Conduct user interview

Conduct User Interview

Execute a live interview session so the participant does 90% of the talking and every answer traces to real, recalled experience — not polite speculation.

When to use / when NOT to use

Use when:

• Running a discovery/attitudinal interview session (remote or in person) — from opening consent to closing
• Prepping to facilitate: probe repertoire, silence handling, observer briefing
• Recovering mid-session: terse participant, rambling participant, observer interrupting, you caught yourself leading

NOT for — go to the sibling skill:

• Writing or ordering the questions themselves → write-interview-guide (owns the 7-step guide process, coverage matrix, leading/double-barreled lint)
• Watching someone attempt tasks on a product → moderate-usability-session (interviews capture what people SAY; usability tests capture what people DO)
• Interviewing business stakeholders about goals/constraints → conduct-stakeholder-interviews
• Observing people in their real environment → run-contextual-research
• What to do with the notes afterward → synthesize-research-data

The method — session arc

0. Pre-session (day before + 15 min before)

• Guide piloted at least once (owned by write-interview-guide — confirm it happened; if not, dry-run on a colleague now)
• Recording tool tested; backup recording method ready (phone, second device)
• Consent script printed/pinned — recording, data use, right to skip or stop
• Notes template open: two columns — Observation (verbatim quotes, timestamps) | Interpretation (your hunches, marked as yours)
• Observers briefed (see Observer management below) BEFORE the participant joins
• Your own state: single task. Facilitating IS the job — no Slack, no note-perfectionism. If budget allows, a dedicated notetaker; otherwise record and take sparse timestamp notes.

1. Opening (2–3 min)

Say, in your own words:

1. Who you are and the session length.
2. "I'm here to learn from your experience — there are no right or wrong answers, and nothing you say can disappoint me."
3. "You can skip any question or stop at any time."
4. "May I record this? The recording is only for my notes and is [deleted/kept] per [policy]." — wait for an explicit yes before recording.
5. "Any questions before we start?"

2. Rapport warmup (3–5 min)

• Open with easy, factual, self-descriptive questions ("What does a typical Tuesday look like for you?"). Never open with the hardest research question.
• Mirror lightly: match their pace and energy, nod, keep eye contact (camera lens on remote calls).
• Do not interrupt. Ever. Let every sentence land fully before you speak.
• Rapport is not chit-chat minutes — it's the participant learning, from your behavior, that long answers are welcome.

3. The funnel — broad to narrow

Run each topic in the guide as its own funnel:

1. Grand tour (broadest): "Walk me through the last time you [did the activity]."
2. Critical incident (anchor in a real memory): "Tell me about a specific time when [X] went badly / surprisingly well." Real recalled events beat generalities every time.
3. Narrow follow-ups: only after the broad layers, ask about the specific details the research questions need.
4. Never lead the funnel downward with your own vocabulary or hypothesis. Use THEIR words for THEIR things.

Phrasing law: open questions start with how / what / tell me about. Questions starting with did / was / is / do you are closed and often leading — save them only for factual confirmation ("Was that on your phone or laptop?").

4. Probing — the repertoire

Probe MORE than feels polite. Insufficient probing is a named NN/g facilitation mistake; the surface answer is almost never the finding.

Probe	How it works	Use when
Echo	Repeat their last key phrase with rising intonation: "…it felt like a chore?"	They said something loaded and stopped. Cheapest probe; invisible.
Boomerang	Return their question as a question: "What should this button do?" → "What would you expect it to do?"	They ask YOUR opinion or how something works. Never answer mid-session.
Columbo	Play naive, trail off, let them finish: "So you ended up calling instead, because…?"	You suspect a reason but must not supply it. Half-sentences pull complete stories.
Neutral expand	"Tell me more about that." / "What do you mean by [their word]?" / "What happened next?"	Any answer that's shorter than the experience it describes.
Significance	"Why is that important to you?"	They state a preference or ritual without stakes attached.
Silence	Say nothing. Count 5–10 seconds. Attentive face.	ALWAYS try this first after an answer. People fill silence with the good stuff.

