UX researcher role-specific interview questions

As asked

A PM says activation is low and asks you to run user interviews. How do you decide whether interviews, usability tests, surveys, or product analytics are the right method?

Sample answer outline

Clarify the decision the research must inform before choosing a method. If the team does not know why users fail, start with analytics to locate drop-off and usability tests to observe the flow. Interviews help understand motivation and context, but they are weak evidence for measuring prevalence. Surveys can quantify patterns after qualitative work has generated good hypotheses. A strong researcher pushes back on 'run interviews' as a pre-selected method and builds a mixed plan that matches the risk, timeline, and confidence needed.

Expect these follow-ups

• What would you do if the PM needs an answer in three days?
• How do you avoid turning a usability test into a feature-request session?
• What decision would make a survey the wrong tool here?

As asked

A product manager comes to you with a question: why are users dropping off on step three of our onboarding flow? How do you decide which research method to use, and what would your first move be?

Sample answer outline

Strong answers frame the question as a 'why' problem that qualitative methods are better suited for, then mention usability sessions or contextual inquiry to observe the drop-off. They distinguish between validating a hypothesis versus generating one, and note when to layer in funnel analytics to size the problem before recruiting. Bonus if they mention combining diary studies for longitudinal context.

Expect these follow-ups

• How many participants would you recruit for usability sessions on this, and why that number?
• What if the PM says there is no time for research and wants a quick answer?

As asked

How do you determine the right number of participants for a usability study, and how do you defend that number to a stakeholder who thinks five people is not enough?

Sample answer outline

A strong answer cites Nielsen's 85% threshold finding and explains diminishing returns per additional participant in formative studies. They distinguish between formative and summative contexts, noting that summative studies need statistical power. They explain saturation, not head-count, as the stopping criterion for qualitative work, and reframe the stakeholder concern around insight quality vs. representativeness.

Expect these follow-ups

• When would five participants genuinely not be enough, even for qualitative work?
• How do you handle when saturation happens at three participants?

As asked

Walk me through how you actually facilitate a think-aloud usability test. What do you say in the intro, how do you prompt without leading, and what do you do when a participant goes silent?

Sample answer outline

Good answers cover the consent and framing script, the distinction between concurrent and retrospective think-aloud, and neutral prompts like 'what are you thinking right now?' rather than 'does that make sense?'. They mention handling silence with silence before prompting, and address how to respond when participants ask for help mid-task without contaminating data.

Expect these follow-ups

• What is your protocol when a participant says they would never actually use this product?
• How does your facilitation change for remote sessions versus in-person?

As asked

A designer hands you a draft survey they wrote and one of the questions reads 'How much did you enjoy using our new feature?' What issues do you flag, and how do you rewrite it?

Sample answer outline

The answer names leading/loaded framing, acquiescence bias risk, and the assumption embedded in 'enjoy'. A strong rewrite could be a neutral satisfaction item on a Likert scale or a separate open field. They also mention ordering effects if this question appears before a general satisfaction item, and the importance of pretesting survey language with a small pilot group.

Expect these follow-ups

• How would you test whether your rewritten question actually measures what you intended?
• What is your go-to approach for validating a survey instrument before fielding it broadly?

As asked

You need to recruit 12 participants for a study on a B2B expense-management tool. How do you write a screener that gets you the right people without tipping them off to the 'right' answers?

Sample answer outline

Strong answers describe writing screener questions that are behaviorally observable and not self-reported attitude items, using decoy questions to avoid gaming, and mixing the qualifying criteria among disqualifying options. They define their segment criteria precisely (role, frequency of use, company size) before writing, and note the importance of over-recruiting by 20 to 30 percent to handle no-shows. They also mention IRB or consent considerations for sensitive populations.

Expect these follow-ups

• How do you handle participants who made it past the screener but turn out to be wrong for the study once they arrive?
• What is your approach when your panel or recruiting vendor cannot find people who meet your criteria?

As asked

Tell me about the remote research tools you use regularly. For each one, tell me what you use it for and one meaningful limitation you have run into.

