Episode
How to Do AI-Powered Discovery (Step-by-Step with Live Demo) | Caitlin Sullivan
- Podcast
- The Growth Podcast
- Published
- Feb 13, 2026
- Duration seconds
- 4360
- Processing state
not_requested- Canonical source
- https://www.news.aakashg.com/p/caitlin-sullivan-podcast
Actions
POST https://stenobird.com/v1/public/podcasts/the-growth-podcast-6993153/episodes/how-to-do-ai-powered-discovery-step-by-step-with-live-demo-caitlin-sullivan/transcription-requests
Idempotently request low-priority transcript generation for this episode.GET https://stenobird.com/podcast/the-growth-podcast-6993153/how-to-do-ai-powered-discovery-step-by-step-with-live-demo-caitlin-sullivan.md
Read the agent-friendly Markdown representation of this episode resource.
Summary
Today’s Episode Discovery might be the most important core PM skill for building great products. But most PMs are unprepared to do discovery in AI . PMs run surveys incorrectly, conduct interviews poorly, and end up with poor insights. Today will give you the roadmap to avoid all those mistakes . Caitlin Sullivan is a user research expert who runs courses teaching PMs how to do AI-powered discovery. And in today’s episode, she shows you exactly how she does it . We’re talking live demos. Step-by-step workflows. Real survey data. Real interview transcripts. This is a masterclass in discovery. The kind that moves the needle. ---- Brought to you by: Maven : Get 15% off Caitlin’s courses with code AAKASHxMAVEN Pendo : The #1 software experience management platform Jira Product Discovery : Plan with purpose, ship with confidence Kameleoon : AI experimentation platform Amplitude : The market-leader in product analytics ---- Key Takeaways: 1. Replicate the human process - Good AI analysis mirrors how experienced researchers work: comb through data first, then synthesize. Never jump straight to "give me themes."2. Use multi-step prompting - Load context in one prompt, run per-participant analysis in the next, then verify. Cramming everything into one prompt degrades quality.3. Code before you count - For surveys, apply inductive coding labels to every response before asking for patterns. Skipping this step leads to miscategorized, unreliable results.4. Always audit AI's work - Force the model to re-check its own analysis. It catches contradictions, overexaggerated intensity ratings, and miscoded responses regularly.5. Claude wins on nuance, Gemini wins on frequency - Claude gives more thorough, complete analysis by default. Gemini surfaces top-frequency themes faster but misse…