Silence rule: the discomfort you feel at second 4 is yours, not theirs. If you jump in early you buy their headline and lose the story. Wait the full count; only then probe verbally.

Probe decision tree:

• Answer feels complete and concrete → silence anyway (once) → move on.
• Answer is abstract/general ("usually", "people tend to") → critical incident: "Tell me about the last specific time."
• Answer contradicts something earlier → Columbo, never confrontation: "Earlier you mentioned [X], and just now [Y] — help me put those together…"
• Answer is a question to you → boomerang.
• Answer is emotional but unexplained → echo the emotion word.

5. Live mistake recovery

You notice	Fix, in the moment
You just asked a leading question ("Don't you find that annoying?")	Don't apologize at length. Re-ask neutrally: "Let me ask that better — how did that feel?" Mark the tainted answer in notes.
You rephrased their observation in YOUR words and they agreed	Their agreement is worthless (acquiescence). Re-elicit: "Say it back to me in your own words?"
Participant is terse (one-word answers)	You narrowed too fast. Climb back UP the funnel to a grand-tour question; re-warm rapport; slow your own pace.
Participant rambles off-topic	"That's really helpful — I want to make sure we have time for [topic]. Earlier you said [bridge]…" Redirect via THEIR earlier words, not your agenda.
Participant asks "am I giving you what you need?"	"You're doing exactly right — there's no wrong answer here." Then continue; do not evaluate specific answers.
You're behind on time	Cut whole topics (lowest-priority per the guide), never compress probing on the topics you keep. Ten shallow answers < four deep ones.

6. Observer management

Poorly managed observers are a named NN/g facilitation mistake — they can wreck rapport in one interjection.

• Before the session: brief observers — silent during; cameras off/muted on remote; questions go to a back channel (chat/doc) and YOU ask them at the end, rewritten neutrally.
• Introduce honestly, minimally: "A colleague is joining to take notes." Never a wall of silent faces — cap visible observers at 1–2; the rest watch the stream/recording.
• If an observer jumps in with a leading question: absorb and repair — "Building on that: how do YOU think about [topic]?" Debrief the observer after, not in front of the participant.
• Observer questions at the end: filter them. Rewrite leading ones; drop out-of-scope ones. You own the session's neutrality.

7. Closing (5 min)

1. "Is there anything I should have asked you about, but didn't?" — reliably yields a top-3 insight.
2. Check the back channel for observer questions; ask the filtered ones.
3. Thank them, state what happens next, deliver/confirm the incentive.
4. Stop recording ONLY after asking "anything else?" — post-"official-end" comments are often the most honest; if they say something valuable after recording stops, ask permission to note it.

8. Debrief (within 30 minutes, non-negotiable)

Memory decays in hours. Before opening anything else, write the debrief (format below).

The say-vs-do caveat (interpretation law)

Interviews are attitudinal self-report. People misremember, rationalize, and perform social desirability — not from dishonesty, but because memory works that way.

• An interview answer is evidence of what the participant believes and how they frame it — never proof of what they actually do.
• Any claim shaped like "users do X" or "users will do X" sourced from interviews alone is under-evidenced by definition — handle it per craft-critique's evidence-discipline protocol. The data that strengthens it is behavioral (logs, observation, usability sessions), never more interviews.
• If the real question is "can people use this?" or "what do people actually do?" — an interview is the wrong instrument. Say so and route to moderate-usability-session or run-contextual-research.
• Prediction questions ("Would you pay for this?") are banned. People cannot predict their own behavior. Ask about past behavior instead.

Worked example — annotated exchange

Context: discovery interview about how salon clients book appointments (guide already written via write-interview-guide).

