Learn · methods

By Fieldwork Research Team · Published 2026-05-05 · Updated 2026-05-11

How to write a qualitative research discussion guide

A discussion guide is the structure that makes a qualitative interview productive rather than a conversation that happens to be recorded. It defines the topics to cover, the sequence to cover them in, the depth required at each topic, and the questions that open each section. A well-written guide produces sessions that are consistent across participants and rich enough to support meaningful analysis. A poorly written one produces transcripts full of interesting tangents that don't answer the research question.

What a discussion guide is and is not

A discussion guide is not a script. Reading from a list of fixed questions in a fixed order produces a semi-structured survey, not a qualitative interview. The guide is a map, not a route. It tells you where you need to get to without dictating exactly how you get there.

The key distinction is between the guide as a set of topics and the guide as a set of questions. A topic-based guide says "we need to understand what happened at the document upload step." A question-based guide says "ask question 7, then question 8." The first gives the interviewer the goal and the freedom to reach it through whatever conversational path makes sense. The second turns the interview into a questionnaire.

For AI-conducted interviews, this distinction matters even more. Sofi works from a topic structure with defined depth requirements and resolution criteria, not from a question list. The intelligence is in the structure, not the wording of individual questions.

The components of a good discussion guide

Introduction and framing

The opening of the interview sets participant expectations. It should explain what the session is about (without revealing your hypotheses), how long it will take, that there are no right or wrong answers, and that the participant can decline to answer any question. This takes two to three minutes and is not optional. Participants who don't understand what they're doing in an interview give worse answers.

Warm-up questions

Start with easy, concrete questions that don't yet address the core research topic. These serve two purposes: they ease the participant into the conversation and they establish important context. "Walk me through your role and what a typical research day looks like" tells you something useful and gets the participant talking before you ask them anything that requires real reflection.

Core topics

These are the substantive sections of the guide, the ones designed to answer the research question. Each core topic needs:

A clear topic statement that tells you what this section needs to achieve. Not "ask about onboarding" but "understand where confidence drops during document upload and what is driving it."

An opening question that anchors the conversation in specific, recent experience rather than general impressions. "Walk me through what happened when you got to the document upload step" produces better data than "how was the document upload experience?"

Probe questions for the most common responses. If the participant says it was fine, what does fine mean specifically? If they say it was confusing, what was the first thing that confused them? These are not questions to ask in sequence. They are fallback options when the opening question doesn't produce sufficient depth.

Depth and resolution criteria. How much do you need to learn about this topic before you can move on? A quick check-in needs one question and one probe. A behavioural deep dive needs six to eight turns of genuine exploration. Knowing the difference before you go into the field prevents the two most common time management problems in qualitative interviews: spending too long on low-priority topics and leaving high-priority ones underdeveloped. If you have not locked the study structure yet, start with the study design — the topics, depth settings, and resolution criteria belong there, not in the guide itself.

Closing

End with an open invitation: "Is there anything about your experience with this that you expected me to ask but didn't?" This is one of the highest-value questions in the guide because participants sometimes hold their most important insight back, waiting for the right question that never comes. This is the right question.

Best practice for opening questions

The opening question for each topic is the highest-leverage sentence in the guide. It determines whether the participant gives you a story or a summary.

Summaries are useless for qualitative analysis. "It was pretty straightforward overall but there were a few confusing bits" is a summary. "When I got to the upload screen, I spent about five minutes trying to figure out whether I needed a certified copy or a scan" is a story. The story contains specific, analysable information. The summary contains a disposition.

Opening questions that produce stories:

Anchor in a specific instance: "Tell me about the most recent time you had to..."
Ask for chronology: "Walk me through what happened from the moment you..."
Ask about a specific moment: "What were you thinking when you got to..."

Opening questions that produce summaries:

Ask for general impressions: "How was your experience with..."
Ask evaluation questions first: "Did you find the process easy or difficult?"
Ask hypothetical questions: "Would you normally..."

Write your opening questions and then check each one: does it invite a story or a summary? If it invites a summary, rewrite it.

How Sofi builds the discussion guide for you

When you write a research brief in Fieldwork, Sofi generates a complete study structure from it. This is the equivalent of a discussion guide, built for AI-conducted interviewing: a set of topics, depth settings for each, resolution criteria that tell Sofi what she needs to learn before moving forward, and interview logic that governs when and how she moves between topics.

Sofi's generation is not template-based. She interprets your research brief, identifies the implied methodology and the questions that need answering, and builds a topic structure calibrated to those goals. A brief about onboarding drop-off produces a different structure than a brief about feature adoption or pricing sensitivity, even if the surface-level questions look similar.

The process takes less than a minute. What comes back is a complete interview structure: topics named and sequenced, depth settings assigned, resolution criteria written, and edge conditions defined so the interview knows how to handle common response patterns.

This does not mean accepting whatever Sofi generates without review. It means you start the design process with a well-formed structure rather than a blank page.

How to assess Sofi's topic graph before you launch

The generated structure is a starting point, not a finished design. Before any study goes live, review it against your actual research goals. Here is what to check.

Are the topics scoped correctly?

Each topic should be resolvable in 5 to 10 minutes of focused conversation. If a topic reads like it could take 30 minutes to cover properly, it needs to be broken into two or three narrower ones. If a topic is so narrow it could be answered in a single sentence, it probably belongs as a probe question within a broader topic rather than its own section.

Do the resolution criteria describe observable evidence?

