What is a discussion guide?
A discussion guide, also called an interview guide, is the document a moderator uses to run a conversation. It keeps the interview on topic without turning it into a script. You decide in advance what you need to learn, in what order to explore it and what to ask if the first answer is thin.
It is a guide and not a questionnaire. Good interviews wander a little when something interesting appears. The guide tells the moderator where to come back to.
Why does a guide matter more with an AI moderator?
A human moderator can improvise around a weak guide. An AI moderator works from what you give it, so a clear goal and well worded questions have a direct effect on the quality of the interview. Writing the guide carefully is the most useful hour you spend on the study.
Part 1, study goals
Write two to four sentences at the top. They are for you and the moderator, not for the participant.
- Decision this study supports: for example, whether to change the onboarding flow.
- What we need to learn: for example, why new customers stop before the first project is finished.
- Audience: for example, customers who signed up in the last 60 days.
- What we are not studying: for example, pricing opinions.
Keep to one main goal. A guide with six goals gives you six shallow answers.
Part 2, opening script
The opening sets expectations and puts the participant at ease. Keep it under a minute.
"Thanks for taking part. I am going to ask about your experience with [topic]. There are no right or wrong answers, and I am interested in what actually happened, not what you think I want to hear. It takes about [number] minutes. Your answers are used to improve [product or decision]. You can skip any question."
Add one line on how the answers will be used, because people share more when they know why.
Part 3, warm-up questions
Start with an easy question about the person and their context.
- Tell me a little about your role and what a normal week looks like.
- How do you currently handle [the area you are studying]?
These questions also help you read later answers, because you know the background.
Part 4, core questions
Write five to eight questions in the order you want them asked. Start broad and narrow down. For each, add one or two probes.
- Question: Tell me about the last time you [did the thing]. Probe: What led up to it? What happened next?
- Question: What was the hardest part? Probe: What did you try? How did that go?
- Question: How did you decide to [choose, switch or stop]? Probe: What else did you consider?
- Question: What would have made it easier? Probe: Can you give an example?
- Question: If you could change one thing, what would it be? Probe: Why that one?
Notice that each question asks for a specific recent event. Specific stories carry more reliable information than opinions.
Part 5, probes
Probes are the small follow-ups that turn a short answer into a useful one. Keep a list nearby.
- Can you say more about that?
- What do you mean by [their word]?
- Can you give me an example?
- What happened just before that?
- How did that make the work harder or easier?
- What did you do instead?
Part 6, closing
End with a question that catches what you missed, then thank the participant.
"Is there anything I did not ask that you think I should know?" Then: "Thank you. That was very helpful."
How do I write questions that get useful answers?
Follow a few rules and most weak guides improve at once.
- Ask open questions that begin with what, how or tell me about. Avoid questions that can be answered with yes or no.
- Ask about the past, not the future. "Would you use this?" produces optimism. "When did you last try to do this?" produces facts.
- Do not lead. "Don't you find the dashboard confusing?" plants the answer. "How did you find your way around the dashboard?" does not.
- Ask one thing at a time. Two questions in one sentence get a half answer.
- Use the participant's words. If they say "report", do not switch to "analytics view".
- Leave the solution out. Describe the problem and listen before showing an idea.
The same principles apply to topic lists for customer interview questions, where you will find a long list of examples.
How long should the guide be?
A conversation of twenty to thirty minutes fits roughly five to eight core questions with probes. If you have more questions than that, cut. Rank them by how much they affect your decision and drop the bottom third. Participants notice when an interview drags, and answers get shorter.
A rough split for a thirty minute interview:
- Opening and warm-up: three minutes.
- Core questions: twenty minutes.
- Concept or prototype reaction, if any: five minutes.
- Closing: two minutes.
Adjust the numbers, but keep the closing, because the last open question often produces the best quote.
How do I use the template with an AI moderator?
With Moderators you describe what you want to learn and who you are talking to, and the guide is drafted from that. You can then edit goals, questions and probes before you send the link. The moderator follows the order and depth you set, asks follow-ups when an answer is short, and records the conversation so findings can point back to quotes. See how AI moderated research works for the full sequence.
Before sending the link to everyone, run it yourself once as a participant. If a question confuses you, it will confuse them. Fix the wording first, then invite your customers.
- Check the guide fits in the time you promised.
- Read your own opening aloud and trim it.
- Make sure each core question maps to a goal.
- Remove any question you would not act on.
To compare how different interview styles handle a guide, see moderated vs unmoderated testing.
What mistakes should I avoid?
- Too many goals: pick one decision per study.
- Questions that hide a pitch: if you want to sell, do not call it research.
- Hypothetical questions: swap "would you" for "when did you last".
- No probes: without follow-ups the answers stay at the surface.
- Skipping the pilot: one test run catches most wording problems.
- Reading answers only as summaries: go back to quotes to check that the theme is really there.