Running UX research
UX research reduces an uncertainty that can change a decision. It is not a collection of attractive quotes or a way to make a preselected solution look validated. Useful research connects a question, relevant people or situations, a method, observations and a limit on confidence.
Simple rule — Do not ask only what people think. Observe what they try to do, in which context, using which clues, and with what result.
The research loop
Decision → uncertainty → question → method → observation
↑ ↓
← synthesis ← cautious interpretation ← data --
Before recruiting, write the decision that could change: vocabulary, journey order, feature priority or abandonment of a hypothesis. If no decision can move, research can become endless collection without learning.
1. Ask a question that teaches you something
A good question is open, situated and tied to behaviour. It does not contain the expected answer.
| Weak question | Useful question |
|---|---|
| “Do users like our new page?” | “How does someone find a course when they do not know its name?” |
| “Should we add a motivation button?” | “What makes someone hesitate or abandon a review session?” |
| “Is our menu intuitive?” | “Which words help someone predict what each category contains?” |
Use this frame:
We want to understand: [behaviour / situation]
For: [people concerned]
Because we must decide: [choice that can change]
We will know more when we observe: [signal]
We are not trying to prove: [non-goal]
2. Match the method to the uncertainty
The question chooses the method, not fashion.
| Uncertainty | Useful method | It can reveal | It cannot prove alone |
|---|---|---|---|
| Understand a practice or constraint | semi-structured interview, contextual observation | stories, strategies, vocabulary, context | frequency in the whole population |
| See whether a journey is doable | moderated or unmoderated usability test | actions, hesitation, errors, recovery | a final causal explanation |
| Compare groupings | open or closed card sort | categories and expected vocabulary | final visual quality |
| Check findability in a structure | tree test | findability in a text structure | visual and emotional experience |
| Measure a trend | product data, targeted survey | frequency, change, correlation | the cause of a behaviour |
| Check known rules | heuristic review, accessibility audit | foreseeable defects and priorities | every real user’s experience |
Methods can complement each other. Someone who says “it is clear” but searches for the button three times gives a useful signal: words illuminate experience, while behaviour makes the cost visible.
3. Recruit for situations
More participants do not compensate for the wrong profile. Define differences that can change the decision: level, usage frequency, device, language, assistive technology, goal, time constraint or domain experience.
A recruitment profile should contain:
- inclusion criteria: the situation the person has actually experienced;
- exclusion criteria: expertise that would distort the task, when relevant;
- wanted variation: mobile and desktop, beginner and regular user, low and high time pressure, for example;
- reason for selection: the difference must connect to the uncertainty.
Do not present a convenience sample as a general population. State the limit: “five mobile beginners recruited from our network” is more honest and useful than “users think…”.
4. Prepare a neutral protocol
A protocol reduces improvisation and protects the participant.
Possible introduction
We are testing the journey, not your skills. There is no right answer.
Describe what you would naturally do; we may ask what you are looking for,
but we will not guide you towards a solution. You can stop at any time. With
your permission, we will note actions and hesitation.
Before the session, specify consent, possible recording, data retention, anonymisation and withdrawal. Do not collect personal data simply because a tool makes it possible.
A well-written task
A task gives a goal and context, not the button to click:
You want to review ten concepts before an interview tomorrow morning.
Start a session, answer the first card, then explain how you would know what
to do after an interruption.
Avoid: “Click Start and tell me whether the page is clear.” That instruction reveals the path and measures obedience instead of findability.
Interview guide
Prepare themes, then let the person tell the story:
- “Tell me about the last time you did this.”
- “What were you trying to achieve then?”
- “What made you choose that action?”
- “How did you know you could continue?”
- “What would you have done if that option had not worked?”
Use “what happened next?” rather than “you meant that… right?”. Leading questions manufacture part of the answer.
5. Observe without over-interpreting
Record short, verifiable units:
| Type | Note example | Certainty |
|---|---|---|
| Verbatim | “I do not know whether it is finished.” | reported speech |
| Action | opens the menu, then goes back | observed behaviour |
| Time | hesitates eight seconds before the first click | situated measure |
| Error | selects a category that does not contain the course | observable result |
| Interpretation | does not understand progress | hypothesis to compare |
Keep separate columns for fact, interpretation, and next question. Do not turn an observed emotion or isolated sentence into a psychological diagnosis.
A light session grid is often enough:
Participant / context:
Task:
First choice:
Clue read or ignored:
Hesitation / error:
Request for help:
Result: succeeded, partial or abandoned:
Important phrase:
Hypothesis to check:
6. Synthesize by themes and evidence
Synthesis is not a vote for the prettiest quotes. Group observations by issue, context or mechanism. For each theme, write:
Theme: navigation vocabulary
Facts: 3 people searched for “course” under “Resources”; 2 used the breadcrumb
to return.
Interpretation: “Resources” may not match the intention “learn by topic”.
Confidence: medium; beginners, same task, small sample.
Decision: test an intention-based label and compare findability.
Open question: does the term work on mobile and in the other language?
Triangulate when possible: story, behaviour, product data, expert review or test. Convergence increases confidence; disagreement is learning, not a defect to hide.
Prioritising an issue
priority ≈ severity × observed frequency × reach × confidence
This is not mathematical truth. It is a tool for making discussion explicit. Document impact on the person, situations affected, evidence quality and cost of correction.
7. Limit common biases
- Confirmation bias: look only for validation; add a question that could disconfirm the solution.
- Recruitment bias: study only expert colleagues; include the situations that motivated the research.
- Desirability bias: people want to help; observe actions and remind them they are not being assessed.
- Novelty effect: a prototype attracts attention; test the task after a minimal explanation and in a realistic context.
- Survivorship bias: listen only to finishers; seek abandonment, failure and exit states.
- Solution bias: confuse an observed problem with a requested feature; restate the desired progress before designing.
8. Make a conclusion falsifiable
A research conclusion should say what could contradict it:
We think [situation] comes from [assumed mechanism], because [observations].
We will try [intervention] with [profile / context].
We will weaken the hypothesis if [contrary signal].
We will measure [metric or behaviour], without claiming beyond [study limit].
This turns research into a learning loop. Research does not always choose the solution; it makes the next choice less blind.
FlashLearning workshop: from signal to decision
- Choose a real friction in a flashcard journey.
- Write the decision that could change and the non-goal.
- Formulate a research question without a solution in the sentence.
- Prepare a neutral task, three probes and an observation grid.
- Simulate a session with someone who does not know your interface.
- Separate facts and interpretations, then group three observations.
- Write a hypothesis, a next test and the signal that could weaken it.
Checklist before sharing
- A decision can genuinely change.
- The question describes a situation or behaviour.
- The method matches the uncertainty.
- Profiles and limits are explicit.
- The task does not reveal the solution.
- Consent, privacy and withdrawal are covered.
- Facts, quotes and interpretations are separate.
- Every recommendation points to evidence.
- A limit and an open question remain visible.