User Interviews
Date: 2026-08-17
A structured conversation to understand someone’s context, goals and past behaviour. The discipline is entirely in the questions — people will confidently tell you what they’d do in future, and that answer is close to worthless.
A user interview is a semi-structured conversation with an individual, aimed at understanding what they do, why, and what constrains them.
Semi-structured is the operative word: a prepared guide of topics, asked in a flexible order, with follow-ups driven by what the person says. Fully scripted becomes a spoken survey; fully unstructured isn’t comparable between participants.
Past behaviour, not future intention
The single rule that separates a useful interview from a misleading one.
WORTHLESS USEFUL
"Would you use this?" "Tell me about the last
"Would you pay £30?" time you bought one"
"How often would you..." "What did you do next?"
"Do you like this?" "What were you trying
to achieve?"
People are poor predictors of their own behaviour and excellent reporters of it. They want to be helpful, they underestimate friction, and they answer hypotheticals with an idealised self.
The fix is to anchor every question to a specific real episode. “The last time” is the most valuable phrase in the guide, because it forces recall rather than invention.
Question shapes to avoid
LEADING
"How frustrating was the checkout?"
→ assumes the answer
DOUBLE-BARRELLED
"Was it fast and easy to find?"
→ which half are they answering?
BINARY TOO EARLY
"Did you like it?"
→ yes/no, and you learn nothing
JARGON
"How do you find the faceted nav?"
→ they guess what you mean
WHY, repeatedly
→ people rationalise. Ask what they
DID, then what happened next
“Why” is weaker than it looks. Asked directly, it invites a post-hoc justification the person constructs on the spot. “Walk me through what you did” produces the same understanding without the invention.
Running one
1 WARM UP easy, factual questions
→ build rapport, calibrate
2 CONTEXT their role, situation,
constraints
3 THE EPISODE one specific recent
instance, in detail
4 PROBE "what happened then?"
"what did you expect?"
5 WRAP anything I should have
asked?
Silence is a technique. After an answer, wait. People fill pauses with the qualification they were editing out, and it’s frequently the most useful sentence in the session.
Record, with consent, and take light notes. Trying to transcribe live means missing the follow-up worth asking — Research Ethics.
What interviews are bad at
- Anything about a specific interface. Watching someone use it tells you far more than asking them about it — Usability Testing
- Estimating frequency or prevalence — Sample Size in Qualitative Research
- Preference between designs. Stated preference and revealed behaviour diverge routinely
- Pricing. People will name a number and not pay it — Willingness to Pay
Where it goes wrong
- Interviewing the wrong people. Convenient participants — colleagues, engaged power users, whoever answered the recruitment email fastest — are systematically unrepresentative
- Confirming a decision already made. If the interview exists to justify a roadmap, the questions will find the answer
- Only speaking to current customers. They are, by definition, the people the product already works for. The people who left, or never converted, hold the information you don’t have — and they’re harder to recruit, which is why it doesn’t happen
- One person’s summary standing in for the evidence. Findings should be traceable to what someone actually said — Qualitative Coding, Research Repositories
The number of them
Five to eight per distinct audience segment is the usual working range, stopping when new interviews stop producing new information. That’s saturation, and it’s the honest stopping rule rather than a fixed count — Sample Size in Qualitative Research.