A research question is narrow enough when it tells you what evidence you need, but broad enough that answering it still requires genuine analysis. If the question could be answered with one fact, it is probably too narrow. If it could support three dissertations, it is too broad.
The easiest way to narrow a question is to move through a series of deliberate constraints: phenomenon, relationship, population, context, timeframe and method. You do not need to use every constraint. Each one should make the project clearer rather than merely smaller.
If you have not yet established what the literature leaves unresolved, start with how to find a research gap. A beautifully narrow question that does not address a meaningful problem is still a weak research project.
What makes a good research question?
Monash University’s research-question guidance stresses that a useful question should be researchable, appropriately scoped and analytical rather than simply descriptive.
Use five tests:
| Test | Ask |
|---|---|
| Clarity | Can a reader understand the key concepts without guessing? |
| Scope | Can the project answer the question within the available degree, time and word limit? |
| Evidence | Can you obtain data, texts, cases or other material capable of answering it? |
| Analysis | Does the question require interpretation, explanation, comparison or evaluation? |
| Contribution | Would a credible answer improve understanding of a defensible research gap? |
The research-question funnel
Start broad, then make one meaningful decision at a time.
- Topic: What broad field are you interested in?
- Problem: What is uncertain, contested or practically important?
- Relationship: What are you trying to explain, compare, evaluate or understand?
- Population or material: Who or what will you study?
- Context: Where or under what conditions?
- Time: Does a period or event matter?
- Feasibility: What can you realistically access and analyse?
Do not add a population, place and year simply to make the title look precise. A boundary belongs in the question when it matters to the explanation or design.
Worked example 1: from “AI and students” to a question
| Stage | Version | Problem |
|---|---|---|
| Topic | AI and university students | Too broad to determine evidence or method. |
| Problem | Does generative AI affect student learning? | “Learning” and “use” are still undefined. |
| Relationship | How does generative-AI feedback affect revision? | Better, but population and evidence remain open. |
| Population/context | How does optional generative-AI feedback influence revision behaviour among first-year computing students? | Now the project has a clearer unit of study. |
| Operational version | How does access to optional generative-AI feedback change the number and type of revisions first-year computing students make between draft and final submission? | Specific enough to suggest data and analysis. |
The final question is not universally “better.” It is better for a particular study because the concepts can be observed and compared.
Worked example 2: when narrowing goes too far
Suppose your initial topic is remote work and early-career software developers.
Too broad: How does remote work affect software developers?
More focused: How does fully remote work affect informal learning among early-career software developers?
Possibly too narrow: How does working remotely on Tuesdays affect informal learning among first-year Java developers at three fintech firms in Melbourne during June 2027?
The last version contains impressive detail but may reduce the project to an arbitrary slice unless Tuesdays, Java, fintech and June are theoretically or practically important.
1. Replace vague outcomes
Words such as “impact”, “success”, “experience”, “effectiveness” and “wellbeing” can hide several different questions.
Ask what the outcome actually means.
- “Academic success” could mean grades, persistence, learning or graduation.
- “Engagement” could mean attendance, platform activity, participation or cognitive investment.
- “Sustainability” could mean energy, carbon, hardware life, cost or social effects.
You do not always need a numerical measure, but you do need conceptual clarity.
2. Decide what kind of answer you need
The opening words of a question often reveal its logic:
| Question type | Typical form |
|---|---|
| Description | What patterns or characteristics occur? |
| Comparison | How does X differ between A and B? |
| Relationship | How are X and Y associated? |
| Explanation | Why or through what mechanism does X occur? |
| Experience | How do participants understand or experience X? |
| Evaluation | How, for whom and under what conditions does intervention X work? |
A descriptive question can be entirely valid. The problem is asking a descriptive question when your thesis needs to explain a mechanism, test a relationship or evaluate an intervention.
3. Let access narrow the design
Feasibility is not an embarrassing compromise; it is part of research design. If you cannot access the population, data, equipment, archive or organisation required by the question, the project does not become stronger by ignoring that fact.
Ask early:
- Can I recruit enough appropriate participants?
- Can I access the dataset or archive?
- Does the project require permission from an organisation?
- Will ethical review fit the timeline?
- Do I have the methods skills, or realistic time to learn them?
- Can the question survive if one planned source of evidence disappears?

4. Avoid questions that assume the answer
“Why does social media damage student concentration?” already assumes damage. A neutral version might ask how particular patterns of social-media use relate to concentration, or how students describe managing interruptions during study.
The question can still emerge from a concern. It should not pre-decide the finding.
5. Check the question against the literature
Read the closest studies and ask whether your supposedly new question has already been answered. The literature review matrix and literature-review guide help you compare how other researchers define the same concepts and where their designs stop.
Often, literature makes a question narrower in a useful way. You discover that one relationship is already well established, but the mechanism or boundary condition is not.
6. Test whether the method and question agree
Write down the evidence you would need to answer the question. If you cannot complete that sentence, the question may still be too vague.
Examples:
- A causal question usually needs a design capable of supporting causal inference.
- A question about lived experience needs data that can capture experience, not only administrative records.
- A question about change over time needs repeated or historical evidence.
- A question comparing groups needs comparable measures and a reason those groups matter.
A quick narrowing worksheet
Broad topic:
Specific problem or uncertainty:
What the literature already establishes:
What remains unresolved:
Main phenomenon/concept:
Relationship or comparison:
Population/material:
Context:
Relevant timeframe:
Evidence I can access:
Method that could answer it:
What I am deliberately excluding:
Provisional question:
What would make the answer useful?
Signs your question is still too broad
- You need several unrelated methods to answer different parts.
- Every noun in the question could become its own literature review.
- You cannot identify the main outcome or phenomenon.
- The project requires access to multiple organisations you do not have.
- Your supervisor keeps asking “Which part of that do you actually mean?”
- Your proposal objectives read like separate research projects.
Signs you may have narrowed too far
- The answer is a single number or factual lookup.
- Your restrictions are arbitrary rather than meaningful.
- The available sample or material becomes too small to support the planned analysis.
- The question has lost connection to the research gap.
- Even a strong answer would have little implication beyond the exact case.
Once the question is stable enough to design around, use it as the spine of your PhD research proposal. It can still evolve later. A good provisional question gives the project direction without pretending that doctoral research never changes.