Resolution criteria tell Sofi what she needs to learn before moving forward. Weak criteria: "participant has discussed the onboarding experience." That gives her no real guidance. Strong criteria: "participant has named at least one specific moment of uncertainty, identified what caused it, and indicated what would have resolved it." That tells her exactly what evidence is needed.

Read each criterion and ask: if I saw this in a transcript, would I know the topic was substantively covered? If the answer is no, rewrite the criterion to describe specific, observable evidence.

Are there any dead ends?

A dead-end topic is one where the conversation has nowhere to go after the opening question. Read each topic and ask: if the participant gives a two-sentence answer, what does Sofi ask next? And after that? If you cannot answer those questions clearly, the topic needs redesigning before it reaches participants.

Sofi flags potential dead ends and other structural issues during generation. Review those flags seriously. They are based on the same structural logic that makes interviews produce usable data. A study that launches with known structural problems will produce known structural gaps in the findings.

Does the topic sequence make sense?

Topics should move from specific and concrete to broad and interpretive. Start with recent, tangible experience. Move toward meaning, motivation, and implication. Participants warm up through a session. The topics that require the most reflective depth should come after the participant has been talking for 10 to 15 minutes, not at the beginning.

Does every topic connect to the research question?

Read the research question and then read each topic. Can you explain in one sentence why this topic is in the study? If you cannot, the topic may be a distraction or a habit from previous research rather than something this study needs.

The pre-launch checklist for your discussion guide

Every topic is scoped to 5 to 10 minutes of focused conversation
Every opening question anchors in specific, recent experience, not general impressions
Every topic has a clear depth setting matched to what you need from it
Resolution criteria describe specific, observable evidence, not vague coverage
No topic is a dead end: there is a clear forward path for the most common responses
Topics sequence from concrete and specific to broad and interpretive
The closing section includes an open invitation for anything not covered
Every topic connects directly to the research question

What this looks like in practice

A product manager at a payments company needs to understand why merchants are not using the bulk reconciliation feature three months after launch. She knows from analytics that merchants are finding the feature. They are clicking into it. They are not completing their first reconciliation run.

She writes a brief: "Merchants are finding bulk reconciliation but not completing a first run. We need to understand what stops them at the point of use: whether the barrier is understanding, confidence, process fit, or something else."

Sofi generates a four-topic structure: first impressions of the reconciliation tool, the first attempted run and where it stalled, what the merchant was trying to accomplish and whether they found another way, and what would need to be true for them to trust the tool with a live reconciliation.

The PM reviews the structure. The third topic's resolution criteria are too abstract: "participant has explained their workaround." She rewrites them: "participant has named the specific alternative they used, described why it felt more reliable, and indicated what would have to change about the reconciliation tool for them to switch." The fourth topic looks like a dead end: if the merchant says "just make it simpler," there's nowhere obvious to go next. She adds a probe path: what does simpler mean specifically, and what is the most complex thing about the current flow?

The study goes live. Eleven sessions complete over three days. The finding is specific: merchants are abandoning at the account mapping step because the tool requires them to match their internal account codes to the platform's codes, and they don't have that mapping documented anywhere. The reconciliation isn't failing because the tool is confusing. It is failing because it requires information merchants don't have in front of them. That is a solvable problem, and the PM knows exactly what to build.

Frequently asked questions

How long should a discussion guide be?

Length should be determined by the number of topics and the depth required at each, not by a target word count or page count. A two-topic quick pulse guide might be one page. A six-topic behavioural study might be four. The test is whether a researcher unfamiliar with the project could pick it up and conduct a session that meets the research objectives. If yes, it is long enough. If it requires background knowledge to use, it is incomplete.

Should you pilot your discussion guide before full fieldwork?

Yes, whenever possible. A pilot session with one or two participants before the main study reveals whether opening questions produce stories or summaries, whether topic depth settings are calibrated correctly, and whether the total session length is realistic. Changes to the guide after a pilot are much cheaper than discovering the same problems after 10 sessions.

How much should the guide vary between participants?

The topics should be consistent. The questions within topics can vary based on what each participant brings to the conversation. A guide that prescribes identical questions for every participant regardless of what they say produces the consistency of a survey without the validity of a qualitative study. The guide should hold the topics constant and leave the path through each topic responsive to the participant.

What is the difference between a discussion guide and a topic guide?

The terms are often used interchangeably. Some researchers use "topic guide" to describe a lighter, more flexible structure with just topic headings and a few prompt questions, and "discussion guide" to describe a more detailed structure with opening questions, probes, and transition language. The distinction matters less than having a structure that serves the research question and the interviewer's needs.

How do you handle participants who go off topic?

Note the tangent and return to the guide. If what they raised is genuinely relevant to the research question but not in the guide, make a note to probe it when there is a natural opportunity. If it is not relevant, acknowledge it briefly and redirect: "that is useful context. I want to make sure we cover [topic], can you tell me about..." A discussion guide is permission to redirect, not just a reminder of what to ask.

How does Sofi's topic structure differ from a traditional discussion guide?

Functionally they serve the same purpose: defining what topics to cover, in what depth, and with what resolution criteria. The difference is that Sofi's structure includes machine-readable depth settings and resolution criteria that govern her behaviour in real time, not just notes for a human moderator to interpret. The research design logic is the same. The implementation is more precise because it needs to be executed consistently across every session without a human making judgment calls in the room.

Related on Fieldwork

Last updated: 2026-05-11