razbakov/skills/youtube-notebooklm-research

YouTube NotebookLM Research

Search YouTube for videos on a topic, add them to a NotebookLM notebook, and ask questions about the content.

Trigger

When the user asks to research a topic using YouTube videos, find videos and add to NotebookLM, or learn about something from YouTube content.

Prerequisites

• notebooklm CLI installed (pipx install "notebooklm-py[browser]")
• yt-dlp installed (brew install yt-dlp)
• Authenticated: notebooklm login (one-time browser auth)
• Chromium for playwright: pipx run --spec "notebooklm-py[browser]" playwright install chromium

Process

Step 1: Search YouTube

Use yt-dlp to find relevant videos on the topic:

yt-dlp "ytsearch${COUNT}:${TOPIC}" --flat-playlist --print "%(id)s | %(title)s | %(duration)s | %(channel)s | %(view_count)s" 2>/dev/null

• Default COUNT: 5 (adjustable by user)
• Filter out videos shorter than 2 minutes (likely shorts/clips)
• Filter out videos longer than 120 minutes unless user wants deep dives
• Prefer videos with higher view counts when relevance is similar

Step 2: Present results and confirm

Show the user the found videos with title, channel, duration, and view count. Ask which ones to add (default: all).

Step 3: Create or select NotebookLM notebook

# Create a new notebook for the topic
notebooklm create "Research: ${TOPIC}"

# Or list existing to reuse
notebooklm list
notebooklm use <notebook_id>

Step 4: Add YouTube videos as sources

# Add each selected video
notebooklm source add "https://www.youtube.com/watch?v=${VIDEO_ID}"

Wait for each source to be indexed before adding the next. Check with:

notebooklm source list

Step 5: Ready to query

# Ask questions about the content
notebooklm ask "What are the key points across all videos?"
notebooklm ask "${USER_QUESTION}"

Step 6: Optional - Generate artifacts

# Generate a podcast-style audio summary
notebooklm generate audio "Focus on practical takeaways" --wait

# Generate a study guide
notebooklm generate report --style study-guide --wait

# Generate quiz
notebooklm generate quiz --difficulty medium

# Download artifacts
notebooklm download audio ./research-${TOPIC}.mp3

Output

• NotebookLM notebook with YouTube videos as sources
• Answers to user's questions grounded in video content
• Optional: audio summary, study guide, quiz

Notes

• NotebookLM uses Gemini to process video transcripts — answers are grounded in actual content, not hallucinated
• The add_youtube method extracts transcripts automatically, no need for manual transcript download
• Auth token stored at ~/.notebooklm/storage_state.json — refresh with notebooklm login if expired
• For batch research across multiple topics, create separate notebooks per topic