Video summary

The Only NotebookLM 2.0 Tutorial You'll Ever Need

Main summary

Key takeaways

Technology

Summary of the tutorial (NotebookLM 2.0)

  • Biggest change: Google rebuilt the core engine behind NotebookLM. The UI largely looks the same, but the underlying system can now do real work, not just summarize sources.
  • Model upgrade: NotebookLM 2.0 runs on Gemini 3.5, improving:
    • Accuracy
    • Consistency across long, messy notebooks

Agentic system (“antigravity”)

NotebookLM adds an agentic framework called antigravity, meaning the AI can take actions and use tools rather than only generating text.

This agentic capability lets NotebookLM:

  • Open/run things
  • Process tasks step-by-step
  • Perform work itself instead of only answering questions

Secure cloud computer per notebook

Each notebook includes a secure cloud “sandbox” computer attached to it. The notebook can run code and calculations on your behalf with 100+ built-in skills.

Practical outcome: NotebookLM can stop guessing and instead compute and generate real files from your sources/data.

What’s new in outputs: generating real editable/downloadable files

A major visible upgrade appears when you ask for a file.

  • Old behavior (described): could generate items like slide decks/images and tables, but with limitations (e.g., non-editable slides; tables exporting only to Google Sheets).
  • New behavior:
    • You can request PowerPoint, Excel, PDF, charts, and structured data like CSV
    • NotebookLM builds the real working documents
    • Downloads-only: files download to your computer rather than opening inside NotebookLM
    • Generated content is structured sensibly (e.g., an investor pitch deck includes a title slide, problem, numbers, closing ask)

File-generation examples described

  • Investor pitch deck from research (returns a real .pptx-style deck)
  • Excel spreadsheet for cost/revenue projections (live/editable cells in Excel/Sheets)
  • One-page PDF summary including tables/structured content
  • Word document option and CSV for clean data transfer to other tools

How NotebookLM finds and imports sources (gap-filling research)

The tutorial emphasizes that NotebookLM can now help with sourcing more intelligently.

Instead of blindly searching the web, the workflow:

  • Reads what’s already in your notebook
  • Identifies missing “gaps”
  • Searches for sources that specifically fill those gaps

It can also search for opposing views (counterarguments), not just supporting material.

No source is added automatically—you review and approve, then import.

Example “missing sources” prompts described

  • Find gaps like TAM/industry growth, valuation multiples, exit strategies, operating expenses
  • Find opposing arguments to test your assumptions

Using NotebookLM for analysis on your own data (not just other people’s content)

A key section explains that NotebookLM can work with your raw files, not only articles/PDFs/videos.

You can upload data (e.g., ad spend, newsletter sign-ups) and ask questions like:

  • Compare datasets
  • Determine which channels convert

Because it can run code:

  • It can handle messy/mismatched formatting
  • It can align different data shapes
  • It performs actual calculations (not visual/eyeball reasoning)

It can generate:

  • Charts (e.g., spend vs. sign-ups by channel)
  • Narrative explanations of what the chart means (e.g., losing vs. winning channels)

Example: “reporting from the same chat”

After generating a chart and analysis, you can ask for a short PDF report with the chart embedded—so analysis + final document come from one chat.

Guidance on getting better results (habits/workflow)

The tutorial recommends several practical usage habits:

  1. Be specific in prompts
    • “Make a deck” vs. “Make an investor pitch deck” changes structure and tone.
  2. Use one notebook per project
    • Avoid mixing unrelated tasks; focused notebooks ground the work better.
  3. Let NotebookLM do research before you narrow
    • Start with a rough idea → let it suggest gaps → then choose what to import.
  4. Chain features in one continuous chat
    • Research → analyze → build the document/file sequentially, so each step feeds the next.

Trust and rollout notes

The speaker addresses trust concerns:

  • Even though NotebookLM can search for new sources, it only uses them when you approve imports.
  • When chatting, it still works from what’s in your notebook to preserve grounding.

Rollout mentioned:

  • AI Ultra plan first
  • Pro plan later (cheaper tier)

Main speakers / sources

  • Main speaker: The video creator/host (“I’ll take you through everything that’s new…”) from the tutorial.
  • Primary product source: Google NotebookLM / Gemini (Gemini 3.5) and Google’s described antigravity agentic system.

Original video