NotebookLM & Invideo Skill
What NotebookLM Does
NotebookLM is a research and content synthesis tool powered by Google. It can:
- Audio overviews -- generate podcast-style conversations between two AI hosts that discuss your uploaded sources in a natural, engaging format
- Research synthesis -- combine multiple sources into a unified understanding, surfacing connections and key themes
- Q&A over sources -- ask questions grounded in your uploaded documents and get cited answers
- Study guides -- auto-generate structured study materials from academic or technical content
- FAQ generation -- extract common questions and answers from source material
- Timelines and briefing docs -- produce chronological summaries or executive-level briefings
What It Does NOT Do
- Cinematic video production -- for video ads, explainers, or social content, use Invideo (see below)
- Real-time data -- NotebookLM works only with uploaded/linked sources, not live feeds
- Interactive content -- no quizzes, polls, or clickable outputs
- Code execution -- it cannot run code or generate working software artifacts
- Private/sensitive processing -- sources are sent to Google servers; do not upload credentials, PII, or confidential business data
Source Preparation
NotebookLM accepts up to 50 sources per notebook. Supported formats:
- PDF documents
- Google Docs (linked directly)
- Websites / URLs
- Pasted text (plain or markdown)
Tips for best results:
- Keep all sources focused on one topic per notebook -- mixing unrelated subjects dilutes quality
- 3-7 sources is the optimal range for audio overviews -- enough diversity without overwhelming
- Quality over quantity -- one well-written 20-page report beats ten shallow blog posts
- Remove boilerplate, headers/footers, and navigation text from PDFs before uploading
- For websites, prefer article pages over homepages or index pages
Audio Overview Tips
Audio overviews work best with:
- Factual and educational content -- research papers, technical docs, reports, how-to guides
- Diverse source types -- mixing a PDF report with a blog post and a data table produces richer conversation than three similar PDFs
- Specific instructions -- tell it the target audience, desired tone (casual, academic, journalistic), and which aspects to emphasize
- Moderate complexity -- topics that benefit from explanation and discussion, not simple facts
Audio overviews are weaker for:
- Highly subjective or opinion-based content
- Very short source material (not enough to discuss)
- Content requiring visual explanation (charts, diagrams, code walkthroughs)
Video Content (Invideo)
For video ads, explainers, social clips, and other video content, use the Invideo tool:
mcp__claude_ai_Invideo__generate-video-from-script
Writing effective Invideo prompts:
- Duration -- specify target length (e.g., "30-second ad", "2-minute explainer")
- Visual style -- describe the aesthetic (minimalist, bold, corporate, playful, cinematic)
- Text overlays -- write exact text for titles, subtitles, and call-to-action screens
- Color palette -- specify brand colors or mood-based palette (warm tones, dark mode, neon accents)
- Soundtrack mood -- describe the audio feel (upbeat electronic, calm ambient, dramatic orchestral)
- Pacing -- fast cuts for energy, slow transitions for authority, mixed for storytelling
- Structure -- break the script into scenes with visual directions for each
The more specific the prompt, the better the output. Vague prompts produce generic results.
Output Formats
| Format | Tool | Best For |
|---|
| Audio overview (podcast) | NotebookLM | Educational synthesis, research digests, team updates |
| FAQ document | NotebookLM | Knowledge base articles, onboarding docs |
| Study guide | NotebookLM | Training material, exam prep, technical learning |
| Timeline | NotebookLM | Project history, event sequences, changelog narratives |
| Briefing doc | NotebookLM | Executive summaries, stakeholder updates |
| Video (ad, explainer, social) | Invideo | Marketing, social media, product demos |
When NOT to Use This Skill
- Simple factual questions -- just search the web or ask directly; no need to create a notebook
- Code generation -- use the bot's native coding capabilities
- Real-time data -- use web search or API integrations
- Private/sensitive content -- sources are processed on Google/Invideo servers
- Quick one-off answers -- NotebookLM is for multi-source synthesis, not single-question lookups