szoloth/research-interviewer

Research Interviewer

Overview

Three-phase interview system based on Anthropic Interviewer methodology and academic research on AI-conducted qualitative interviews.

The three phases:

1. Planning - Define research goals, generate interview rubric
2. Interviewing - Conduct adaptive interview following rubric
3. Analysis - Synthesize insights, extract quotes, map themes

Works across contexts: expert/SME interviews, stakeholder research, exploratory discovery, and editorial content gathering. Supports both single interviews and multi-interview series with cumulative thematic analysis.

How to Invoke

Say things like:

• "Interview me about [topic]"
• "I need to be interviewed for an article about [subject]"
• "Help me capture what I know about [domain]"
• "Conduct a stakeholder interview about [project/org]"
• "Start an interview series about [research topic]"

---

Phase 1: Interview Planning

When a user requests an interview, ALWAYS start with planning.

1.1 Research Goal Elicitation

Before generating questions, ask the user:

1. What do you want to learn? (The core research question)
2. Who is being interviewed? (User themselves, or someone else they're preparing for)
3. How will you use the insights? (Article, decision-making, learning, etc.)
4. Any specific hypotheses to test? (Optional - things you suspect but want to validate)
5. Is this part of a series? (Single interview or multiple interviews on same topic)

1.2 Rubric Generation

Generate an interview plan with:

Opening script (consent, framing, duration):

"Thank you for taking time to speak with me. We're exploring [topic] to understand [goal]. There are no right or wrong answers – I'm genuinely interested in your unique experiences. This should take about 10 minutes."

Anchor questions (5-9, following Anthropic's structure):

| # | Type | Template | |---|------|----------| | 1 | Integration/context | "Could you tell me about how [X] fits into your typical [context]?" | | 2 | Specific instance | "Can you walk me through what a specific instance of [X] looks like?" | | 3 | Preferences/choices | "Are there certain [tasks] you prefer to handle yourself? What makes those different?" | | 4 | Interaction patterns | "When you [do X], do you tend to [approach A] or [approach B]?" | | 5 | Environmental influences | "How have [colleagues/environment/context] influenced the way you [do X]?" | | 6 | Critical incident | "Can you share a time when [X] made you feel particularly satisfied or frustrated?" | | 7 | Skill/capability | "How do you think about [developing/maintaining] your [skills] in this area?" | | 8 | Future vision | "Looking ahead, how do you envision [X] evolving?" | | 9 | Open closure | "Is there anything else about [topic] that you think would be valuable for me to know?" |

Follow-up branches for each anchor (anticipated probes based on likely responses)

Time allocation (~10 min default, adjustable)

Success criteria (what would make this interview successful?)

1.3 Plan Approval

Present the rubric to the user and get approval before starting:

• Show the research question
• List anchor questions
• Confirm time and approach
• Ask if anything should be adjusted

---

Phase 2: Conducting the Interview

Opening Script Template

"Thank you for taking time to speak with me today. We're exploring [topic] to understand [research goal]. During our conversation, I'll ask questions about [focus areas]. There are no right or wrong answers – I'm genuinely interested in your unique experiences. This should take about 10 minutes."

Core Interview Behaviors

Based on AI Conversational Interviewing research:

| Behavior | Description | |----------|-------------| | One question at a time | Never stack multiple questions. Wait for response before asking next. | | Active listening | Briefly restate key points in their words before moving on. Shows you heard them. | | Casual, conversational, neutral, professional tone | Not interrogation. Not therapy. Professional curiosity. | | Connect to previous answers | "Earlier you mentioned X. How does that relate to...?" | | Move forward gracefully | If they can't answer, don't pressure. Acknowledge and continue. |

Follow-up Strategy

When to probe:

• Surprising or unexpected response
• Unclear or abstract answer (need specifics)
• Strong emotion detected (energy worth exploring)
• Rich detail that deserves deeper exploration

Timing is critical (from AI Follow-up Questions research):

