# Podcast Transcript Extraction Prompt
You are extracting structured information from a Lenny Rachitsky podcast transcript. Be EXHAUSTIVE - these transcripts are gold mines of product wisdom. Missing content is worse than over-extracting.
## Output Format
Return ONLY valid JSON with this schema:
```json
{
"episode": {
"guest": "string",
"expertise_tags": ["string"],
"summary": "string (150-200 words)",
"key_frameworks": ["string"]
},
"topics": [{ "id", "title", "summary", "timestamp_start", "timestamp_end", "line_start", "line_end" }],
"insights": [{ "id", "text", "context", "topic_id", "line_start", "line_end" }],
"examples": [{ "id", "explicit_text", "inferred_identity", "confidence", "tags", "lesson", "topic_id", "line_start", "line_end" }]
}
```
## MINIMUM REQUIREMENTS (you MUST meet these)
| Element | Minimum Count | Notes |
|---------|---------------|-------|
| Topics | 10-20 | Every distinct conversation segment gets a topic |
| Insights | 15-30 total | 2-4 per topic, more for rich topics |
| Examples | 10-25 total | Every story, anecdote, case study, company mention |
If your output has fewer than these minimums, you have under-extracted. Go back and find more.
## What to Extract
### Topics
A new topic starts when:
- The conversation shifts to a new subject
- Lenny asks a new question on a different theme
- The guest transitions with "Another thing..." or "Let me tell you about..."
Topic titles should be searchable: "How Airbnb eliminated traditional PMs" not "Discussion about teams"
### Insights (BE THOROUGH)
Extract EVERY piece of wisdom including:
- Contrarian takes ("Most people think X, but actually Y")
- Tactical advice ("The way to do X is...")
- Frameworks and mental models
- Lessons learned from failures
- Non-obvious observations
- Quotable statements
BAD insight: "Product management is important"
GOOD insight: "Engineers prioritize projects that give them visibility. Create internal newsletters celebrating their contributions to earn discretionary effort without formal authority."
### Examples (CRITICAL - extract ALL of them)
**Type 1: Explicit stories** - Guest names the company/person
- "At Airbnb, we did X..."
- "When I was at Facebook..."
- "Companies like Stripe do X..."
**Type 2: Implicit stories** - Guest doesn't name but you can infer
- "At my previous company..." → Use guest's LinkedIn to infer (e.g., Brian Chesky = Airbnb)
- "One marketplace I know..." → If guest worked at Uber, likely Uber
- "A famous social network..." → Facebook, Twitter, Instagram based on context
**Type 3: Anecdotes** - Personal stories that illustrate a point
- "I remember when Herbie Hancock called me..."
- "We stayed in the office for 22 days straight..."
For EACH example, ask: "What would someone search to find this?" and put those terms in tags.
### Tags for Examples (5-10 per example)
Include:
- Company name (explicit or inferred)
- Industry (marketplace, SaaS, consumer, B2B)
- Problem being solved (retention, growth, pricing, hiring)
- Tactic used (A/B test, reorg, launch strategy)
- Role (PM, founder, executive)
- Outcome keywords (10x growth, saved company, failed)
## Line Numbers
- Line numbers are 1-indexed
- Topic line ranges should NOT overlap
- Topics should cover the ENTIRE transcript (no gaps)
- Example/insight line numbers should fall within their parent topic
## Quality Checklist Before Returning
1. Do I have at least 10 topics? If not, find more conversation segments.
2. Do I have at least 15 insights? If not, re-read for tactical advice and contrarian takes.
3. Do I have at least 10 examples? If not, look for every story, anecdote, and company mention.
4. Did I capture implicit examples? Check for "at my previous company" type phrases.
5. Are my tags searchable? Would someone find this example by searching those terms?
6. Are my lessons specific? Not "shows importance of X" but "demonstrates how to achieve X by doing Y"
## Transcript to Process
The following transcript has line numbers. Extract comprehensively:
