an assigned GenAI category, and at Cambridge the assessment brief can explicitly allow or restrict AI. What you must declare follows from that document, so read it before you write a word.
- A short acknowledgement statement. Edinburgh asks for a minimum of the name and version of the tool, the publisher, the URL and a one-sentence description of how you used it. Its model wording: "I acknowledge the use of ELM (University of Edinburgh, Edina, https://elm.edina.ac.uk/elm/elm) to help me generate initial ideas and proof read my final draft."
- An appendix when more detail is required. For some assessments Edinburgh expects the prompts you used, the date the output was generated, the output or an extract of it, and how it was edited and incorporated into the work.
- A formal reference for quoted output. If you quote or closely paraphrase an AI answer, cite it like any other source. Referencing styles commonly used at English-speaking universities, such as APA and Harvard, publish formats for generative AI; confirm with your department which one applies.
A more detailed guide to reference formats is in our article on citing online sources.
Why does AI invent sources?
This is the most practical risk for a thesis, and universities name it directly. Edinburgh explains that GenAI tools are "language machines rather than databases of knowledge" and warns that "They can occasionally produce fake citations and references." Its golden rule for facts: "Always check GenAI output for factual accuracy, including references and citations."
In practice it looks harmless: you get a citation with a real author's name, a credible journal title, a year and page numbers, and the study does not exist. An examiner will spot it when checking your reference list. The rule is therefore simple: never cite a source you have not opened yourself. Trace every AI-supplied claim back to the original and cite the original.
Are AI detectors reliable?
They are not infallible, in both directions. They flag text you wrote yourself and miss edited model output. That is why the university documents above put the weight on acknowledgement and on your ability to explain your work, not on a percentage from a tool. Edinburgh notes that while you may use GenAI to aid your learning, you may be required, for example in an exam, to demonstrate your abilities without such tools. Where misconduct is suspected, it is investigated under formal procedures in which people, not detectors, make the decision. How detection works and where its limits are is covered in our article on how AI-generated text is detected.
What happens if you break the rules?
The exact process depends on your university's academic misconduct procedures. Edinburgh defines academic misconduct as "the use of unfair means in any University assessment" and investigates suspected cases under its formal investigation procedures. Cambridge treats unacknowledged AI content in summative assessment as academic misconduct outright.
There is no single statute behind this; each institution applies its own regulations. The outcomes usually range from a mark penalty or failing the assessment to, in the most serious or repeated cases, exclusion from the programme. The severity typically depends on the extent of the misconduct, your intent and whether it is a first offence.
Checklist: what to ask your supervisor in advance
Five minutes in a consultation protects you more than any article. Find out:
- Does our department or programme have its own AI guidance, or does only the university-wide policy apply?
- Which activities can I do with AI without declaring them, and which must I acknowledge?
- Where exactly should the acknowledgement go: in a statement at the front, in the methodology, or in an appendix?
- What format do you want for the declaration and references (tool, version, date, prompts, an appendix with the conversation)?
- Is it acceptable to use AI for language editing of my finished text?
- Will I be expected to explain at the viva or defence exactly how I used AI?
Write the answer into your consultation notes and keep the email. If anyone asks later, you have proof that you followed instructions.
Frequently Asked Questions
Can I have ChatGPT write my whole dissertation?
No. Cambridge classes unacknowledged AI content in summative assessment as academic misconduct, UCL states it is not acceptable to have GenAI write your entire assessment and present it as your own, and Edinburgh treats pasting AI output into assessed work as misconduct. Even if the text passed an originality check, it would still breach academic integrity, with disciplinary consequences.
Do I have to declare that I only used AI to fix grammar?
Edinburgh lists improving grammar and spelling among acceptable uses and asks you to credit tools where used; UCL expects you to acknowledge GenAI wherever it assisted the process. Requirements differ between universities and even modules, so confirm with your module lead or supervisor.
Is it plagiarism if I rewrite the AI output in my own words?
Plagiarism arises when you use someone else's content without crediting the source. Rewriting the output does not release you from two duties: verifying the facts in the original sources and acknowledging the tool in the way your university requires. Without acknowledgement it can count as academic misconduct.
How do I reference an answer from ChatGPT?
As a minimum, Edinburgh expects the tool's name and version, the publisher, the URL and a one-sentence description of how you used it; some assessments also require the prompts and outputs in an appendix. If you quote the answer directly, use the AI format of your referencing style. Confirm the exact form with your supervisor.
Do the same rules apply to coursework and exams?
Not necessarily. At UCL the permitted use depends on the category assigned to each individual assessment. Edinburgh reminds students that in an exam you may need to demonstrate your abilities without AI tools. Always check the brief for the specific piece of work.
What if I am not sure and my supervisor has no clear view?
Ask for a written position from the module lead or your department office, and document your AI use in more detail than seems necessary. Transparency is the common denominator of all three university policies.