Turning an interest into a question you can answer.

Almost everyone arrives with a topic. "I want to research AI and education." "Something about mental health in students." "Climate policy in Nigeria." These are fine as starting points and useless as research questions, because none of them tells you what to go and do on Monday morning.

The difference is not that a question is narrower. It is that a question contains instructions. Read a good question and you know who to study, what to measure, and what you would compare it against. Read a topic and you know nothing except the general area of the library.

The three things a question has to contain

1. A population. Who or what. Not "students" / second-year undergraduates in one department, or secondary school teachers in two states.
2. An outcome. The thing that might change, expressed so that two people would measure it the same way.
3. A comparison or a variation. Compared with whom, before and after what, or across what range.

If your question is missing one of the three, it is not ready. That is the entire test.

Worked conversion one

Topic: AI and education.

First attempt: How does AI affect learning? This has no population, no measurable outcome, and no comparison. It is also a question no single project could answer.

Second attempt: Do students who use ChatGPT write better essays? Better on what dimension, judged by whom? "Better" is doing far too much work.

Ready: Among second-year undergraduates in one department, do essays drafted with an AI assistant differ from unassisted essays in the number of cited sources and in tutor marks for argument structure?

Notice what happened. The scope shrank and the project became possible. You now know you need essays, a marking rubric you did not invent yesterday, and two groups. You can picture the spreadsheet.

Worked conversion two

Topic: Student mental health.

Ready: How do final-year students describe the point at which academic pressure became unmanageable, and what support did they try first?

This one is qualitative, and it still has all three parts: final-year students, their account of a threshold moment, and variation across the support routes they tried. It does not measure anything numerically and it is not weaker for that. It answers a "how" and "why" question, which is what interviews are for.

Worked conversion three

Topic: Climate policy.

Ready: Between 2015 and 2024, how did the stated commitments in Nigeria's national climate plans change in specificity, comparing the energy and agriculture sections?

Population: the documents. Outcome: specificity, which you would need to define / deadlines named, figures given, agency named. Comparison: two sectors, across a decade. No fieldwork, no ethics approval, and a real finding at the end of it.

The four questions to ask your own draft

• Could two people disagree about the answer? If not, you have a fact-finding exercise, not research.
• Do I know what data would answer it? Name the actual thing: 30 interviews, 200 essays, 12 policy documents.
• Could I get that data in the time I have? This kills more projects than anything else. Cut the scope now, not in month four.
• If the answer turned out to be no, would it still be worth writing? A question that is only interesting if it confirms your hunch is a bad question.

Do this on paper, once

Write your topic at the top. Underneath, write three candidate questions / not one, three. Then fill in population, outcome and comparison for each. One of them will be obviously easier to picture doing. That is usually the one to take, and the other two become your limitations paragraph and your future-work paragraph later.

Ten minutes of this saves weeks. The most expensive research mistake is not a bad method; it is six weeks of reading in service of a question you had not yet written down.