AI Discussion Guide Generator: Auto-Generate Interview Guides From Your Research Goals

The Bottom Line

An AI discussion guide generator is a tool that takes a plain-English research goal and returns a complete, moderator-ready interview guide — warm-up questions, ordered core questions across multiple response formats, follow-up probing rules, and a closing section — in under a minute. Instead of staring at a blank Google Doc trying to remember the right Mom Test phrasing, you describe what you need to learn and the AI drafts an entire guide grounded in real research methodology.

The two big shifts in 2026: (1) generators no longer just produce static templates — they produce branded, methodology-aware guides that adapt to your study type (discovery, evaluative, generative, JTBD, NPS follow-up), and (2) the best generators don''t just hand you a doc — they convert the guide directly into a live AI-moderated interview that runs itself, freeing you from being the moderator at all.

This guide covers what AI discussion guide generators do, the methodology frameworks they should support, what separates a real generator from a glorified template gallery, and how Koji''s AI Consultant agent produces guides that are good enough to publish without manual rewriting.

What a Discussion Guide Actually Contains

Before evaluating any generator, get clear on what a good guide looks like. A complete moderator-ready interview guide has six sections:

1. Header — Study name, research question(s), target participant profile, expected length (typically 30–45 min live, 10–15 min async text, 20–30 min async voice).
2. Warm-up (3–5 minutes) — 2–3 low-stakes questions to build rapport and confirm the participant matches the screener. "Tell me a bit about your role" type prompts.
3. Context and current state (8–12 minutes) — Open-ended exploration of how the participant operates today, with adaptive probing.
4. Core research questions (15–25 minutes) — The questions that directly answer the research objective. Mix of open-ended, scales, and choice questions.
5. Reaction and reflection (5–10 minutes) — If you''re testing a concept, prototype, or pricing, this is where it goes. Otherwise, reflective synthesis questions.
6. Wrap-up (2–3 minutes) — "Anything I didn''t ask that I should have?", thank-you, next steps.

Each question should carry a tag: type (open-ended, scale, single-choice, etc.), must-ask vs. nice-to-have, and probing depth (0 = ask once, 3 = probe up to three times). Without these tags, a "discussion guide" is just a list of questions, not a research instrument.

A serious AI discussion guide generator produces all six sections fully populated and tagged. A weak generator gives you a numbered list and calls it done.

How an AI Discussion Guide Generator Works

The good ones follow a four-step internal loop:

1. Goal intake

You describe the study in 1–3 sentences:

• "Why are paying customers downgrading from the Pro plan?"
• "What jobs are RevOps leaders hiring our category to do?"
• "Test the appetite for a usage-based pricing model among design partners."

2. Methodology selection

A real generator detects the methodology that fits the goal. "Why are customers downgrading" → churn / Mom Test framing. "What jobs are leaders hiring our category for" → Jobs-to-be-Done switch interview structure. "Test appetite for usage-based pricing" → concept validation with willingness-to-pay anchors.

Each methodology has its own phrasing rules. JTBD asks about the moment of switch and the forces of progress; Mom Test forbids hypotheticals and pricing-feasibility questions; Generative explores broad context before narrowing.

3. Question drafting

The generator writes questions in the selected methodology''s voice. It picks the right response format per question (scale for benchmarks, single-choice for current state, open-ended for narratives), assigns probing depth, and orders them for low cognitive load. A good rule of thumb: easy → medium → hard → easy. Open with low-stakes warm-ups, escalate to harder reflective questions in the middle, end with synthesis.

4. Validation pass

The generator self-checks the draft against the methodology rules. Did it accidentally introduce a leading question? Rewrite. Is the same construct asked twice? Consolidate. Is the guide longer than the target duration? Trim.

The output: a guide that''s ~80–90% publish-ready out of the box. You can edit any question, change any type, or rewrite the order — but you don''t have to start from zero.

Methodology Frameworks Every AI Generator Should Support

Different studies need different question styles. A serious AI discussion guide generator natively supports at least these five frameworks:

Koji''s AI Consultant agent embeds all five as runtime principles — not just labels. When you select Mom Test methodology, the Consultant will literally rewrite any hypothetical-future question into a past-behavior question. When you select JTBD Switch, it will insert the four forces of progress (push, pull, anxiety, habit) as probing rules.

See working with the AI consultant for how this works under the hood.

