A great discussion guide bridges the gap between what you want to learn and how you ask it. By structuring your guide correctly, you ensure high-quality data collection, clean analysis, and seamless AI moderation.
This quick-start guide is your blueprint for building high-impact discussion guides in Marvin. Whether you’re running a live interview or deploying an AI moderator, you'll learn how to structure your questions, optimize your AI settings, and collaborate with your team.
Follow the steps below to create your discussion guide:
Step 1: Set up your guide's structure
To ensure Marvin automatically detects answers within interview files and links them back to your guide, you must follow specific formatting rules.
Use the forward slash (/) for every question
The most critical formatting rule: place a forward slash / at the start of every question.
This signals Marvin to recognize the text as a question, enabling automatic answer detection within interview files and linking responses to the correct guide. It also makes questions taggable as note labels during analysis, which is key to surfacing insights efficiently.
📝 Example:
/ What are your biggest frustrations with your current workflow?
Use "Detect Questions" for faster setup
Instead of manually formatting everything, use the Detect Questions feature to automatically scan your text and convert it into the proper slash format.
Add visual organization for moderators
Nesting, bullets, and numbered lists don't affect AI analysis, but they make the guide significantly easier to follow during a live session. Use them to group related questions, indicate branching paths, and separate sections.
Implement branching logic
Use conditional logic to direct the interview flow based on participant responses. For example, route a frustrated participant to a different set of questions, or adjust the focus based on a participant's role or tenure.
Pro tip: Write Instructions Without Slashes. Text without a forward slash acts as background information, coaching, or direct instructions for the AI (e.g., In the next question, we want to understand their seniority level).
Learn more: Create and use discussion guides for interviews
Step 2: Design effective questions
Well-designed questions prevent participant burnout and keep the AI from getting confused.
Tie every question to a research goal
Before finalizing, check that each question helps answer a project research goal. If you can't connect it to one, cut or reframe it. Start broad by understanding the user's role and overall goals. When asking for product feedback, logically group your questions by specific features or functions.
Ask one topic at a time using open-ended questions
Keep each question to a single topic. Avoid double-barreled questions and yes/no questions; both produce shallow data and make it harder for the AI to match responses accurately.
❌ "Do you find the current process difficult?"
✅ "Walk me through how you currently handle that process and where it gets challenging."
Use conversational language
Write questions as they'll actually be spoken. If your guide says "desired outcomes" but the moderator says "goals," the AI may miss the match. Include alternate phrasings when the language might vary. In follow-ups, name the specific feature or topic the participant mentioned rather than using vague phrases like "tell me more about that."
Limit your guide to ~30 questions
To get the fastest, most accurate insights out of Marvin, keep your discussion guide to approximately 30 questions. Sticking to this baseline ensures full analysis coverage from start to finish, maximizes post-interview processing speeds, and maintains crisp context tracking so your live notes and timestamps stay perfectly synced.
Pro tip: For usability tests, break the workflow into chunks and prompt participants to think aloud at each step rather than completing the whole flow at once.
Step 3: Organize your content strategically
Where you put information matters as much as what you write. Marvin's guide has two distinct sections: to keep your research workspace organized, intentional, and optimized for long-term data analysis.
Separate research goals from the discussion guide
Research goals page: Research Goals section: The "why": context, objectives, hypotheses, and background that help the AI ask smarter follow-up questions without cluttering the guide itself.
Discussion guide page: Discussion Guide section: The "how": exact questions, step-by-step instructions, task descriptions, and moderator prompts.
Detailed research goals allow the AI moderator to ask smarter follow-ups without you having to anticipate every possible follow-up question in the guide.
Add rich context to the guide
The more context you provide, the better Marvin performs across auto notes, Ask AI, and insight matching. Beyond research goals, consider including:
An opening script the moderator or AI will use to introduce the session
Potential follow-up questions for key topics
Interviewer prompts and guidance for sensitive or complex areas
Background on the participant type or product area being discussed
Prepare relevant labels
As you write each question, think about how you'd categorize the answers. This helps you decide which existing label template fits your study, or helps you identify new labels to create. Well-matched labels make it much easier to isolate and compare responses during analysis and Ask AI queries.
Include compliance information
Use the advanced settings to add disclaimers, privacy notices, or "read me" links that participants must accept before the interview begins. You can also configure participant limits and incentive redirects here.
Learn more:
Step 4: Leverage AI for guide creation
Marvin’s native AI tools speed up your discovery process and enhance your follow-up precision.