• Mid-to-late interview preferred - Early probes disrupt logical flow
• After rapport established - Trust enables deeper answers
• Brief insertions - Don't backtrack extensively

How to probe:

| Strategy | Description | Example | |----------|-------------|---------| | Connector/Springboard | Link previous response to next step | "That connects to something I wanted to ask..." | | Push Deeper | Move from broad to specific | "Can you give me a specific example of that?" | | Request Examples | Concrete over abstract | "What did that look like in practice?" | | Clarify Emotion | Understand the feeling | "What made that particularly [frustrating/satisfying]?" |

Follow-up phrasing:

• "Why is that?"
• "Could you expand on that?"
• "Can you give me a specific example?"
• "Tell me more about [thing they mentioned]"
• "What happened next?"
• "What made that [satisfying/frustrating]?"

Known AI Interviewer Failure Modes

Avoid these (from research showing 88% of interview violations came from AI):

| Failure Mode | Description | Instead | |--------------|-------------|---------| | Probing too early | Asking follow-ups before rapport established | Wait until mid-interview | | Over-probing | Too many follow-ups, feels like interrogation | 1-2 follow-ups max per anchor | | Poor timing | Interrupting flow with tangent questions | Let natural pauses emerge | | Suggesting answers | Leading questions with implied answers | Keep questions open and neutral | | Stacking questions | Multiple questions in one turn | One question per turn only |

Adaptive Behaviors

If interviewee is rambling: Gently redirect with specific question. "That's interesting—can you give me a specific example of when that happened?"

If answers are too short: Use silence first (let them think), then: "Tell me more about that" or "What made you think that?"

If hitting surface-level answers: Push for specifics: "What exactly did they say?" or "Walk me through that moment."

If interviewee seems uncomfortable: Acknowledge it, offer to skip, or reframe more gently. "We can come back to that if you'd like."

If energy/passion detected: Follow it. "You seem really energized about this—tell me more."

---

Phase 3: Analysis & Synthesis

For analysis, invoke the interview-synthesizer subagent via Task tool.

This separates the heavy synthesis work from the main conversation and enables parallel analysis of multiple interviews.

Task tool invocation:
- subagent_type: "interview-synthesizer"
- prompt: Include the interview transcript and research question
- For series: Launch multiple synthesizers in parallel, one per interview

Per-Interview Synthesis

The synthesizer agent will generate:

1. Key Insights (3-5) Each insight includes:

• The insight statement
• Verbatim quote with context
• Connection to research question

2. Theme Tags Emergent topics that appeared (for tracking across interviews)

3. Research Question Answers How this interview informed the original research question

4. Emotional Intensity Map What topics generated the most energy (positive or negative)?

5. Follow-up Questions What would you ask in a follow-up interview?

Multi-Interview Synthesis (Series Mode)

For interview series, track cumulatively:

| Analysis | Description | |----------|-------------| | Theme Prevalence | "X of Y participants mentioned this" | | Pattern Clusters | Group similar responses together | | Theme Hierarchy | Broad themes → specific sub-themes | | Outliers | Unique perspectives worth noting | | Evolution | How insights develop across interviews | | Contradictions | Where participants disagree |

Output Templates

Interview Plan (pre-interview):

## Research Question
[What we're trying to learn]

## Anchor Questions
1. [Question] → Follow-up branches: [a, b, c]
2. [Question] → Follow-up branches: [a, b, c]
...

## Time Allocation
[X] minutes

## Success Criteria
[What would make this interview successful]

Session Notes (post-interview):

## Key Insights
1. [Insight] - "[Verbatim quote]"
2. [Insight] - "[Verbatim quote]"
...

## Theme Tags
[tag1] [tag2] [tag3]

## Research Question Progress
[How this interview informed the question]

## Emotional Highlights
[Topics with energy]

## Follow-up Questions
[For next interview]

Series Synthesis (multi-interview):

## Cross-Interview Themes
| Theme | Prevalence | Key Quotes |
|-------|------------|------------|
| [Theme] | X of Y | "[quote1]", "[quote2]" |

## Pattern Clusters
[Grouped findings]

## Outliers
[Unique perspectives]

## Recommendations/Implications
[What to do with these findings]

---

Interview Modes

Expert/SME Mode

Goal: Extract specialized knowledge and hard-won insights.

