How to use NotebookLM for product research (Full workflow + Prompts) | Episode 22
AI research tools become powerful when you know exactly how to prompt them.
In this episode of Human × Intelligent, Madalena Costa walks through the workflow she uses to analyse user research with NotebookLM, turning raw interview transcripts into structured insights, personas, empathy maps and a final research report.
Instead of asking AI generic questions about your data, this workflow uses a sequence of 16 purposeful prompts to unlock the full potential of NotebookLM as a qualitative research engine. The working example throughout is a Spotify UX research study: 8 participants, 60-minute sessions, covering personalisation, discovery and what's frustrating real users.
For UX researchers, product managers and AI-curious builders, this episode shows how AI can move from surface-level summaries to deep, structured research synthesis, inside your actual documents.
In this episode, you'll learn:
Why prompting strategy matters more than the tool itself
How to ground NotebookLM in your data before any analysis begins
How to build user personas directly from interview transcripts
How to create empathy maps grounded in real participant language
How to apply the Jobs to Be Done framework with AI
How to generate and prioritise How Might We questions
How to map research to a feature opportunity matrix
How to build a 3-level insight hierarchy from affinity clusters
How to generate an executive summary, full report and stakeholder deck outline
I also share what the AI does well, where human judgment still matters and how to know when this workflow creates real leverage.
This episode is part of Human × Intelligent, a series exploring how humans and intelligent systems collaborate across product design, AI workflows and intelligent tooling.
Chapters
00:00 Introduction - why this workflow matters
02:00 What is NotebookLM and why it changes research
06:00 Step 1 - grounding the data
09:00 Step 2 - building user personas
13:00 Step 3 - creating empathy maps
16:30 Step 4 - Jobs to Be Done framework
19:30 Step 5 - How Might We questions
22:30 Steps 6 & 7 - opportunity mapping and affinity clustering
25:30 Step 8 - generating the final report
28:30 What AI does well and where judgment still matters
Example workflows mentioned
Data grounding: Summarise participant demographics, study goals and surface clusters of frustrations mentioned by 2+ participants before any analysis begins
Persona generation: Identify archetypes, build full persona profiles and map tensions between user needs
Empathy mapping: Create SAYS/THINKS/DOES/FEELS maps per persona grounded in actual participant quotes
Jobs to Be Done: Extract functional, emotional and social jobs. Then rank those that are most underserved
How Might We: Generate 15+ opportunity questions across categories, then prioritise by frequency, intensity and strategic fit
Affinity clustering: Group all observations into a 3-level hierarchy, observation → insight → opportunity
Report generation: Produce an executive summary, full UX research deliverable and stakeholder presentation outline
Free resources
→ Full prompt guide (all 16 prompts, PDF): https://docs.google.com/document/d/1q...
→ NotebookLM: notebooklm.google.com
Links
Episode page: https://humanxintelligent.com/episode...
LinkedIn: / madalenafigueirasdacosta
LinkedIn: / human-x-intelligent
Instagram: / designwithmaddie
Instagram: / humanxintelligent
Subscribe to the newsletter: https://substack.com/@humanxintelligent
🎙️ Human × Intelligent explores how humans and intelligent systems evolve together across product design, behaviour and technology.
Subscribe for more episodes on:
AI research workflows
UX and product design
Agentic systems
Human-AI collaboration
Future of product development