User Interviews

The full lesson

User interviews are the workhorse of generative UX research. When you need to understand the “why” behind behavior — the mental models, goals, frustrations, and contextual pressures that shape how people relate to a problem — a well-structured conversation is unbeatable. But a poorly prepared or poorly analyzed interview is worse than no data. It produces false confidence. This lesson covers the full arc from study design to synthesis, with emphasis on the habits that separate rigorous practice from research theater.

When to Use Interviews (and When Not To)

Interviews are a generative, attitudinal method. “Generative” means they help you discover and explore. “Attitudinal” means they surface what people think, believe, remember, and feel. They do not directly observe what people actually do — and that distinction matters.

The say/do gap is well documented: people misreport their own behavior. They round up how often they do things, invent reasons after the fact, and give answers that sound good socially.

Use interviews when:

• You are in discovery mode and need to understand a problem space before you have a solution to test.
• You want to understand motivations, decision criteria, and mental models that analytics cannot show.
• You need to map a workflow or experience as users perceive it, including steps that happen outside any product you control.
• You are exploring a new domain or user segment where you lack baseline knowledge.

Do not rely on interviews when:

• You want to know what users will actually do with an interface — run a usability test instead.
• You need statistically reliable data — use a survey with a validated instrument like SUS, UMUX-Lite, or SEQ, sized for 40 or more respondents at 95% confidence.
• Your question is “which of these two designs performs better” — that calls for a comparative usability test or an A/B experiment.

Study Design: Starting With the Right Question

A common failure is running interviews without a clearly scoped research question. “Let’s talk to some users” is not a research question. A good interview study starts with a question that is:

• Specific enough to be answerable. “What mental model do first-time home buyers use to understand mortgage options?” is answerable. “What do users think of our product?” is not.
• Generative, not evaluative. If you already have a solution to test, an interview is the wrong primary method.
• Focused on behavior, context, and motivation — not on preferences and opinions.

From that research question, build a discussion guide. A discussion guide is not a script. It is a structured set of topic areas and opening questions. The guide ensures you cover what you need to cover, while still leaving room to follow where the participant takes you.

Sample Sizing for Qualitative Research

The modern standard for qualitative problem-finding research is 5 participants per distinct user segment. This number comes from Nielsen and Landauer’s 1993 analysis, which showed that 5 users surface roughly 85% of the most common usability problems. The key qualifier: this applies to qualitative, problem-finding work with a relatively homogeneous user group.

A common mistake is applying this number everywhere. Quantitative benchmarking studies need 40 or more participants for reliable statistics at 95% confidence. Know which mode you are in. If your research question requires detecting differences or establishing percentages, five interviews will not answer it.

If you have multiple distinct segments — say, both enterprise admins and end users — recruit 5 per segment, not 5 total.

Recruiting the Right Participants

Bad recruitment is the most common cause of useless interview data. Convenience samples — colleagues, friends, anyone who responds quickly — produce findings that reflect those people, not your actual users.

Best practices for participant recruitment in 2026:

• Define screening criteria before writing the screener. Who are the people whose experiences will answer your research question? Define them by behavior (“bought a home in the past 18 months”), context (“uses the product for work, not personal use”), or expertise (“has filed taxes independently at least twice”).
• Use a screener survey with open-ended behavioral questions, not self-reported skill levels. “Describe the last time you had to dispute a charge on a credit card” reveals far more than “How often do you use online banking (1–5)?”
• Avoid recruiting only from existing customers. If you need to understand why people churn or never convert, you need participants who left or never arrived.
• Recruit to your repository. Modern research operations maintain a participant panel — a research repository — that captures opt-ins, past study participation, and segment data. Tools like Dovetail, Maze, or a well-maintained CRM integration can cut recruitment lead time without sacrificing quality.

Writing a Discussion Guide That Works

The discussion guide structures the session without scripting it. A rigid script produces predictable answers. A guide lets the conversation follow the participant’s reality.

Structure

A one-hour interview typically follows this arc:

1. Introduction (5 min). Consent, recording permission, a brief framing of the session. Make clear you are studying their experience, not testing them — there are no right or wrong answers.
2. Warm-up (5–10 min). Broad, easy questions to build rapport and set context. “Tell me a bit about your role and what a typical work week looks like.”
3. Core topics (35–40 min). Three to five topic areas, each opened with a broad behavioral prompt, then followed by probing questions.
4. Wrap-up (5–10 min). “Is there anything important about your experience with X that I haven’t asked about?” This is a high-yield question — participants often surface their most pressing concern here.

