AI-moderated interviews require a fundamentally different approach to discussion guide design. Traditional guides script every question, probe, and transition in sequence — because a human moderator needs that roadmap. AI-moderated guides define research frameworks that the AI uses to probe adaptively — because the AI generates follow-up questions dynamically based on what each participant actually says.
This distinction matters practically. Over-scripting an AI-moderated guide turns it into a survey. Under-specifying it produces shallow, unfocused conversations. The skill is designing frameworks that give the AI enough structure to maintain research objectives while allowing enough flexibility to follow interesting threads 5-7 levels deep.
What Makes an AI-Moderated Discussion Guide Different
In human-moderated research, the guide is a script. It includes:
- Exact question wording for each topic
- Pre-written follow-up probes
- Transition language between topics
- Time allocations per section
- Moderator notes on what to listen for
This makes sense when a human is conducting the conversation. The moderator needs structure to stay on track, hit every topic, and manage the 45-60 minute session.
In AI-moderated research, the guide is a framework. It includes:
- Research objectives — what the study needs to learn
- Topic areas — 5-8 themes to explore
- Opening prompts — 1-2 questions per topic that initiate the conversation thread
- Probing directives — what kinds of depth the AI should pursue (motivations, comparisons, emotions, trade-offs)
- Methodological parameters — laddering depth, non-leading language constraints, time allocation
The AI handles everything else: generating follow-up probes, adapting to unexpected responses, managing transitions, and maintaining conversational flow. You design the “what to explore” — the AI handles the “how to explore it.”
The 5-7 Level Laddering Framework
Laddering is the core technique that separates depth interviews from surveys. It’s a structured approach to probing progressively deeper into a participant’s reasoning:
Level 1-2: Surface response (What happened)
- “I switched from Brand A to Brand B”
- “I liked the new feature”
Level 3-4: Functional and emotional drivers (Why it matters)
- “Brand B saves me time on my morning routine” (functional)
- “I felt frustrated that Brand A kept changing their formula” (emotional)
Level 5-7: Core values and identity (What it means)
- “Being efficient in the morning means I can spend that time with my kids before school” (value)
- “I’m the kind of person who does research before buying — brands that change without telling me feel disrespectful” (identity)
Most surveys stop at level 1-2. Focus groups typically reach level 3-4. Depth interviews with skilled moderators reach level 5-7. AI-moderated interviews achieve 5-7 levels consistently across every conversation because the laddering framework is built into the moderation methodology — not dependent on moderator skill or energy levels.
How the AI Ladders
When a participant says “I switched to Brand B because it’s cheaper,” a survey records that response. An AI moderator probes:
- “What made the price difference noticeable to you?” (Context)
- “How did you feel about paying more for Brand A before the switch?” (Emotion)
- “What did you expect to get for the higher price?” (Expectation)
- “When those expectations weren’t met, what specifically disappointed you?” (Specificity)
- “How did that disappointment affect your trust in Brand A overall?” (Broader impact)
- “What would Brand A need to do for you to consider coming back?” (Action threshold)
Each probe builds on the participant’s actual response — not a pre-scripted follow-up. The AI identifies which thread to pursue, when to probe deeper, and when to move on.
Designing Your Opening Questions
Opening questions set the direction for each topic area. They need to be:
Open-ended — no yes/no questions. “Tell me about…” or “Walk me through…” instead of “Did you…” or “Do you prefer…”
Non-leading — don’t suggest the answer. “How did you decide which product to buy?” instead of “What made you choose our product?”
Experiential — anchor in concrete experience rather than abstract opinion. “Think about the last time you made a purchase in this category. What happened?” instead of “What do you think about this category?”
Singular — one question at a time. Don’t combine: “Tell me about your experience and what you’d change.” Split into two separate opening prompts.
Examples of Strong Opening Questions
Brand perception research:
- “Think about the last time you heard someone mention [Brand]. What was the context?”
- “If [Brand] were a person at a dinner party, how would you describe them?”
- “Walk me through the last time you considered buying from [Brand]. What was going through your mind?”
UX discovery research:
- “Show me the last thing you tried to accomplish in [product]. Walk me through exactly what happened.”
- “Think about a time when [product] frustrated you. What were you trying to do?”
