How to write a research question (three worked examples)
Nobody needs another list of adjectives a research question should have. What you need is to watch someone take a topic the size of a syllabus and cut it down, draft by draft, with the reason each draft failed stated out loud. That is what this page is: three topics narrowed in public, plus a two-minute test that tells you whether the literature can answer your question before you spend a week finding out it cannot.
- The five tests a question has to survive
- The two-minute literature dipstick
- Narrowing a climate topic to a question
- Narrowing an ocean plastics topic to a question
- Narrowing a remote work topic to a question
- The five moves that do all the narrowing
- Why the frameworks do not produce a question
- From question to thesis
- Testing a question against what exists
A research question is the one sentence that decides what you read, what you argue, and what counts as a good answer. Get it right and the paper has a spine. Get it wrong and no amount of clean prose rescues it, because you are answering the wrong thing carefully.
The five tests a question has to survive
Run any draft past these. The one it fails tells you the repair.
- Specific. You can name what is being measured or examined, in what setting. If the question contains an "and" joining two separate inquiries, you have two questions.
- Researchable. Evidence, not values, would settle it. "Should cities ban cars from their centres?" is a debate. "Has the car ban changed footfall on the streets it covers?" is a question.
- Open. A reasonable person could expect a different answer than you do. If it resolves in one lookup, there is no paper in it.
- Feasible. Answerable in the length and time you actually have. Ambition is where most projects lose their first fortnight.
- Grounded. Real published work already engages with it. This is the one students assume rather than check, and the next section is how to check it in two minutes.
The two-minute literature dipstick
Here is a trick nobody teaches. Before you commit to a question, search its exact phrases in a free academic index and read the number of results as a signal about the question itself. Not the papers, the count. It tells you in seconds whether you are holding a topic, a workable question, or a phrase the field does not use.
We used OpenAlex, which is free, needs no account, and has an open API, searching titles and abstracts for the exact phrases. All the counts on this page were run on 11 August 2026 and anyone can rerun them: the query is a request to api.openalex.org/works with a title and abstract filter, or the same phrases typed into the search box. Google Scholar works too, with the caveat that its counts are rounded estimates and include far more grey material.
| Hits on your exact phrases | What it means | What to do |
|---|---|---|
| Over 10,000 | A topic, not a question | Add a population, a mechanism, a place or a period |
| 2,000 to 10,000 | A subfield | Narrow once more, on the dimension you actually care about |
| 200 to 2,000 | The workable zone for a student paper | Read the ten most cited and the five most recent |
| 20 to 200 | A specific question | Check the hits are genuinely on your question, not just sharing words |
| Under 20 | A gap, or the wrong words | Try the field's own terms first. It is usually the words |
The bands are our judgement, calibrated on the searches below, not a rule from anywhere. The last row is the one that saves the most pain. Students read a near-empty result set as evidence they have found something original, and for an undergraduate paper it almost never is. It means you have written "buying behaviour of young people" where the literature says "consumer socialisation", and the fix is fifteen minutes with the vocabulary of the field.
Narrowing a climate topic to a question
Assigned topic: climate change in cities.
Draft 1. "How does climate change affect cities?"
Dipstick: 58,161 papers mention climate change and cities in their title or abstract. Fails specific and feasible. This is a reading list for a career. There is nothing wrong with it as a starting direction, and everything wrong with it as a question you have three weeks to answer.
Draft 2. "Are urban heat islands dangerous?"
Dipstick: 25,443 for "urban heat island". Better, because it names a mechanism instead of a phenomenon the size of the weather. Still fails researchable, because "dangerous" is not something evidence settles without deciding first what harm you are counting, to whom, and against what baseline. The vagueness is hiding a design decision.
Draft 3. "Does tree canopy cover reduce urban heat island intensity?"
Dipstick: 327. Now we are in the workable zone, and the question names two things that can be measured. But it fails open: the answer is yes, established, and a paper that spends 3,000 words confirming it has told nobody anything.
Draft 4. "Does street tree canopy reduce daytime air temperature enough to matter for pedestrians in dense neighbourhoods?"
Dipstick: 21 for those three phrases together. This is the trap the table warns about. It reads like a sharpening, but "enough to matter for pedestrians" is a value judgement wearing a measurement's clothes, and the thin result set is a symptom of that, not a discovery.
Draft 5, the keeper. "How much does street tree cover lower daytime air temperature in dense urban neighbourhoods, and does the size of the effect depend on street geometry?"
Dipstick: 70 papers mention urban heat island, street trees and air temperature together, sitting inside the 327 on canopy and heat. Specific, because it names what is measured. Researchable, because "how much" has units. Open, because the geometry half is genuinely contested: shading a narrow street and shading a wide one are not the same intervention. Feasible, because 70 papers is a literature you can actually read. Grounded in a field with a founding paper you can build on, T. R. Oke's account of why urban surfaces store and release heat the way they do (Quarterly Journal of the Royal Meteorological Society, 1982, doi.org/10.1002/qj.49710845502).
Watch what changed between drafts 2 and 5. Every edit replaced a word that hid a decision with a word that states one: "dangerous" became "daytime air temperature", "reduce" became "how much", and a second clause arrived carrying the part that is actually arguable. That is the whole move.
Narrowing an ocean plastics topic to a question
Assigned topic: ocean plastic pollution.
Draft 1. "Is plastic bad for the ocean?"
Fails open before any search is needed. The answer is known and nobody is arguing.
Draft 2. "How much plastic enters the ocean each year?"