Sample answer outline

Strong answers mention at least three distinct tools by name covering different methods: for example Lookback or UserZoom for moderated sessions, Maze or UserTesting for unmoderated prototype testing, Optimal Workshop for IA, Typeform or Qualtrics for surveys. They name real limitations like Maze not supporting complex Figma interactions, or Qualtrics being overkill for small teams. Weak answers stay generic ('I use Zoom and Google Forms').

Expect these follow-ups

• If you had to cut your tooling budget in half, which tools would you keep and why?
• Have you ever had a tool fail mid-session? What was your contingency?

As asked

Before you start a study, you write a research brief. What does yours include, and why is each section there?

Sample answer outline

A thorough answer covers: background and business context, research objectives, specific questions the study will and will not answer, methods and rationale, participant criteria, timeline and stakeholders, and how findings will be used. They emphasize that listing what is out of scope prevents scope creep and that the 'how findings will be used' section forces stakeholders to commit to actioning research before it starts.

Expect these follow-ups

• How do you handle it when stakeholders want to add more objectives to the brief after it is approved?
• Walk me through the difference between a research question and a survey question.

As asked

Your team wants to run usability tests but is short on researcher time. A designer suggests just using an unmoderated tool like UserTesting or Maze. How do you evaluate that suggestion?

Sample answer outline

Strong answers name concrete tradeoffs: unmoderated wins on speed and scale but loses the ability to probe unexpected behavior, recruit to specific criteria, or catch confusion that participants work past silently. They recommend unmoderated when the prototype is high-fidelity and tasks are well-defined, and moderated when the research question involves nuanced reasoning or early-stage concepts. They also mention completion-rate noise in unmoderated tools.

Expect these follow-ups

• Have you ever had an unmoderated study give you misleading results? What happened?
• How would you combine both within a single research sprint?

As asked

You are interviewing a user who says 'I just want everything to be simpler.' How do you go deeper to understand what that actually means for them, without putting words in their mouth?

Sample answer outline

Good answers describe laddering or the five-whys technique: asking 'what would simpler look like for you?' then 'and why does that matter?' iteratively. They distinguish between probing for examples versus abstracting to values too early. They mention noting non-verbal cues and pausing before redirecting, and recognize when a participant is giving a socially desirable answer versus a real behavioral one.

Expect these follow-ups

• How do you know when you have gone deep enough and are starting to create artifacts rather than uncover real needs?
• What do you do when two participants define 'simple' completely differently?

As asked

You have just finished eight contextual interviews and have a wall of sticky notes. Walk me through how you run an affinity mapping session, including how you involve design and product without letting them override the data.

Sample answer outline

Strong answers describe silently reading and de-duplicating raw observations, grouping bottom-up by similarity rather than top-down by existing product categories, and labeling groups with observation statements not interpretations. They address facilitating cross-functional sessions by setting norms around 'move it, don't debate it' and using dot voting only after grouping is complete. They note how to surface disagreements as research questions.

Expect these follow-ups

• How do you handle it when the PM wants to name clusters after their roadmap items?
• How do you decide when a cluster is big enough to be meaningful versus just noisy?

As asked

How does a Jobs-to-Be-Done framing change the questions you ask in a user interview compared to a persona-based approach? Give me a concrete example from work you have done.

Sample answer outline

A strong answer explains that JTBD interviews focus on the purchase or adoption event, not demographics, by asking about the last time someone took a specific action and what prompted it. They contrast this with persona research that asks who the user is rather than what they are trying to accomplish. A concrete example shows how the same user population splits into different jobs that require different solutions.

Expect these follow-ups

• How do you identify the 'functional' versus 'emotional' versus 'social' dimensions of a job?
• When does JTBD break down as a framework and you need something else?

As asked

You are redesigning the navigation of a content-heavy B2B platform. When would you use open card sort versus closed, and how do you go from sort results to actual IA recommendations?

Sample answer outline

Strong answers distinguish open sorts for discovery of mental models from closed sorts for validating a proposed structure. They describe analyzing dendrograms and similarity matrices from tools like Optimal Workshop or Maze, identifying consensus clusters and outlier items. They note that card sort data shapes groupings but tree testing is needed to validate navigation paths. They also mention the difference between participants' labels and the vocabulary appropriate for business or legal constraints.

Expect these follow-ups

• How do you handle items that participants consistently place in multiple categories?
• What do you do when card sort results contradict the existing IA that the engineering team has already built?

