AI is now part of the research landscape. The challenge is not whether to use it, but how to use it well. Used thoughtfully, AI can support learning and sharpen thinking. Used uncritically, it can undermine the very skills researchers are trying to develop.

Most researchers reach a point where AI produces something genuinely impressive — a summary, a structure, a first draft — in the time it takes to make a cup of tea. The reaction is often mixed: relief, and then a flicker of unease. If it can do that, what is my role here?

This guide does not sidestep that question. It is built around it. Because how you use AI — and what you allow it to do on your behalf — will shape the kind of researcher you become.

There is a distinction worth noting as you read this guide: the difference between doing research and directing it.

Doing is the execution layer — the searching, drafting, summarising, and formatting. AI is already competent here, and increasingly so. Directing is something else: deciding which question deserves your attention, recognising which direction within a project is most promising, knowing when a result is genuinely good rather than merely plausible. Directing requires a deep familiarity with your field — a feel for the terrain that tells you where to go next.

This is important for both ethical and practical reasons. AI is most powerful in the hands of researchers who can direct it well — who know what to ask for, how to evaluate what comes back, and what it means for the work they are actually trying to do. That directing capacity does not come from having access to the tools. It comes from having developed the judgement to use them. And judgement develops through engagement with the substance of research — exactly the engagement that AI makes it easiest to skip.

Evidence-based practice depends on a hierarchy of evidence — systematic reviews and meta-analyses at the top, expert opinion and anecdote at the bottom. AI outputs occupy no recognised place in this hierarchy. They are not a source of evidence. They are a tool for processing, organising, and communicating evidence that must originate elsewhere.

In practice, this means:

The distinction often comes down to whether you remain cognitively active — interpreting, questioning, and deciding. This applies across all phases of research, not just the writing-up stage: data collection, instrument design, qualitative analysis, and knowledge translation into practice are equally governed by these principles.

A useful test: your name is on the work. The expectation is that the ideas, reasoning, and interpretations are your intellectual contribution. If AI is constructing the argument or speaking in your place, it is no longer supporting your work — it is replacing it.

Systematic reviews are the gold standard of evidence synthesis in health research. They operate under strict methodological protocols — including PRISMA reporting standards, PROSPERO registration, structured database searching, and formal risk of bias assessment. AI is increasingly being used at various stages of this process, and the risks are specific and significant.

What AI cannot do in a systematic review

The interpretive, reflexive engagement with participants' words and experiences is not a preliminary step in qualitative research — it is the research. Using AI to generate themes from interview data, summarise field notes, or suggest codes risks introducing researcher distance from the data and undermining the methodological rigour that gives qualitative work its validity and trustworthiness.

This does not mean AI has no role in qualitative research. It may legitimately support literature searching, help structure a methodology section you have already drafted, or improve the clarity of your writing. But the analytic process itself — the sustained, reflexive engagement with data — must remain yours.

Do not input unpublished data, participant information, interview transcripts, or any confidential material into public AI systems. Doing so may breach your ethical approval, your institution's data protection policy, and GDPR.

Disclose AI use as required — but note that disclosure does not make a use acceptable. Policies vary across institutions, programmes, and journals, and are still evolving. Disciplinary norms also vary: what is standard in computational research may be inappropriate in qualitative health research. Seek field-specific guidance alongside this framework.

Journal and publisher policies on AI-assisted writing differ, including on which tools must be declared. Check these before submission.

Before inputting anything that touches participant data into any AI tool, ask three questions:

When using AI, or including AI-supported work in your research:

If any answer is 'no' or uncertain, pause and re-engage with the task.

The researchers who benefit most from AI are not those who use it most — but those who use it most wisely, keeping their own intellectual development and professional accountability at the centre.

You are the researcher. AI is the tool. Keep it that way.