Dipstick: 1,309 for "marine plastic pollution". Researchable, and it looks like a real question, but it fails open in a subtler way: it asks for a number that published estimates already supply, so the paper becomes a report of somebody else's figure.
Draft 3. "Which sources of ocean plastic should policy target first?"
Fails researchable on the word "should". Evidence can tell you where the plastic comes from and how concentrated it is; the ranking of what to do about it is a values argument sitting on top.
Draft 4, the keeper. "How concentrated are riverine plastic emissions in a small number of rivers, and why do published estimates disagree?"
Dipstick: 200 papers on riverine plastic, 108 on rivers and plastic emissions. This one passes every test, and it passes open emphatically, because the field openly disagrees. Lebreton and colleagues estimated in 2017 that the top 20 polluting rivers account for 67 percent of the global total (Nature Communications, doi.org/10.1038/ncomms15611). Meijer and colleagues, in 2021, put it very differently: their title is "More than 1000 rivers account for 80% of global riverine plastic emissions into the ocean" (Science Advances, doi.org/10.1126/sciadv.aaz5803). Same quantity, two models, answers with completely different policy consequences.
Draft 4 is the strongest question on this page and it is worth saying why. A question whose published answers disagree hands you your paper's structure for free: here is estimate A, here is estimate B, here is what the models assume differently, here is what would settle it. You are not required to resolve the disagreement. You are required to explain it, which is a paper an undergraduate can genuinely write.
Narrowing a remote work topic to a question
Assigned topic: remote work.
Draft 1. "Is remote work good for companies?"
Dipstick: 14,883 for "remote work". Fails specific and researchable at once. "Good" is undefined, "companies" is every organisation on earth, and no evidence answers it as written.
Draft 2. "Does remote work increase productivity?"
Dipstick: 2,368. Sounds like a real question and is the version most students submit. It fails on a hidden ambiguity: productivity means output per hour to an economist, tickets closed to a manager, and self-reported effectiveness in most survey work, and those three measures do not agree with each other. A question that changes its answer depending on an undeclared definition is not yet a question.
Draft 3, the keeper. "For software teams that moved to fully remote work, does the reported productivity effect depend on whether productivity is measured as delivered output or as self-reported effectiveness?"
Dipstick: 213 papers on remote work, productivity and software together. Specific about the population, researchable because both measures are reported in the literature, open because the disagreement between them is the finding, and feasible because 213 papers is a fortnight of screening, not a career.
The move in draft 3 is one worth stealing. When a term in your question is ambiguous, do not pick one meaning quietly. Make the ambiguity the question. Papers that ask "does the answer depend on how we measure it?" are more interesting than papers that ask "what is the answer?", and they are much easier to write honestly, because you cannot be wrong about a disagreement that exists.
The five moves that do all the narrowing
Every edit above is one of five moves. When a question is stuck, run down this list and try each one on it.
- Name the population or the setting. Which people, which organisations, which cities, which period. "Cities" became "dense urban neighbourhoods"; "companies" became "software teams".
- Replace the vague verb with a measurable one. "Affect" and "impact" hide the decision about what you are counting. "Lower daytime air temperature" cannot.
- Swap "whether" for "how much" or "under what conditions". Yes-or-no questions are usually already answered. Magnitude and conditions are usually not.
- Surface the ambiguity instead of resolving it privately. If a key term has three meanings in the field, the question is which meaning changes the answer.
- Find the disagreement. Two credible published answers to the same question is the best gift a topic can give you, and the dipstick's job is partly to find it.
Why the frameworks do not produce a question
You will be pointed at acronyms. FINER asks whether a question is feasible, interesting, novel, ethical and relevant. PICO frames a comparison as population, intervention, comparison and outcome, and it genuinely helps when your work weighs one course of action against another: "among first-year undergraduates, does weekly peer tutoring, compared with drop-in office hours, raise end-of-term exam scores?" fills the four slots and comes out sharper than it went in.
Our honest view, though, is that these are graders of questions rather than generators of them. FINER will tell you a question is not feasible, which you knew; it will not tell you which of the five moves above to apply. If you have half an hour to spend on your question, spend twenty-five minutes on the dipstick and the narrowing and five on the acronym, not the other way round. The reason is simple: a checklist can only evaluate the drafts you have already thought of, and the difficulty of writing a research question was never the evaluation.
From question to thesis
The question asks; the thesis answers. You write a working question before you read much and a working thesis once you have read enough to take a position. "How concentrated are riverine plastic emissions, and why do published estimates disagree?" is the question. "The disagreement between the 2017 and 2021 estimates is driven by how each model treats small rivers, which means the policy implication depends on a modelling choice rather than on the data" is the thesis it earns. Turning a question into a defensible claim is covered in how to write a thesis statement, and the sections that answer it in how to build a research paper outline. For a longer project, the same narrowing runs at chapter scale in how to write a bachelor thesis.
Testing a question against what exists
The dipstick tells you how much literature exists. It does not tell you what that literature says, and that is the next hour of work: reading enough to know whether your question has already been answered, and by whom. CiteOwl does that part on a verify-first principle. It searches real academic databases and reads what it finds, so when you are circling a question it can show you what the field has already covered and where the disagreements sit, with the quote behind every claim. The papers are real, not generated from memory, which matters here more than anywhere: a question shaped against invented literature is a question shaped against nothing.
From there it can take the question you settle on, structure the paper around it, find and cite real sources, and draft sections you review as plain diffs you accept or reject. The question and the argument stay yours. Finding sources by hand works too, and the dipstick above needs nothing but a browser.