Table of Contents
Quick Answer
AI turns user research from a 3-week project into a 3-day sprint. Use it for discussion guide design, live transcription, theme synthesis, and affinity mapping — while you focus on building rapport and asking good follow-ups.
- Design: AI drafts a non-leading discussion guide in 10 minutes
- Capture: Otter.ai or Fireflies transcribe automatically
- Synthesize: Dovetail or custom GPT clusters themes from 12 interviews in 1 hour
What You'll Need
- Interview goal (one clear research question)
- 8-12 participants matching criteria
- Zoom or Google Meet
- Otter.ai, Fireflies.ai, or Grain for transcription
- Dovetail, Notably, or Claude for synthesis
Steps
- Write the research question. Example: "Why do trial users drop off before activation?" One sentence. AI can't help if this is fuzzy.
- Draft the discussion guide with AI. Prompt: "Write a 30-minute user interview guide for [research question]. Include 5 open-ended questions, no leading language, Jobs-to-be-Done framing, and follow-up probes."
- Recruit participants. User Interviews.com, Respondent, or your own user base. Offer $75-150 for 30 min.
- Run interviews. Always ask "tell me about the last time you…" — specific memory beats hypothetical.
- Transcribe automatically. Otter.ai or Fireflies runs in the background.
- Synthesize with AI. Paste 12 transcripts into Claude 3.5 (200K context) or Dovetail. Prompt: "Identify top 10 themes, frequency, and supporting quotes with speaker IDs."
- Build the insight doc. Top 5 themes + 3 verbatim quotes each + recommended actions.
Discussion Guide Template
Intro (2 min): Thank you, record consent, explain this is not a sales call.
Context (5 min):
- Tell me about your role.
- Walk me through a typical day.
Problem exploration (15 min):
- Tell me about the last time you tried to [job to be done].
- What did you do? What was hard?
- What did you try before? Why didn't that work?
Solution reactions (5 min):
- [Only if testing a concept — show and ask open questions]
Wrap (3 min):
- What didn't I ask that I should have?
Common Mistakes
- Leading questions ("Would you pay for a tool that did X?") — always yes, meaningless
- Too few interviews (3-4) — patterns emerge at 7-12
- Jumping to solutions in the interview — stay in problem-space
- AI-only synthesis — miss nuance; always sanity-check
- No written insight doc — research evaporates
Top Tools
Tool
Best For
Pricing
Dovetail
End-to-end research repo
$39/user/mo
Otter.ai
Live transcription
$17/mo
Notably
AI research synthesis
$24/user/mo
User Interviews
Participant recruiting
$45/participant
Claude 3.5 (200K context)
Custom synthesis
$20/mo
FAQs
How many interviews for solid signal? 7-12 for B2B; 15-20 for consumer. Themes saturate fast.
Can AI interview people directly? Yes — Outset.ai and User Interviews now offer AI moderators. Good for quant screening, not nuanced discovery.
Should I record without consent? Never. Always get explicit consent on camera at the start.
What about bias in AI synthesis? Real. Always read 2-3 transcripts yourself to ground-truth AI themes.
How do I share insights? Short Loom video + 1-page doc + Slack post. Long reports die unread.
Can AI replace researchers? No — it accelerates them 5x. The skill is question design and listening.
What about international / non-English interviews? Otter and Fireflies now support 30+ languages. Claude translates and synthesizes natively.
Conclusion + CTA
User research is the highest-leverage activity in product development — and AI just made it 5x faster. No more "we don't have time for research" excuses.
Book 3 customer calls this week. Use the discussion guide above. Paste transcripts into Claude Friday afternoon. Ship insights Monday.