Interview Guide Builder

Interview Guide Builder

A discussion guide is scaffolding, not a script: it keeps the researcher oriented while leaving room for the participant to lead. The costly failure it prevents is manufactured confirmation - a guide full of leading questions returns exactly the answers the team already believed, and the product ships into a market that was never actually heard.

Operating procedure

Step 1: Gather inputs

Collect before drafting:

• Research goal in one sentence, and whether the round is generative (understand behavior and needs) or evaluative (react to a concept or prototype). Default: generative.
• The 3-5 things the team must learn - these become topic blocks; anything beyond five gets cut or deferred to another round.
• Session length (default 60 minutes), participant profile, and whether sessions are recorded.
• The team's current hypotheses - written down explicitly so they can be kept OUT of the question stems.

Label any assumed goal as a guess and confirm.

Step 2: Build the structure

Every guide follows this arc, in order, because trust must be established before depth is possible:

1. Intro (2-3 minutes): who you are, why you are talking, that there are no right answers, consent to record if applicable.
2. Warm-up: 1-2 low-stakes context questions that get the participant talking about themselves.
3. Core topics: 3-5 areas maximum, ordered most-important-first - participant energy fades, so front-load what the team cannot leave without.
4. Edge probes: deeper follow-ups on the richest threads.
5. Close: "Anything else you wish I had asked?" - this question routinely produces the best quote of the session.

Step 3: Write open, non-leading core questions

Every core question must pass three tests: (1) it cannot be answered yes or no, (2) it does not contain the answer or an assumption, (3) it invites storytelling about specific past behavior, not hypotheticals. Prefer "Tell me about the last time you..." and "Walk me through how you..." stems.

Strip evaluative adjectives - "easy", "hard", "frustrating", "helpful" - from every question stem. Those words are the team's hypotheses, and planting them harvests them back as fake findings.

Contrast pair:

Bad: "Do you find it frustrating when the export takes a long time?" (yes/no, contains the hypothesis "frustrating", supplies the topic "export speed", invites polite agreement)

Good: "Walk me through the last time you got data out of the tool. What happened?" (open, past-behavior, no planted evaluation - if export speed is a real pain, the participant will bring it up unprompted, and that unprompted mention is the actual evidence)

Step 4: Attach a probe bank

After each core question, attach 2-3 probes. Reliable neutral probes: "What happened next?", "Can you say more about that?", "How did that make you feel?", "What were you trying to accomplish at that point?", "Was that typical for you?". Avoid "Why?" as a standalone probe - it reads as accusatory; reframe as "What was driving that for you?". Add silence to the kit: a 3-4 second pause makes participants fill the gap with the unrehearsed version.

Step 5: Budget time and flag must-covers

A 60-minute interview fits 4-5 core questions with probes - more means every answer stays shallow. Write a soft time budget in the margin of each topic block. Flag the one or two "must cover" questions the team cannot go without; when the conversation runs long (it will), everything unflagged is what gets dropped.

Step 6: Run the bias audit before fielding

Review the final draft line by line for:

• Confirmation bias - questions that assume the product works or the problem exists.
• Leading language - any planted adjective or suggested answer.
• Compound questions - two questions joined by "and"; participants answer only one.
• Closed questions - anything answerable with yes/no.
• Read every question aloud; awkward phrasing rarely survives being spoken, and a question you stumble over will derail live.

Step 7: Add mode-specific blocks

For concept or prototype testing, append a "react aloud" task block after the core questions - reactions come after unprimed behavior questions, never before, or the concept contaminates everything. For longitudinal studies, version the guide across waves and mark which questions repeat verbatim for comparability.

Deliverable

Produce a fielded-ready guide containing: intro script, warm-up, 3-5 core topic blocks each with an open lead question, 2-3 attached probes, a margin time budget, must-cover flags, a close question, and a completed bias-audit checklist confirming zero leading or compound questions.

Do NOT

• Do NOT ask about hypothetical future behavior ("would you use...", "would you pay...") - stated intent does not predict action; ask what they did last time.
• Do NOT put the concept reaction before the behavior questions in evaluative rounds; priming poisons the baseline.
• Do NOT write more than 5 core questions for an hour - coverage at the cost of depth returns nothing usable.
• Do NOT let the guide become a script read verbatim; the order can flex when the participant opens a rich thread.
• Do NOT skip the consent/recording line to save time.

Quality bar

• Every core question passes all three tests: not yes/no, no embedded assumption, invites a story.
• Zero evaluative adjectives in any question stem.
• Every core question has at least two probes attached and a time budget.
• Must-cover questions are flagged and appear early.
• The guide was read aloud once before fielding.