Can AI write a dissertation? What is allowed, chapter by chapter

1Start with the rules

A dissertation is usually covered by more than one rule: your university's policy, your department's handbook and any instructions from your supervisor. UCL's framework is a useful example of how these fit together. It sorts assessments into three categories (no AI use, AI in an assistive role, AI as an integral part), says the student must remain the author in the assistive category, and tells students that instructions in the assessment brief take precedence over general guidance.

Ask your supervisor directly which category, or local equivalent, applies, and write down the answer.

2Proposal and planning

Planning is where tools are most often allowed: testing whether a question is answerable, brainstorming search terms, building a timeline. The question itself and the case for it should be yours. See using AI on a research proposal.

3Literature review

The biggest risk is references that do not exist. Walters and Wilder's 2023 study of ChatGPT literature reviews found 55 per cent of GPT-3.5 references and 18 per cent of GPT-4 references were fabricated. Every source in your review must be one you have found and read. More in literature reviews with AI.

4Methodology

A tool can explain a method in simpler terms, but the choice and justification of your method are assessed judgement. If your research involves people, be careful about data: UCL's guidance warns students not to upload sensitive data to generative AI tools without meeting data protection requirements, and UK Research and Innovation's policy says personal data about others must never be entered without consent. Interview transcripts and survey responses usually count.

5Results

Some courses allow AI-assisted analysis, such as help writing or testing code, and UCL lists data analysis among possible assistive uses. Others do not. Either way, you are responsible for every number: check the analysis yourself, keep the code and outputs, and make sure you can explain each result in a viva or meeting.

6Discussion and conclusion

These chapters carry your interpretation: what the findings mean, how they relate to the literature, and what follows. They are where your own thinking is most visibly assessed, and where generated text is hardest to defend if you are asked about it. Keep AI out of the writing here unless your course explicitly allows otherwise.

7Language and proofreading

Many universities allow help with language. UCL's Academic Manual, as quoted in its AI guidance, permits a third party to check structure, fluency, presentation, grammar, spelling and punctuation, but warns that it may be misconduct if substantive changes to content are made by the reviewer or software or at their recommendation. The same line applies to tools: fixing a sentence is one thing, rewriting the argument is another. See proofreading, editing or ghostwriting.

8Keeping a record

Keep a simple log as you go: the date, the tool, what you asked, and what you did with the result. Keep your drafts and version history too. If your course asks you to acknowledge AI use, the log makes the statement easy to write; if your work is ever questioned, it is your evidence. Before you start, read how to read your university's AI policy, and for structure see how to write a dissertation.

More on this in AI and writing.

9FAQ