UX researcher domain knowledge interview questions

As asked

What is the difference between a descriptive research question, an evaluative research question, and a generative research question? Give an example of each.

Sample answer outline

Descriptive: 'Who are our current users and what tasks do they do most often?' Evaluative: 'Can users complete the checkout in under two minutes without errors?' Generative: 'What are the unmet needs in how people manage monthly expenses?' Strong answers connect each type to the appropriate method and explain that conflating them leads to poorly designed studies that answer the wrong question.

Expect these follow-ups

• Which type of question does your current role generate most often and why?
• What happens when a stakeholder asks a descriptive question but the team actually needs a generative answer?

As asked

Define internal validity, external validity, and reliability, and give a concrete example of how each can be compromised in a usability study.

Sample answer outline

Internal validity threat: demand characteristics in a moderated session where participants behave differently because they know they are being watched. External validity threat: testing with a convenience sample of tech-savvy employees when the product targets seniors. Reliability threat: different moderators using different prompts across sessions, making results inconsistent. A strong answer shows the researcher has internalized these as real concerns, not textbook definitions.

Expect these follow-ups

• Which of these three is hardest to achieve in a startup context and why?
• How does ecological validity differ from external validity in the context of a diary study?

As asked

What is confirmation bias and what are three specific ways it can enter a UX research study at different stages?

Sample answer outline

Three entry points: (1) Screener design that over-recruits enthusiastic users and excludes critics. (2) Facilitator probing that follows up on positive feedback more than negative. (3) Analysis that codes data to fit a pre-formed theme rather than letting themes emerge. A strong answer also mentions stakeholder review sessions where leadership selectively remember the findings that support their plans, and how to surface this by recording and sharing raw insights before interpretation.

Expect these follow-ups

• How do you run a synthesis session in a way that actively fights confirmation bias?
• Have you ever caught yourself falling into confirmation bias? What did you do?

As asked

Describe the Double Diamond model and explain what type of research belongs in each diamond. Where do most product teams skip a step and what goes wrong when they do?

Sample answer outline

First diamond: Discover (generative, open research) and Define (synthesis into a focused problem statement). Second diamond: Develop (ideation and concept testing) and Deliver (usability testing and iteration). Teams most often skip the first diamond entirely and jump to solutions without defining the right problem, resulting in well-executed products that solve the wrong need. They also skip the 'Define' phase by treating raw data as conclusions.

Expect these follow-ups

• Where does your current team skip, and how do you work around it?
• How do you adapt the Double Diamond to a fast-moving agile sprint cycle?

As asked

Walk me through the six phases of Braun and Clarke's thematic analysis and explain where most researchers cut corners.

Sample answer outline

The six phases: familiarize with data, generate initial codes, search for themes, review themes, define and name themes, and write up. Most researchers cut the familiarization phase (reading transcripts without coding) and jump straight to highlighting, which means they miss latent patterns. They also conflate themes with topics (frequency-based labels) rather than constructing a meaningful claim about the data. Strong answers show the researcher has actually done this, not just read about it.

Expect these follow-ups

• What is the difference between a semantic theme and a latent theme?
• How do you handle disagreement between two researchers coding the same transcripts?

As asked

What is a mental model in the context of UX research, and how do you surface and document a user's mental model of a complex process?

Sample answer outline

A mental model is the user's internal representation of how a system works, which may differ from how it actually works. Methods to surface it: card sort for structural models, interview probing for process steps using the draw-a-picture or explain-it-to-me-like-I-am-new technique, and protocol analysis during task performance. Documentation uses mental model diagrams (Indi Young style) with behavioral patterns below the line and feature solutions above. Strong answers note that mental models are partial and context-dependent.

Expect these follow-ups

• How do you use a mental model diagram in a stakeholder presentation?
• When a user's mental model is just wrong, is it your job to correct it or the design's job to match it?

As asked

What is the difference between a user journey map built from assumptions and one built from research, and what data sources do you use to populate each stage?