Question Quality

Weak question type	Why it fails	Stronger alternative
Leading: “Do you find it frustrating when…”	Suggests the answer	”Walk me through what happened when you last tried to do X.”
Hypothetical: “What would you do if…”	Self-report on imagined behavior	”Tell me about a time when you had to deal with Y.”
Opinion/preference: “Which do you prefer?”	Attitudinal, low predictive value	”Describe the last time you chose between A and B — what drove that decision?”
Multi-part: “How do you do X, and do you find Y useful?”	Forces the participant to choose what to answer	Ask one question at a time
Closed: “Was that easy?”	Yes/no answer ends the thread	”What was that like for you?”

The most powerful interview technique is the simplest: the follow-up probe. “Tell me more about that.” “What did you do next?” “Why did that matter to you?” Silence is also a probe. Most interviewers fill silence too quickly and cut off the participant’s reflective thinking.

Running the Session: Moderation Craft

The quality of your data depends heavily on how you moderate. Several techniques separate competent from excellent interviewers.

Neutrality without coldness. Your job is to understand the participant’s experience, not to validate or challenge it. “That’s really interesting — can you say more?” is fine. “That’s a great point!” signals approval and shapes subsequent answers. Be warm and genuinely curious while staying neutral on content.

Laddering. When a participant gives a surface-level answer, ladder down. Keep asking “why” — or softer variants like “what made that important to you?” — until you reach the underlying motivation or value. Surface answers produce surface insights.

Paraphrasing to check understanding. “So if I’m hearing you right, the biggest friction point is not knowing whether your change was saved — is that right?” This surfaces misunderstandings during the session, where you can correct them, rather than during analysis, where you cannot.

Managing dominant participants. Some participants answer every question with a confident monologue that drifts from lived experience into opinions and abstractions. Gently redirect: “That’s helpful context — I want to make sure I understand the specific situation you mentioned earlier. Can we go back to that?”

Analysis and Synthesis

Raw interview data — even excellent data — is not insight. Synthesis is where research creates value. It is also where research most commonly goes wrong.

Affinity Diagramming

Affinity diagramming is the most widely used qualitative synthesis method. Write each distinct observation on a card (digital or physical), then cluster cards by emergent theme — not by which question was asked. The key discipline: cluster by what participants said or did, not by your pre-existing mental model of the problem. Forcing observations into a framework you brought in contaminates the synthesis.

Modern tools like Dovetail, Maze Analyze, or EnjoyHQ offer AI-assisted tagging and clustering. These tools speed up the mechanical work of coding but do not replace the judgment of deciding which clusters represent truly distinct themes versus surface-level topic groupings.

Avoiding Common Synthesis Failures

• Frequency as a proxy for importance. A theme mentioned by 4 of 5 participants is significant. But a theme mentioned by just 1 participant who represents a highly underserved segment may be more strategically important than a commonly voiced minor irritation. Count mentions, but do not let counting replace judgment.
• Confirmation bias in theme selection. It is easy to notice observations that confirm existing beliefs and downplay contradictions. Involve a second analyst in independent coding and compare results before merging clusters.
• Jumping to solutions in synthesis. Affinity clusters describe the problem space. They are not design requirements. The move from insight to implication to design direction is a separate, deliberate step.

Deliverables: From Findings to Decisions

Research that does not change decisions is wasted effort. Shape your deliverable around the decision it needs to inform, not around convention.

Common formats and their appropriate uses:

• Research readout (slide deck or document). Best for sharing findings broadly with stakeholders who were not in the sessions. Lead with insights and implications, not methodology. Include direct quotes and representative clips.
• “How Might We” statements. Translate insights into opportunity framing. Useful as input to ideation workshops. Each HMW should map to a specific research insight so it stays grounded.
• Journey map or experience map. Synthesizes interviews with behavioral data to show the end-to-end user experience, including emotional highs and lows. Most valuable when interviews surfaced significant context that occurs outside the product.
• Jobs-to-be-Done summary. If your interviews reveal that users are hiring your product (or a competing solution) to accomplish a higher-order goal, a JTBD framing helps product and strategy teams prioritize.

The modern standard for research operations is storing synthesized findings in a shared repository — Dovetail, Notion, or equivalent — with tagged observations linked to source transcripts. This keeps findings retrievable and auditable months later, rather than locked in a one-time presentation that everyone has forgotten.

Interviews Within a Mixed-Method Program

A single qualitative method is a starting point, not a complete picture. Modern research practice treats user interviews as one node in a triangulated mixed-method program:

• Interviews reveal the “why” and the mental model.
• Analytics and session recordings reveal the “what” at scale.
• Usability testing reveals friction in the designed experience.
• Surveys (SUS, UMUX-Lite, SEQ) quantify satisfaction and effort with statistical confidence.

When behavioral data and interview findings diverge — users say something is easy but analytics show high drop-off — that divergence is itself a high-value research signal worth investigating. It is not a problem to explain away in favor of one data type.