From a broad topic to a feasible research question

A topic is not yet a question

“Artificial intelligence in healthcare” identifies a field, not a study. A research question needs a defined problem, context and kind of evidence. Start by describing what is not understood or not working, for whom, and why it matters. Resist the temptation to add a fashionable method before the problem is clear.

Write a short problem statement in ordinary language. Explain what is already known at a broad level and the uncertainty your proposed work would address. This first version will change as you read; its purpose is to guide reading rather than to claim novelty prematurely.

Read to refine, not only to collect citations

Build a small reading matrix with the research question, setting, methods, findings and limitations of each relevant study. Distinguish a limitation mentioned by an author from a gap that actually matters for your proposed context. Check whether newer research has already addressed it.

Keep a note of how each paper changes your thinking. A long reference list is not evidence of a clear project if you cannot explain the relationship between the studies. Use original research and appropriate reviews, and record full citation details as you read.

Choose a manageable boundary

Specify the population or material, setting, outcome or concept, and time or comparison where relevant. The exact structure depends on the discipline; do not force an exploratory qualitative study into a clinical-trial template. A useful question should suggest what evidence would count as an answer.

For illustration, replace “How can AI improve hospitals?” with a narrower question about how a defined group interprets uncertainty in a specific decision-support task. This is only an example of narrowing, not a validated novel topic. The next step would be to check the literature, access and ethical implications.

Test feasibility before polishing

Ask what data, participants, permissions, tools and expertise the study would need. Separate access you already have from access you hope to negotiate. Consider ethics review, privacy, recruitment and the time available. An interesting project can become a poor proposal when its essential data cannot realistically be obtained.

Sketch the smallest credible version of the study. What could be completed if recruitment is slower, a dataset is restricted or one method proves unsuitable? A contingency should preserve the core question rather than silently change the study into something unrelated.

Prepare a one-page concept note

Use five short sections: problem, question, proposed evidence or methods, feasibility, and expected contribution. Add a few well-chosen references and the uncertainties you still need to resolve. This is often more useful for an early academic conversation than a lengthy draft full of confident but unsupported claims.

Ask a potential supervisor or reviewer whether the question is important, answerable and appropriate for the degree. Treat feedback as a way to improve the project. A proposal demonstrates informed thinking and a workable direction; it does not need to pretend that every research decision has already been settled.

Practical editorial guidance, not a guarantee of an outcome. The current programme and provider instructions take precedence.