As asked

After a card sort you have a draft IA. Walk me through how you set up and analyze a tree test, including how you decide which tasks to write and what success looks like in the results.

Sample answer outline

Good answers describe writing tasks around user goals, not menu labels, to avoid cueing. They explain directness scores, success rates, and where-to-first-click data. They know that high success with low directness (backtracking) is a warning sign. They interpret results by task, not just aggregate, and identify specific nodes where users bail out. They also mention recruiting 50 to 100 participants to get stable task-level scores.

Expect these follow-ups

• What do you do when two tasks have the same correct destination but very different success rates?
• How do you decide when to iterate on the IA versus when the tree test results are good enough to move to prototype?

As asked

You ran a usability study and the SUS score came back at 67. How do you interpret that, and what do you tell the product team?

Sample answer outline

A strong answer explains the SUS scoring formula (alternating scoring, multiply by 2.5), places 67 in the C minus to C range on Bangor et al.'s adjective and letter-grade scale, and notes that the commonly cited adjective for this band is 'OK' rather than poor. They caution against over-interpreting a single score without knowing the sample size and confidence interval, and they pair the score with qualitative findings from the same sessions to prioritize what to fix. They also explain SUS as a comparative tool, not an absolute threshold.

Expect these follow-ups

• How many participants do you need to get a reliable SUS score?
• When would you use SUS versus NPS versus a custom satisfaction scale?

As asked

Leadership wants to use NPS as the primary UX quality metric. What are the strengths and limitations of NPS for measuring the user experience, and what would you propose alongside it?

Sample answer outline

Strong answers note that NPS measures likelihood to recommend, not usability or task success, and is sensitive to factors outside the product like price and support. They mention the binary passive-zone issue and cultural response bias in international products. They propose pairing NPS with task success rates, CSAT, CES, or UMUX-Lite to triangulate, and suggest pairing with open-text to diagnose drivers. They may also note that NPS requires large samples for reliable segmentation.

Expect these follow-ups

• How would you design a lightweight measurement program that gives you reliable UX signal without burdening users?
• Have you ever pushed back on a metric leadership loved? How did that go?

As asked

You are designing a survey to measure attitudes toward three redesign concepts. Walk me through how you decide between a Likert scale, a semantic differential, and a forced ranking, and what pitfalls each has.

Sample answer outline

Good answers describe Likert for agreement intensity, semantic differential for bipolar attitude measurement along adjective pairs, and forced ranking for relative preference when acquiescence bias is a concern. They note that 5-point versus 7-point scales affect discriminability. They flag that forced ranking does not tell you absolute preference and can frustrate respondents when all options feel similar. They address max-diff as a more efficient alternative for comparing many concepts.

Expect these follow-ups

• When would you use a max-diff study instead of ranking questions?
• How do you handle neutral midpoints on Likert scales when your audience is cross-cultural?

As asked

You want to know whether users can find the 'export to CSV' function on a new dashboard. Why would you use a first-click test rather than a full usability session, and what does a 70% first-click success rate actually tell you?

Sample answer outline

Good answers explain that first-click tests are faster and cheaper for isolated navigation decisions, and reference research showing first-click accuracy is a strong predictor of task completion. A 70% rate is marginal; the answer should address what the remaining 30% clicked instead and whether those clicks indicate a competing mental model. They would then look at click maps to identify the competing target and potentially redesign or relabel.

Expect these follow-ups

• How many participants do you need for a reliable first-click test result?
• What would change in your recommendation if the 70% correct first-click group completed the full task but the 30% incorrect first-click group also eventually succeeded?

As asked

Engineering is about to A/B test two checkout flow designs. What do you need to review before the test launches to make sure the results will actually be interpretable?

Sample answer outline

Strong answers cover: a pre-set primary metric and success criterion, a power calculation to determine minimum detectable effect and sample size, randomization strategy and segment consistency (no novelty effect contamination), a pre-defined test duration based on business cycle, guardrail metrics to catch harm, and agreement on what 'no winner' means for the roadmap. They also ask whether the variants are actually different enough to expect a measurable difference.

Expect these follow-ups

• What do you do when you get a statistically significant result but the effect size is trivially small?
• How do you handle an A/B test where the conversion metric improved but the qualitative feedback was overwhelmingly negative?

