Using AI on a research proposal: what supervisors allow and notice

1Parts AI can help with

Where your course allows it, these uses are the least risky:

• Brainstorming search terms and related topics.
• Asking whether a draft question is too broad, then deciding yourself how to narrow it.
• Explaining a research method before you read about it properly.
• Improving the clarity of sentences you have written, or cutting a draft to the word count.

Each keeps you as the author. The tool responds to your work; it does not produce the work.

2Parts it cannot do for you

A research proposal has to prove that your question is worth answering and that you can answer it. The question, the case for why it matters and the choice of method are where that proof lives. USC Libraries' proposal guide stresses that a proposal must show the gap in the literature and a method tied to the study's aims.

A tool cannot do this for you honestly. It has not read your sources, it does not know your skills, time or access to data, and it can invent literature that seems to support a gap. One 2023 study found fabricated references in 18 to 55 per cent of citations from ChatGPT literature reviews, depending on the version.

3What funders and universities say

Research funders have published unusually clear rules, and they are a good guide to the thinking behind university policies. UK Research and Innovation's policy, published in September 2024, says:

• Applicants must not use generative AI to generate an entire application, or sections of one, without human involvement.
• Applicants are expected to be transparent about substantive use, such as help generating ideas or hypotheses, summarising data or text, comparing literature, or generating an abstract. Disclosure does not affect assessment.
• Minimal use, such as improving the standard of English or reducing the word count, does not need to be declared.
• Sensitive or personal data about others must never be entered into a tool without consent.

Universities set their own rules for student proposals. UCL, for example, sorts assessments into three categories: no use, use in an assistive role with the student remaining the author, and use as an integral part. Its guidance tells students that the assessment brief takes precedence. Find out which applies to yours before you start; how to read your university's AI policy explains what to look for.

4A safe workflow

1. Write your own draft question and a paragraph on why it matters, before opening any tool.
2. If allowed, ask a tool to challenge the question: too broad, unclear, already answered?
3. Use any suggested sources only as search terms, and check each one in the library. See literature reviews with AI.
4. Choose and justify the method yourself, using your course materials and supervisor.
5. Keep a note of what you used, when and why, and disclose it as your course requires.

If the proposal leads to a dissertation, the rules may change by chapter: see AI in a dissertation, chapter by chapter.

More on this in AI and writing.

5FAQ