AI Workflows · Guide
Draft User Research Questions With AI Without Leading People
Let AI propose interview wording, then remove assumptions that tell participants what to say.
AI can produce a long interview guide in seconds. The danger is that plausible questions may already contain the answer the team hopes to hear. “How much easier is the new dashboard?” assumes it is easier. “Would you use an AI summary every day?” asks for a prediction about a hypothetical future. Both can produce confident responses that tell little about real behavior. A stronger guide starts with the decision the research should inform. If the team needs to know whether people can find a project decision later, ask how they do that today. AI can help vary the wording and suggest follow-ups, but the researcher chooses questions that let participants surprise the team.
Write the learning goal first
Describe the behavior, context, and uncertainty. “We need to learn how team leads recover a decision made in a meeting” is specific enough to guide questions. List what you already know and what you are merely assuming. This keeps the model from turning assumptions into facts in the interview guide. Ask for a short guide: opening context, a few experience questions, prompts for concrete examples, and a close. More questions are not automatically better. Leave time for the participant's story and for the interviewer to follow an unexpected detail.
Illustrative prompt · adapt before use
Draft neutral interview questions for the stated learning goal. Prefer recent behavior, avoid presupposing tool use or benefit, and explain which assumption each revised question removes.Prefer recent behavior over predictions
Questions about a specific past event usually yield richer detail. “Tell me about the last time you needed to find a decision from a meeting” invites a sequence: what triggered the search, where they looked, what they found, and what they did next. “Would an assistant help you find decisions?” invites speculation. Use neutral follow-ups: “What happened next?”, “How did you know that was the current answer?”, and “Can you show me where you looked?” Avoid praising a response or offering a solution before the participant finishes describing their process.
Audit assumptions line by line
Review the AI draft for adjectives that imply success or failure: easier, frustrating, better, confusing. Check for two questions packed into one, product terms the participant may not know, and a preferred answer hidden in an example. Rewrite “How frustrating is it to search across tools?” as “What did you do when the answer was not in the first place you checked?” Also check whether a question asks participants to disclose information the study does not need. A good guide respects the scope of consent and avoids collecting unnecessary personal or sensitive details.
Pilot the guide before recruiting
Run one practice interview with a colleague who does not know the desired answer. Notice where they ask for clarification, where the conversation becomes abstract, and where the interviewer starts explaining instead of listening. Cut redundant questions. Add a prompt that asks for an artifact or example when memories are vague. During the real interviews, treat the guide as support, not a script. Record what participants actually did or said, including contradictions. AI can help organize notes after the session, but compare any synthesis with the recordings and do not erase minority patterns merely because they are harder to summarize.
Check the guide with a neutral practice interview
Before inviting participants, ask a colleague who did not help write the guide to answer it as if they were describing a recent task. Listen for places where the question suggests a preferred feature or requires the participant to defend a decision. For example, “How did the helpful AI assistant improve your work?” assumes both use and benefit. A neutral alternative is “Tell me about the last time you completed this task. What tools, if any, did you use?” Follow-ups can ask what happened, what was difficult, and what the person did next. They should not smuggle the team's hypothesis into the participant's mouth. During the practice, note whether questions invite concrete stories and whether the interviewer has room to ask an unexpected follow-up. AI can suggest a draft guide, but the researcher owns the learning goal and the live conversation. Revise loaded or repetitive questions, then try the guide once more if changes are substantial. Keep the final guide short enough that the participant has time to describe actual behavior rather than answer a long survey disguised as an interview. Try this next: take five AI-drafted questions and mark each assumption, prediction, or loaded adjective. Replace them with questions about one recent real event, then pilot the revised guide.
Sources and update note
Follow the linked official source before a product, price, plan, or policy decision.