F: Walk me through the last time you booked an appointment at a salon. (grand tour — broad, anchored in a real event) P: Um, I usually just call. F: (silence, 6 seconds) (resists "why not online?" — leading + narrowed too fast) P: …I mean, I tried their website once but I couldn't tell if the person I like was even working that day, so calling is just safer. F: Safer? (echo — one word, their word) P: Yeah — last time I booked online I got someone new and the cut was wrong, and I had a wedding that weekend. Never again. F: Tell me about that time — what happened after? (critical incident + neutral expand) P: I called and they fixed it, but I had to take another morning off work… F (later): So now, before every booking, you… (Columbo — trails off; does NOT supply "call to check first") P: I text the receptionist first. She tells me when Maria's in. THEN I book.

What the funnel bought: the headline was "I usually just call." The finding — a trust-and-recovery workaround routed through a specific stylist and a back-channel text — only surfaced through silence → echo → incident → Columbo. Notes record: Observation: "Never again" + wedding story (quote, 00:14:20). Interpretation (mine): stylist continuity may outweigh booking convenience — check against other participants. And per the say-vs-do caveat: this is her account of what she does; if booking-flow redesign hangs on it, corroborate with booking logs or observation.

Anti-patterns / red flags

Don't	Do
Fill silence at second 3	Count to 8. Their next sentence is the finding.
"Don't you think X is frustrating?"	"How did that feel?" — they name the emotion.
Answer "how does this work?" mid-session	Boomerang: "What would you expect?"
Rephrase their point in your words for confirmation	Ask them to say it in their words. Their nod to YOUR phrasing is noise.
Accept "usually I…" generalities	"Tell me about the last specific time."
"Would you use/pay for this?"	"When did you last [related past behavior]? Walk me through it."
Take verbatim-everything notes while facilitating	Record + sparse timestamped notes; full attention on the person.
Let a stakeholder interject live	Back channel, filtered by you, asked at the end.
Treat interview quotes as behavioral proof in the readout	Label self-report as self-report; triangulate per craft-critique.
Skip the debrief because the recording exists	Write the debrief within 30 min — recall, not transcripts, holds the connections.

Output format — post-session debrief

## Debrief — P[n], [date, session #]
**Top 3 takeaways:** (one line each — what you'd tell the team in the hallway)
**Surprises / contradictions:** (incl. says-vs-earlier-says tensions to test in later sessions)
**Verbatim quotes worth keeping:** (with timestamps)
**Facilitation notes:** (questions that flopped, leading slips → feed back to write-interview-guide)
**Guide changes before next session:** (specific edits, or "none")
**Coverage:** research questions touched / still starving

Sources

• User Interviews 101 — https://www.nngroup.com/articles/user-interviews/
• 5 Facilitation Mistakes (poor rapport, multitasking, leading, insufficient probing, poorly managed observers) — https://www.nngroup.com/articles/interview-facilitation-mistakes/
• Why User Interviews Fail (self-report limits) — https://www.nngroup.com/articles/why-user-interviews-fail/
• The Funnel Technique in Interviews — https://www.nngroup.com/videos/funnel-technique-interviews/
• Probing in User Interviews — https://www.nngroup.com/videos/probing-user-interviews/
• Open-Ended vs. Closed Questions — https://www.nngroup.com/articles/open-ended-questions/
• Leading Questions — https://www.nngroup.com/articles/leading-questions/

Boundaries

• write-interview-guide owns question creation, funnel ORDERING on paper, coverage matrix, and bias lint. This skill owns live delivery and adaptation of that guide.
• moderate-usability-session owns task-based sessions (think-aloud, no premature rescue). If the participant is USING a product during the session, that skill governs.
• conduct-stakeholder-interviews owns business-goal interviews and power-interest context.
• run-contextual-research owns interviews conducted in the user's environment interleaved with observation.
• synthesize-research-data owns everything after the debrief (coding, themes, need statements).
• craft-critique is the single source of the evidence-discipline protocol — this skill applies it to self-report claims, never restates it.