- “If you could change one thing about how [product] works, what would it be and why?”
Churn research:
- “Take me back to the moment you decided to cancel. What was happening?”
- “Before you cancelled, did anything almost make you stay? What was it?”
- “If a friend asked you about [product] today, what would you tell them?”
Writing Probe Sequences
You don’t script individual probes for AI-moderated interviews — the AI generates them dynamically. Instead, you provide probing directives that tell the AI what kinds of depth to pursue:
Motivation probes
“When the participant describes a behavior or choice, probe for the underlying motivation. What triggered the decision? What alternatives were considered? What would have changed the outcome?”
Emotional probes
“When the participant expresses positive or negative sentiment, probe for the emotional experience. How did it make them feel? What was at stake emotionally? How does this connect to their broader life or work?”
Comparison probes
“When the participant mentions alternatives, competitors, or past experiences, probe for comparison dimensions. What’s better, what’s worse, and what’s different? What would need to change for their preference to shift?”
Consequence probes
“When the participant describes a problem or unmet need, probe for downstream effects. What happened as a result? How did it affect other decisions? Who else was impacted?”
These directives give the AI a framework for generating relevant follow-up probes without scripting the exact questions. The AI applies the appropriate probe type based on what the participant actually says.
Branching Logic: How AI Adapts in Real-Time
Traditional branching logic is pre-programmed: “If participant says X, ask Y. If participant says Z, ask W.” This works for surveys but produces rigid, unnatural conversations.
AI-moderated branching is adaptive. The AI detects themes, emotions, and unexpected directions in real-time and adjusts its probing strategy:
Unexpected competitor mention: If a participant spontaneously mentions a competitor you didn’t include in the guide, the AI probes that competitive thread — because unprompted competitor mentions are often more valuable than prompted comparisons.
Strong emotional response: If a participant’s language signals frustration, excitement, or conflict, the AI shifts into emotional probing mode — exploring the experience behind the emotion rather than moving to the next topic.
Contradiction detection: If a participant says “I love the product” but later describes avoiding key features, the AI gently explores the contradiction — “Earlier you mentioned enjoying the product, but it sounds like you don’t use [feature]. Help me understand that.”
Depth vs. breadth management: If a participant is providing deep, rich responses on a topic, the AI spends more time there — even if it means covering fewer topics. Depth on 4 topics beats surface responses on 8.
Common Mistakes in AI Discussion Guide Design
1. Over-scripting
Writing 40 specific questions with exact follow-ups. This turns the AI into a survey bot. Instead: 5-8 topic areas with 1-2 opening questions each.
2. Leading questions in opening prompts
“What do you love about our new feature?” assumes they love it. Instead: “Tell me about your experience with our new feature.”
3. Front-loading too many topics
Trying to cover 12 topics in 30 minutes. No topic gets depth. Instead: 5-6 topics with full laddering. Cut topics ruthlessly.
4. Neglecting warm-up
Jumping straight into research questions. Participants need 2-3 minutes of rapport-building to feel comfortable sharing honest, deep responses. Include a warm-up prompt: “Before we dive in, tell me a bit about your role and what a typical day looks like.”
5. Designing for breadth instead of depth
“Let’s cover product, pricing, brand, competitors, support, and loyalty.” Instead: “Let’s deeply understand the purchase decision, including everything that influenced it.” Five deep topics consistently outperform fifteen shallow ones.
Voice vs. Video vs. Chat: How Guide Design Changes by Modality
The core methodology — 5-7 level laddering, non-leading language, adaptive probing — stays the same across modalities. But guide design adapts to how participants process information:
Voice interviews:
- Opening questions can be more complex and multi-part — verbal processing is natural
- Warm-up is especially important — voice creates intimacy that requires trust
- Probing can reference “what I heard you say” patterns effectively
- Best for: emotional topics, narrative-heavy research, complex B2B decisions
Chat interviews:
- Opening questions should be shorter and simpler — text requires concise prompts
- Multiple follow-ups can feel rapid-fire — pace probing with transitional language
- Participants often write more candidly in text than they speak
- Best for: sensitive topics, large-scale studies, international research in 50+ languages
Video interviews:
- Can incorporate visual stimuli — show concepts, packaging, interfaces
- Non-verbal cues add richness (though AI moderation focuses on verbal content)
- Opening questions should reference what’s being shown: “Looking at this packaging design, what’s your first impression?”
