INTERVIEWER Tell me about the last time you struggled with... PARTICIPANT Every Monday I'd open the app and it would just... NOTES key insight user-interviews · ask less · listen more · past behaviour · specific incidents

User Interviews

Ask less. Listen more. A user interview is a structured conversation designed to understand how someone thinks, what they do, and why — not to validate what your team already believes.

Product DiscoveryProblem ValidationPersona ResearchJourney MappingFeature ScopingOnboarding Evaluation

Two sentences.

A user interview is a structured one-on-one conversation in which a researcher asks open-ended questions to surface how a specific person thinks about, experiences, and navigates the domain your product operates in — with the goal of discovering things the team did not already know. The output is not a list of feature requests; it is observations about goals, behaviours, mental models, and frustrations that inform design decisions more reliably than any internal discussion.

User interviews became standard UX practice in the 1990s through work by Don Norman, Jakob Nielsen, and others who demonstrated that conversation with users produced better products than expert review alone. What makes interviews irreplaceable: no amount of analytics tells you what users were trying to do before they gave up, what alternative they considered, or what they tell colleagues when recommending or discouraging your product.

Apply this when…

You are entering a new problem space and need to understand context, constraints, and goals before any design
Quantitative data shows unexpected behaviour and the team needs to understand the reason
Personas are outdated or built from assumptions — interviews provide grounded raw material
A feature is being scoped and the team needs to understand how users currently solve the problem
A decision affects a user group the team has little direct contact with

When NOT to apply it

Skip it when you need statistical representativeness (five to eight interviews reveal patterns but not percentages), when the question is preference-based (use a survey), when you have no time to synthesise before the decision, or when participants are reluctant and would produce low-quality data.

The mechanism

User interviews create a structured conversation that draws out behaviour, context, and reasoning users would not surface unprompted. The researcher asks about what users actually do, how they do it, what has gone wrong, and how they have worked around it. The value is in the specifics that emerge from real experiences.

01
Behaviour and attitude diverge systematically
What people say they do and what they actually do are consistently different — due to social desirability bias, memory reconstruction, and the tendency to describe idealised behaviour. "Tell me about the last time X happened" asks for episodic memory of a real event, which is more accurate than "what do you do when X happens."
02
Open questions, past behaviour, specific incidents
The three question types that produce the most value: open questions ("tell me about how you currently manage X"), past behaviour questions ("walk me through the last time you did X"), and follow-up probes ("what did you do next?" "how did you feel?"). Questions inviting evaluation, hypotheticals, or leading are unreliable.
03
Users cannot design your product for you
Asking what features users want produces bad answers — not because users are unhelpful, but because they are not designers. Users can describe goals, frustrations, and workarounds with great accuracy. They cannot reliably predict what solution would best serve those goals. The interview surfaces the problem; the design team generates the solution.
04
Pattern frequency and emotional intensity are the signals
A frustration mentioned by six of eight participants warrants immediate attention. One mentioned once may not. After each session, note the three to five most significant observations — the surprises, unexpected workarounds, emotional reactions. Affinity mapping then reveals which cluster across participants.

Not a focus group, not a usability test, not a survey

A focus group captures group dynamics and social consensus. A usability test observes task performance with a specific interface. A survey measures aggregate sentiment. A user interview reveals individual context, mental models, and real-world behaviour. Each answers different questions; none substitutes for the others.

Intercom's jobs-to-be-done interviews and the messaging pivot

In Intercom's early years, the team conducted interviews structured around past behaviour — "walk me through the last time you used Intercom to communicate with a customer." What emerged was that customers were using the same product for three fundamentally different jobs: reactive support, proactive onboarding, and marketing to existing customers. Each had a different mental model, workflow, and measure of success.

This finding came entirely from interviews and could not have come from analytics. The structural decision to reposition Intercom around three distinct jobs — and eventually build separate product lines — was a direct consequence of the interview programme. That is the class of decision user interviews exist to inform.

Intercom · Jobs-to-be-Done Interviews
Three jobs in one product — structural insight from conversation
Reactive support —customer is stuckProactive onboarding —trigger at signupMarketing campaigns —re-engage inactive usersSame product — three different jobsInterview data surfaces three distinct mental modelsIntercom · jobs-to-be-done interviews · structural insight from conversation
Three jobs → three product lines

Test yourself & see real examples

No examples yet — be the first.

Spotted a product decision clearly made without talking to users — or one where the depth of understanding suggests someone did the interview work? Submit what you observed.

✓ Reviewed before publishing✓ Your name on every example you submit✓ Violation or fix — both welcome

Seen User Interviews skipped in a real product? Help grow the evidence base.

Where teams go wrong

Asking hypothetical and preference questions. "Would you use a feature that did X?" invites speculation. "When was the last time you needed X?" asks for real experience. Hypothetical predictions are consistently inaccurate; past behaviour is reliable data.
Leading with the product instead of the problem domain. Once participants see your product, every answer is filtered through it. Discovery interviews should begin with how users currently work, what they are trying to achieve, and what has gone wrong — before the product is ever mentioned.
Recruiting participants who are too similar to the team. Power users and vocal fans over-represent the most engaged segment. Users who gave up, churned, or never activated are the hardest to recruit and the most likely to contain the answers. A participant "like us" produces comfortable data; a genuinely different one produces useful data.
Not synthesising between sessions. Conducting all interviews before reviewing any findings misses the opportunity for early patterns to inform later questions. After each session, update the question framework. A theme from the first interview should be probed more deeply in the second.

Connected ideas

User interviews sit within a broader qualitative toolkit. Understanding the differences helps teams use interviews for the questions they are suited to answer.

The most important pairing is user interviews with affinity mapping. Interviews generate rich individual observations; affinity mapping reveals which represent shared patterns. Without affinity mapping, findings depend on the researcher's memory. Without interviews, affinity mapping has no data of sufficient quality. Together they form the complete generative research loop.

Run it right now

⏱ 10 minutes · Solo · No prep

The Question Audit

1. Write five questions you would typically ask in a user interview about your product. Write them exactly as you would ask them — do not idealise.

2. Classify each as Evaluative (E — asks to assess), Hypothetical (H — imagined future), or Behavioural (B — real past experience).

3. For every E or H question, write a Behavioural replacement: "Tell me about the last time you [did the thing]" or "Walk me through what happened when [the situation]."

4. Compare your originals to your replacements. If the replacements feel harder to answer — requiring real experience rather than opinion — that difficulty is the point. Research that works only on users with real domain experience produces real findings.

10 minutes