Research

Design a non-leading user interview guide

Build a neutral conversation that elicits recent behavior, context, trade-offs, and evidence rather than feature approval

9 min setupTested with:ChatGPTReviewed: 2026-08-28

Add your context

Your text stays in this browser. AILesson Prompts does not send it to a model or server.

Study questions and decisionParticipants and ethicsSession constraints

Your prompt

Unfilled fields remain visible as placeholders, so you can still copy and edit the prompt

Design a neutral user interview guide aligned to the supplied research questions.

Research questions, decision, evidence, hypotheses, unknowns, and concepts:
[study]

Participants, exclusions, recruitment, consent, incentive, privacy, sensitive topics, accessibility, language, and withdrawal:
[participants]

Duration, mode, moderator, observers, recording, artifacts, concept order, tasks, analysis fields, and requirements:
[session]

Allocate time for welcome, informed consent, context, one or more recent concrete instances, current workflow and artifacts, breakdown and workarounds, consequences, variation and counterexamples, concept or task evaluation if approved, and closing. Start with behavior before opinions or future predictions. Use one idea per open question and neutral probes such as “what happened next?” Avoid mentioning the desired answer, asking why in a blaming way, presuming a problem, using praise that signals correctness, or asking whether participants like a solution. When showing concepts, explain their provisional status, counterbalance order where possible, ask participants to interpret and complete realistic tasks, and separate comprehension, usability, value, trust, and preference. Do not collect unnecessary sensitive data. Include permission points, skip and withdrawal paths, moderator cautions, observer rules, exact artifacts to capture, and post-session factual notes versus interpretation. Audit every question for leading premises and direct connection to a research question.

Private by defaultPrompt assembly happens locally in your browser. Avoid placing confidential information into any AI service unless your organization allows it.

From input to outcome

A worked example

See how concrete context turns this recipe into a usable result

Actual input

Study questions and decision: Decision: whether to explore email reminders, in-product ownership cues, or neither for Ready handoffs over 48 hours. Questions: what situations create long Ready states; how roles notice and act; what information enables action; costs and benefits of each concept. Evidence suggests unclear accepters, forgotten status, deliberate waiting, and notification fatigue. Two low-fidelity concepts approved after privacy review.
Participants and ethics: 12 UK/Germany active users: requesters, owners, and accepters, half from workspaces with long Ready events and half comparison. Recruitment with existing research consent. No managers interviewing direct reports. GBP 60 equivalent incentive. Avoid collecting customer content; artifacts must be fictional or participant-redacted. Support screen readers, keyboard navigation, English or German. Participants may skip, stop, or withdraw recording consent.
Session constraints: 60-minute remote moderated session, one researcher and silent note-taker. Recording optional. Use one recent instance plus two clickable prototypes; alternate A/B order. Capture timestamps, exact phrases under short-quote consent, task actions, errors, and confidence. Stakeholders observe only through separate muted room and cannot message moderator.

Example output

0-5 min — Welcome and consent. Explain study purpose without naming a preferred concept; confirm recording separately, short-quote use, fictional/redacted artifacts, accessibility setup, incentive, and right to skip/stop. Observers remain unnamed by role only if approved consent wording says so; otherwise disclose them.

5-10 — Context: “What part do you usually play when work moves from ready to accepted?” “What tools or signals are involved?” Do not ask whether handoffs are frustrating.

10-28 — Recent instance: “Think of the most recent handoff that stayed ready longer than you expected—or, if none, one that moved as expected. Walk me through it from the moment it became ready.” Probes: “What happened next?” “Who knew?” “What did you see?” “What did you do?” “What prevented or enabled that action?” “Was the waiting intentional?” Ask for a redacted artifact only after permission. Capture counterexamples.

28-35 — Variation: “When does this process work differently?” “Tell me about a case where no reminder was needed.” “What information would have changed your next action, if any?”

35-52 — Concepts, order alternated. State: “These are incomplete ideas, not committed features.” Task for each: “Imagine the fictional handoff has been Ready for 52 hours. Show what you think is happening and what you would do next.” Follow with interpretation, actionability, missing information, unwanted consequence, trust, and accessibility questions. Ask preference only after both tasks and require reasoning; do not ask “Would this helpful reminder solve the problem?”

52-57 — Comparison: “In which situations, if any, should neither concept act?” “What would make either option harmful or ignorable?” 57-60 — Recap, correction, questions, withdrawal reminder.

Post-session notes separate participant statements/actions from researcher interpretations and link each to research question. Log concept order, response mode, recording status, task completion, errors, and counterexamples.

Why this works

1
Recent-instance reconstruction yields inspectable behavior rather than polished general opinion.
2
Concept tasks separate what people understand and can do from what they say they like.

Check the result

Does every question avoid a preferred premise and map to a research question?
Does the guide prioritize recent behavior and counterexamples over prediction?
Are consent, privacy, accessibility, withdrawal, observer, and concept-order controls explicit?

More ways to explore