Brainstorm with Marvin's AI
Use the AI generator to build initial drafts and brainstorm question angles. You can easily refine and expand the AI-generated template to fit your exact goals.
Generate Guides from Existing Data
Feed the AI generator external context—such as Product Requirement Documents (PRDs), Notion pages, project overviews, or up to five previous interview files—to help it craft highly relevant hypotheses and questions.
Define Your Target Audience
When generating a guide, provide specific details about the target audience (like job titles or key demographics) to ensure the AI tailors the tone and complexity perfectly.
Learn more: Use Marvin's AI to create discussion guides
Step 5: Collaborate, manage, and iterate
Good research is rarely a solo endeavor. These practices help your team stay aligned, and your guides stay reusable.
Establish naming conventions
Use clear, consistent names that include the persona or moderator type, for example, "Moderator Guide: Enterprise Onboarding" or "AI Guide: SMB Checkout Flow." This makes guides easy to find in dropdown menus and prevents confusion across projects.
Save successful guides as templates
Once a guide performs well, save it as a template so it can be applied to future projects or discovery sessions without rebuilding from scratch.
Maintain version history
When iterating on a research process, create a new guide rather than overwriting the existing one. This preserves a history of versions you can cross-reference when comparing data across different research stages.
Use consistent structure across guides
When you are designing discussion guides for related studies, try to keep the structure and core questions consistent. Maintaining a shared framework makes cross-project analysis significantly more powerful.
By standardizing your approach, you can:
Accelerate Discovery: Surface data more effectively using Ask AI and Deep Research.
Simplify Tracking: Easily compare findings across different participant groups or shifting timeframes.
Streamline Tagging: Reuse the same labels across guides to amplify these benefits.
Consistency across your guides transforms isolated user interviews into a cohesive, searchable repository of insights.
Step 6: Validate before you launch
Before launching your study, take these steps to ensure clean, actionable data collection.
Self-Testing
Preview the guide by acting as a "naive user." Take the interview yourself to test the AI’s performance, check your wording, and catch any logical loops.
Automatic Note Labeling
Your discussion guide questions automatically act as note labels. During analysis, you can instantly isolate and compare answers to the exact same question across dozens of different interview files.
Semi-Structured Flexibility
Don't worry if a participant answers a question out of order. Marvin automatically detects the context and links their response to the correct guide question.
Post-Interview Feedback
Review the automated quality feedback after a session to see where you can improve structure or add more open-ended phrasing in future iterations.
Learn more: Automatically add labels to notes
Pro Tip: Creating discussion guides for AI moderation
When running AI-moderated sessions, specific settings shape how the AI behaves: its tone, how deeply it follows up, and how it handles the conversation flow.
Define tone and persona
Instruct the AI to maintain a specific style: conversational, empathetic, neutral, or direct. Keep instructions clear and concise; the AI performs best with explicit, actionable guidance.
Set follow-up depth per question
Control follow-up depth on a per-question basis to balance insight quality with participant fatigue.
Follow-up Depth | Follow-up Questions | Best used for |
Keep it Brief | 1-2 questions | Warm-up, screening, or lower-priority topics |
Follow Up a Little | 2-3 questions | Standard questions in a typical 20-30 min session |
Follow Up More | 3-5 questions | High-priority topics where depth is critical |
Choose your interaction mode
Turn-based: Turn-based: Participants click a button when they're done speaking. Better for participants who need clear cues.
Real-time conversation: Real-time conversation: The AI detects natural pauses and responds fluidly. More natural, but works best with clear audio.
Understand slash vs. non-slash text
Text with /: Text with /: read verbatim by the AI as a question or response.
Text without /: Text without /: background context or coaching instructions for the AI (e.g., "In the next question, we're interested in knowing the participant's seniority level.")
Manage session length
Aim for AI-moderated sessions between 20 and 30 minutes. Performance may degrade beyond 25 minutes. If your study requires more time, consider splitting it into multiple shorter sessions.
Include Compliance Controls
Use advanced settings to add legal disclaimers, participant limits, privacy "read me" links, and incentive redirects.
Pro tip: When using the AI generator, provide specific audience details (job titles, key demographics, product area) for better-tailored questions and more relevant follow-up questions.
Learn more: Best practices for using AI moderation
FAQs
What if I need more than 30 questions?
If your study requires more depth, simply split your questions into two separate guides of 30 questions each. Because Marvin allows you to run multiple guides sequentially against the exact same interview file, this multi-guide strategy keeps your prompts concise and digestible, guaranteeing rapid turnaround times and comprehensive data mapping without sacrificing the scope of your research.