- Best for: concept testing, design research, product demonstrations
All three modalities achieve the same laddering depth. The choice depends on the research context, participant comfort, and whether visual stimuli are part of the study design.
Template: Brand Perception Study
Here’s a ready-to-use framework for a brand perception study:
Objective: Understand how target consumers perceive [Brand] relative to alternatives
Warm-up (2-3 min): Opening: “Tell me a bit about yourself and how [category] fits into your life.”
Topic 1: Category context (5 min) Opening: “When you think about [category], what brands come to mind first? Walk me through what each one means to you.” Probe directive: Ladder into why these brands occupy top-of-mind position. What experiences shaped the association?
Topic 2: Brand encounter (7 min) Opening: “Think about the last time you encountered [Brand] — whether you bought it, saw an ad, or heard someone mention it. What happened?” Probe directive: Explore the full context. Where were they? What triggered the interaction? How did they feel?
Topic 3: Brand meaning (7 min) Opening: “If [Brand] were a person, how would you describe their personality?” Probe directive: Ladder from surface descriptors to values. What kind of person uses this brand? What does choosing it say about someone?
Topic 4: Competitive comparison (5 min) Opening: “If you couldn’t buy [Brand], what would you choose instead? What would you gain and lose?” Probe directive: Explore switching barriers and drivers. What’s irreplaceable about [Brand]? What’s better elsewhere?
Topic 5: Future relationship (4 min) Opening: “Looking ahead, do you see yourself using [Brand] more, less, or about the same? What would change that?” Probe directive: Ladder into what would strengthen or weaken the relationship. What’s the brand’s biggest risk?
Template: UX Discovery Study
Objective: Identify friction points, unmet needs, and opportunities in [product] experience
Warm-up (2-3 min): Opening: “Tell me about your role and how [product] fits into your workflow.”
Topic 1: Typical usage (5 min) Opening: “Walk me through the last time you used [product]. Start from when you opened it — what were you trying to accomplish?” Probe directive: Map the complete journey. Where did things flow smoothly? Where did they slow down or confuse?
Topic 2: Friction point deep-dive (8 min) Opening: “Think about a time when [product] frustrated you. What were you trying to do, and what went wrong?” Probe directive: Ladder deeply into the consequences. What did they do instead? How much time was lost? How did it affect their work or the people depending on their output?
Topic 3: Workarounds (5 min) Opening: “Are there things you use other tools for that you wish [product] could handle? Walk me through an example.” Probe directive: Explore the gap between expectation and reality. What would the ideal experience look like?
Topic 4: Feature priorities (5 min) Opening: “If you could wave a magic wand and change one thing about [product], what would it be?” Probe directive: Ladder into why this change matters most. What would it enable that isn’t possible today?
Topic 5: Delight moments (5 min) Opening: “Has [product] ever surprised you in a good way? Tell me about that moment.” Probe directive: Understand what creates positive experiences. What made it memorable? How does it affect their overall perception?
Iterating Your Guide Across Study Waves
Discussion guides improve with use. Here’s how to iterate:
Wave 1 (20-30 participants): Run the initial guide. Review 10-15 full conversation transcripts. Note where the AI produced deep, insightful probing and where conversations stayed surface-level.
Between waves: Refine opening questions that produced shallow responses. Add probing directives for themes that emerged unexpectedly. Remove topics that didn’t generate useful depth.
Wave 2 (adjust sample): Run the refined guide. Compare conversation depth and insight quality to Wave 1. The improvement is typically dramatic — because you’re optimizing the framework the AI uses to probe.
Ongoing: Every study refines your understanding of what produces depth with your specific audience. Guides for recurring research (quarterly brand tracking, ongoing UX research) improve continuously. And every conversation feeds into the intelligence hub, building compounding knowledge across waves.
The best discussion guides aren’t written once — they’re evolved through iterative learning about what questions, in what sequence, with what probing frameworks, produce the deepest understanding of your customers.
Ready to design your first AI-moderated study? Start on the platform or explore our complete guide to AI-moderated interviews for methodology deep-dives.