How is a discussion guide different from a survey?

A survey fixes its questions and answer options in advance and every respondent gets the identical instrument. A discussion guide fixes the objectives and core questions but expects the conversation to branch based on answers. Surveys measure known questions at scale; guided interviews explore the questions you did not know to ask.

How long should a discussion guide be?

Plan roughly one core question plus probes per five to seven minutes of session time. A 30-minute interview supports four to six core questions. Guides that try to cover fifteen topics produce shallow answers on all of them; depth requires leaving room to dwell.

Should the participant see the discussion guide?

No. Participants who see upcoming questions rehearse answers and skip ahead. They should know the topic and purpose of the study, for consent and comfort, but the specific questions and probes work best arriving naturally in sequence.

What Is a Discussion Guide in User Research?

A discussion guide is the structured plan a moderator follows in a user interview, and AI moderation changes the craft.

A discussion guide is the structured plan a moderator follows during a user interview or focus group: the research objectives, the questions to ask, the order to ask them in, and notes on where to probe deeper. It is a map rather than a script. A good guide keeps every session anchored to the same learning goals while leaving the moderator free to follow interesting detours, which is where much of the value in qualitative research hides.

What goes in one

Most guides share a recognizable anatomy, whatever the template:

Objectives up top. Two or three sentences stating what the study must learn. Every question below should trace back to these; questions that trace to nothing get cut.
Warm-up. A few minutes of easy, low-stakes questions that establish rapport and context. People give better answers once they have talked a little.
Core sections. The substance, organized by topic and roughly time-boxed. Questions here are open-ended and behavioral: “Walk me through the last time you…” rather than “Would you use a feature that…”. Hypotheticals invite polite fiction; recalled behavior invites truth.
Probes. Pre-written follow-ups attached to key questions: “Why was that frustrating?”, “What did you do next?”, “What did you expect to happen?”. Probes are where guides earn their keep, because unprompted moderators default to moving on.
Cool-down and close. A catch-all (“Anything we didn’t ask about?”), thanks, and next steps.

The craft rules are consistent across the field: never lead the witness, one idea per question, sequence from general to specific so early questions do not contaminate later answers, and plan for fewer questions than the time seems to allow, since good interviews run long on the answers that matter.

How AI moderation changes the craft

In an AI-moderated interview, the discussion guide is no longer a private aid for a human; it becomes the operating instructions for the AI moderator itself. That shifts what you write. Objectives must be explicit enough for a model to judge when an answer is complete versus when to probe. Follow-up strategy becomes a per-question decision: some questions need exact wording with no deviation, others benefit from the AI generating live follow-ups based on what the participant just said.

In unitQ Research, for instance, authors set each question’s follow-up behavior, from none at all to fully live AI probing, and the question itself is asked as written. The guide’s discipline transfers directly: vague objectives produce aimless AI probing just as they produce aimless human interviews. The difference is scale. A guide written once can now moderate hundreds of concurrent sessions in voice, chat, or video, which raises the stakes on getting the guide right, since a leading question is no longer one moderator’s bad habit but a systematic bias applied to every session.

Why it matters

The guide is where research quality is mostly decided, before any participant shows up. It converts a vague ambition (“understand churn”) into answerable questions, keeps multiple sessions comparable enough to synthesize, and protects against the two classic failures of unstructured conversations: leading participants toward the answer you hoped for, and drifting through pleasant chat that yields nothing decision-relevant. Teams that skip the guide do not get more natural conversations; they get unanalyzable ones.

A worked example

A product team wants to understand why trial users fail to convert. The first draft guide asks “What would make you upgrade?”, a hypothetical that produces feature wish lists. The revised guide asks trial users to reconstruct their actual trial: what they tried first, where they got stuck, what they compared the product against, and what happened the day the trial expired. Probes push on each stall point. Twenty sessions later, the synthesized answer is unglamorous and actionable: most non-converters never reached the moment the product demonstrates value, because setup required data they did not have handy. No wish-list question would have found that.