What Separates a Real Generator From a Template Gallery

Many tools that call themselves "AI discussion guide generators" are actually template galleries with a chatbot UI on top. To tell the difference, ask:

1. Does it ask clarifying questions? A real generator asks back: "What decision does this research inform?" "What''s your current hypothesis?" "Who specifically should participate?" A template gallery doesn''t — it just spits out a generic doc.
2. Does it produce structured question types? A real generator tags every question with one of the six structured types (open-ended, scale, single-choice, multiple-choice, ranking, yes/no) so reports can visualize them properly. A template gallery outputs only open-ended prompts.
3. Does it adapt to the methodology? A real generator''s output for "JTBD switch interview" is structurally different from its output for "Mom Test discovery." A template gallery''s output is the same questions with different titles.
4. Does it produce a guide that can run itself? This is the 2026 acid test. A real generator''s output is a brief object that an AI interviewer can read and execute — running the live conversation, asking the questions, probing follow-ups, scoring the transcript. A template gallery just gives you a Word doc.

Koji passes all four checks. The output of the AI Consultant is a StudyBrief object containing structured StudyQuestion[] entries that the AI Interviewer agent executes directly. There''s no "convert your template into a survey" step — the guide and the runnable interview are the same artifact.

Sample Generated Guide: "Why Are Pro Customers Downgrading?"

Here''s a representative output from a real Koji generation for the goal "I want to understand why paying customers are downgrading from our Pro plan to Free."

Header

• Study type: Churn / Mom Test
• Target: Customers who downgraded from Pro → Free in last 60 days
• Expected length: 12 min text / 25 min voice

Warm-up

1. To start — tell me a bit about your role and what you''re working on these days. (open-ended)

Context and current state 2. When you first signed up for Pro, what were you trying to do? (open-ended, probe 2) 3. How often were you using Pro features at peak? (scale 1–5) 4. Which Pro features did you actually use? (multiple-choice)

Core research 5. Walk me through the moment you decided to downgrade. What was happening? (open-ended, probe 3) 6. Were there specific events or changes that pushed you toward the decision? (open-ended, probe 2) 7. What were you doing instead during the period before you actually downgraded? (open-ended, probe 1) 8. On a scale of 1–10, how disappointed would you be if Free was discontinued? (scale, anchor probe)

Reaction and reflection 9. If Pro had cost half as much, would you have stayed? (yes/no, probe) 10. What''s missing from Free that, if added, would''ve kept you on Pro? (open-ended)

Wrap-up 11. Anything I didn''t ask about that you think I should have? (open-ended)

That''s 11 tagged questions across five sections, generated from one sentence in under 60 seconds. Each question type maps to a chart in the report (scales → distribution; multiple-choice → stacked frequency; open-ended → themes + verbatim quotes). See the structured questions guide for how each type renders.

How to Generate Your First Guide With Koji

1. Sign up at koji.so — free tier, no credit card.
2. Click "New study."
3. Type your goal in 1–3 sentences.
4. Answer the AI Consultant''s 2–4 clarifying questions — takes 60–90 seconds.
5. Review the generated guide in the editable brief view.
6. Edit as needed — change any question text, switch types, reorder, add/remove sections. The brief view is fully editable.
7. Click "Publish" — get a shareable link, embed snippet, or API endpoint.
8. (Optional) Pin a custom methodology — Mom Test, JTBD, Customer Discovery, or your own framework loaded as a context document.

For deeper guidance on tweaking the brief, see editing the brief manually and understanding the research brief.

Cost Per Guide

Koji''s generator is included in every plan, including the free tier. There''s no per-guide fee. What you pay for is the interviews the guide collects:

• Free tier — 10 credits, enough for ~10 text or 3 voice interviews end-to-end
• Insights plan — €29/month, 29 credits
• Interviews plan — €79/month, 79 credits
• Per-interview cost — 1 credit text, 3 credits voice, 5 credits report refresh
• Bad interviews don''t cost — sessions scoring 1 or 2 out of 5 quality refund automatically

Generating and publishing a guide costs nothing. You only consume credits when actual respondents complete interviews against the guide.

When to Hand-Write the Guide Instead

AI generation is not always the right move:

• Standardized validated instruments. SUS, UEQ, validated psychometric scales — don''t regenerate, load as a template.
• Highly regulated audiences. Generate the draft, but have legal/compliance review every question before publishing.
• Single-question pulse polls. A simple thumbs-up/down doesn''t need a full guide — publish a single question directly.
• Re-runs of past studies. Clone the prior brief instead of regenerating — preserves longitudinal comparability.

For everything else — discovery, churn, onboarding, NPS follow-up, design partner check-ins, feature prioritization — AI guide generation is faster, more methodologically sound, and easier to iterate on than manual authoring.

Related Resources

• Structured Questions Guide — The 6 question types every generated guide should mix
• Discussion Guide Template for User Interviews — A reference structure for hand-built guides
• User Interview Guide Template — Generic interview guide scaffold
• Customer Interview Questions Examples — 50+ example questions across study types
• Writing Interview Questions — Principles for question phrasing the generator follows
• Working With the AI Consultant — How Koji''s Consultant agent powers the generator
• Editing the Brief Manually — Customizing a generated guide after the draft