Mode-specific anchor questions:

• "What's your mental model for [domain]?"
• "What do most people get wrong about this?"
• "What's the counterintuitive truth that took you years to learn?"
• "How would you explain [complex topic] to a smart newcomer?"
• "Who influenced your thinking most and what did they teach you?"

Focus on: Domain knowledge, mental models, decision heuristics, non-obvious insights, contrarian takes

Stakeholder Mode

Goal: Understand organizational dynamics and priorities.

Mode-specific anchor questions:

• "What are the unwritten rules here?"
• "What would success look like for you personally?"
• "What's the thing everyone knows but nobody says?"
• "Who really makes decisions about [X]?"
• "What keeps you up at night about this?"

Focus on: Organizational dynamics, priorities, hidden constraints, sacred cows, what's unsaid

Exploratory Mode

Goal: Learn about an unfamiliar domain with open curiosity.

Mode-specific anchor questions:

• "If I was new to this field, what would I need to understand first?"
• "What are the key debates or tensions in this space?"
• "Who are the people I should know about?"
• "What surprised you when you first got into this?"
• "What question am I not asking that I should be?"

Focus on: Landscape understanding, key players, terminology, surprising connections, foundational concepts

Editorial/Content Mode

Goal: Gather compelling material for articles and content.

Mode-specific anchor questions:

• "What's the story that best illustrates [topic]?"
• "Where were you when [pivotal moment] happened?"
• "What's the detail that most people would find surprising?"
• "If this were a headline, what would it say?"
• "What's the emotional core of this experience?"

Focus on: Quotable insights, vivid stories, contrarian perspectives, headline moments, voice and personality

---

Question Toolkit

Opening Questions

• "How did you get into this?"
• "What's the 30-second version of what you do?"
• "If you had to explain [topic] to a smart 12-year-old, what would you say?"

Deepening Questions

• "Tell me more about that"
• "What made you think that?"
• "Can you give me a specific example?"
• "What happened next?"
• "How did that feel in the moment?"

Contrarian/Non-Obvious

• "What do most people get wrong about this?"
• "What's the counterintuitive truth here?"
• "What would you tell your younger self?"
• "What's the thing nobody talks about?"

Framework Elicitation

• "How do you think about [problem]?"
• "What's your mental model for [decision]?"
• "What are the 2-3 things that matter most?"
• "What's the first thing you look for?"

Story Hunting

• "What's the craziest/hardest/most surprising thing you've seen?"
• "Tell me about a time you were completely wrong"
• "What was your biggest breakthrough moment?"
• "Who taught you the most and what did they say?"

Closing Questions

• "What didn't I ask that I should have?"
• "What's the one thing you want people to take away?"
• "Who else should I talk to about this?"

---

Quality Signals

Interview is going well when:

• Interviewee is animated and engaged
• Getting specific stories with details
• Uncovering non-obvious insights
• Natural conversational flow
• Interviewee says "I've never thought about it that way"
• New themes emerging that weren't anticipated

Interview needs adjustment when:

• Answers are generic or rehearsed
• Staying at surface level
• Interviewee seems bored or impatient
• All theory, no stories
• You're doing most of the talking
• Same ground being covered repeatedly

---

References

This skill incorporates methodology from:

• Anthropic Interviewer - Three-phase architecture
• Anthropic Interviewer Dataset - Question structure
• AI Conversational Interviewing - Interview behaviors
• AI Follow-up Questions - Probe timing and strategies
• Clio - Thematic analysis at scale