Sample answer outline

An assumption-based map is a stakeholder alignment tool; a research-backed map reflects actual behavior, emotions, and pain points sourced from interviews, diary studies, or log data. Each stage should be grounded in participant quotes or observed behavior, not internal speculation. Data sources include interview transcripts for emotions and motivations, analytics for time-on-step and dropout, and support tickets for recurring pain points. The map should show variance between segments, not just a single happy path.

Expect these follow-ups

• How do you validate a journey map after you have built it?
• When is a journey map the wrong deliverable and you should use something else?

As asked

What is grounded theory and how does it influence how you approach coding interview transcripts?

Sample answer outline

Grounded theory is an inductive approach where theory emerges from data rather than being imposed on it, using open, axial, and selective coding stages. In practice this means starting transcript analysis without a coding frame, generating codes from participant language, grouping them into categories, and iterating until a core category emerges. The alternative is deductive coding which is faster but risks missing new themes. A strong answer notes that pure grounded theory is rare in industry; most researchers use a pragmatic hybrid.

Expect these follow-ups

• When would you choose inductive coding over a pre-set codebook?
• What software do you use for coding transcripts and why?

As asked

You want to detect a difference of 5 percentage points in task completion rate between two design variants with 80% power and a significance level of 0.05. How do you estimate the sample size you need?

Sample answer outline

A strong answer describes using a two-proportion z-test power formula or an online power calculator, inputs being the baseline rate, minimum detectable effect, alpha (0.05), and desired power (0.80). For a 70% baseline and a 5 percentage point lift, the two-proportion z-test produces roughly 1,200 to 1,300 participants per variant, not the 800 to 1,000 figure sometimes cited in informal rules of thumb. They note that smaller effects require many more participants, that one-sided versus two-sided tests affect the number, and that power calculations should be done before data collection, not after.

Expect these follow-ups

• What happens to your sample size requirement if you want 90% power instead of 80%?
• How do you justify the chosen minimum detectable effect to a PM who wants to detect even smaller differences?

As asked

How do Kahneman's System 1 and System 2 thinking frameworks apply to how you design tasks in a usability study?

Sample answer outline

System 1 is fast, automatic, and emotion-driven; System 2 is deliberate and analytical. Usability studies often activate System 2 more than real use does because participants are aware they are being observed and being asked to think aloud. This means you may overestimate users' ability to notice errors and under-estimate how many errors happen through habit or inattention. Design implications: include tasks that are fast, habitual, and dual-task to approximate real-world System 1 use.

Expect these follow-ups

• What design feature would a System 1 user rely on that a think-aloud participant would never notice?
• How does this framework affect how you interpret task completion data?

As asked

What is Microsoft's Product Reaction Cards or Desirability Toolkit and when would you use it in a study?

Sample answer outline

The toolkit consists of 118 product reaction cards with adjectives (both positive and negative) that participants select after interacting with a product. It provides a structured vocabulary for emotional and aesthetic response without relying on open-ended verbal fluency. Best used for comparing emotional profile of two design concepts or tracking brand perception over time. Limitations include cultural bias in the adjective set and the tendency to create a list that does not capture nuance without a follow-up debrief.

Expect these follow-ups

• How do you analyze and report product reaction card data when comparing two concepts?
• What adjective patterns signal a high-usability versus a high-delight result, and what is the difference?

As asked

When do you use a retrospective think-aloud debrief instead of concurrent think-aloud, and what are the tradeoffs in data quality?

Sample answer outline

Concurrent think-aloud is done while performing the task; retrospective is done after, often while watching a replay. Concurrent produces richer moment-to-moment narration but adds cognitive load that can inflate task times and alter behavior. Retrospective reduces in-task interference but suffers from memory reconstruction and rationalization. Retrospective with screen replay reduces recall error. Strong answers note that Ericsson and Simon's protocol analysis research shows concurrent is more valid for capturing actual cognitive processes.

Expect these follow-ups

• How do you choose between the two when you are testing a complex multi-step flow that takes 20 minutes?
• What is the risk of using think-aloud with novice users versus experienced users?