As asked

Your A/B test shows a statistically significant improvement in task completion rate: from 72% to 73.4%. How do you report this finding to the team?

Sample answer outline

A strong answer explains that the result is statistically real (not noise) but practically marginal, and introduces effect size as the right lens. They estimate the business impact of that 1.4 percentage point lift given traffic volume, compare it to the cost of shipping the change, and recommend checking whether the confidence interval's upper bound is practically meaningful. They also recommend checking whether the improvement is uniform or driven by one segment.

Expect these follow-ups

• What is the minimum effect size you would argue is worth shipping, and how would you calculate it?
• How do you communicate this nuance to a PM who just wants to know if the test won?

As asked

A product manager asks you what p less than 0.05 actually means in plain language. How do you explain it, and what common misinterpretation do you need to head off?

Sample answer outline

The correct explanation: a p-value of 0.04 means that if there were truly no difference, you would see a result this extreme or more extreme only 4% of the time by chance alone. The misinterpretation to head off is 'there is a 96% chance this result is real' or 'the null is probably false.' They also flag that p less than 0.05 is a convention, not a truth threshold, and mention the risk of p-hacking from multiple comparisons without correction.

Expect these follow-ups

• What is a confidence interval and how does it give you more information than a p-value alone?
• Your company runs 50 A/B tests a month and adjusts significance thresholds. How do you think about false discovery rate?

As asked

What is the difference between a persona and a behavioral archetype, and when would you push back on a team that wants traditional persona documents?

Sample answer outline

Strong answers distinguish persona (a data-informed fictional composite of demographic and behavioral traits) from archetype (a pattern of motivations and behaviors without demographic fiction). They critique personas that are marketing-demographic rather than behaviorally grounded and explain why archetypes are more durable across product cycles. They would push back on persona requests when the team is early-stage and would be better served by JTBD or when the proposed personas would lock in assumptions before data collection.

Expect these follow-ups

• How do you prevent personas from being created and then ignored?
• Describe how you would validate that a persona actually represents a real segment in your data.

As asked

Your company has three years of research in scattered Confluence pages and Drive folders. How would you build a research repository that teams actually use, and what are the failure modes you are trying to avoid?

Sample answer outline

Good answers name tools like Dovetail, Notion, or Aurelius and describe tagging taxonomy design as a critical upfront decision. They distinguish between a report library and a nugget-level insight repository, and argue for nuggets because they surface faster and age better. Failure modes include over-tagging that no one maintains, storing reports without executive summaries, and no clear ownership for ongoing curation. They mention evangelizing the repo with 'research previously showed...' citations in meetings.

Expect these follow-ups

• How do you decide what level of granularity to store, full sessions versus insights versus themes?
• How do you handle participant data privacy in a shared repository?

As asked

You are the only researcher at a 150-person product company with five product teams. How do you decide what research to do yourself versus democratize to designers or PMs, and what guardrails do you put in place?

Sample answer outline

Strong answers describe a hub-and-spoke or embedded hybrid model where the researcher sets standards and handles high-stakes studies while training practitioners for lightweight discovery. Guardrails include standardized participant consent flows, a screener review process, a 'research request' intake form to prioritize, and a shared playbook with templates. They flag risks of non-researchers leading sessions without training causing anchoring or leading questions.

Expect these follow-ups

• How do you handle a PM who ran their own research incorrectly and is now making decisions based on it?
• What is your process for reviewing another team member's research plan before they go into the field?

As asked

Your survey shows that 60% of users are satisfied with search, but your usability sessions show repeated frustration during search tasks. How do you reconcile this and what do you tell the team?

Sample answer outline

Good answers explain the difference between self-reported satisfaction (retrospective, relative to expectations) and observed behavior (real-time, task-based), and that both can be simultaneously true. They would check whether the usability participants match the survey segment, explore whether users have adapted workarounds that maintain satisfaction despite difficulty, and run follow-up survey items probing specifically search sub-tasks. The recommendation: the usability friction represents headroom for improvement even among satisfied users.

Expect these follow-ups

• What additional data point would most efficiently help you resolve the contradiction?
• How do you prevent a PM from cherry-picking one of the two findings that supports their existing plan?