3 Hidden NotebookLM Features Most People Don’t Use

Build AI apps from YouTube videos, automate research, and create assistants with permanent memory.

NotebookLM is one of those tools I can’t stop writing about.

I’ve covered it twice — NotebookLM Got Crazy Powerful: Here's How I Used It to Learn Something Really Hard and How I Learned Complex Topics 10x Faster with NotebookLM — both became my highest-performing pieces. Then I went even deeper and built a guide showing how to connect NotebookLM directly to Claude using MCP, so I could run research and create content without ever switching tabs.

That guide focused on the Claude side of the equation. How to make Claude your command center while NotebookLM handles the research depth.

But there’s a whole other side I hadn’t explored: what happens when you connect NotebookLM to Google’s own ecosystem.

Gemini Canvas. Gems. Antigravity.

Three connections that turn NotebookLM into something very different from a research tool — an app builder, a permanent AI assistant, and an automated research system.

Today’s guest post is from Gencay, who writes practical AI tutorials for people who’d rather replace than be replaced by AI.

If you want to dig deeper into what he writes about, check out these three latest posts:

1. How to use NotebookLM better than 99% of People?

2. Claude Code MasterClass For Everyone (Part 3): How to Use Claude Code Plugins Better Than 99% of People

3. How I Hired A Free AI Employee (Clawdbot Setup Guide)

Where my guide connected NotebookLM outward to Claude, Gencay maps the connections that stay inside Google’s world — and some of them are surprising. The Antigravity + MCP setup alone opens up 32 different NotebookLM functions you can trigger programmatically.

Here’s Gencay.

---

Hello 👋

NotebookLM was supposed to be simple.

Upload documents. Get summaries. Generate podcasts.

Useful, but limited.

Then Google changed the game. Not once. Not twice. Three times.

They connected NotebookLM to:

• Gemini Canvas

• Gemini Gems

• Antigravity

Same NotebookLM, three hidden connections.

Before getting to discover them, let’s start with the fundamentals.

What Is NotebookLM?

NotebookLM is Google’s AI tool for learning from any source.

Or doing research.

You can upload sources like:

• YT Videos

• PDFs

• Google Docs

• Websites

• Text

• Deep research (done by Gemini)

It reads everything and saves it to memory.

Why should you use NotebookLM

Two reasons make it worth your time:

• Less hallucination: It sticks to your sources. No making things up.

• It’s free: Full access. No credit card. No usage limits.

But there are three more connections that can turn it into an AI command center.

We’ll show you exactly how in this article.

But first, you need to train it properly.

What happens after training

Apart from the use cases we’ll explore, using NotebookLM, you can:

• Generate slide decks using the slide deck feature

• Write blogs using the report feature

• Generate podcasts using audio overview

• Create videos using the video feature

• Generate infographics with Nano Banana integration

• Collect data from the web using the data table feature

The possibilities are wide, so let’s start learning how to train it.

How to train NotebookLM

Let me show you step by step:

Let me also write how I trained NotebookLM:

1. Go to NotebookLM website

2. Click on “Try NotebookLM”

3. Next, click on “Create new”

4. Add any sources you want:

   1. Deep search using Gemini

   2. YouTube videos

   3. Website

   4. Drive

   5. Copied Text

Click “Insert” to add your sources.

Your NotebookLM will start training automatically.

How to Train NotebookLM to Act as a Prompt Advisor

Let’s train NotebookLM to become your personal prompt engineering advisor.

Why?

Most prompt engineering tips online are scattered across videos, articles, and tweets. NotebookLM can absorb all that knowledge and give you instant, source-backed advice.

In this scenario, we’ll use Anthropic’s official prompt engineering video as our source.

1. Go to NotebookLM website

2. Click on “Try NotebookLM”

3. Next, click on “Create new”

4. Paste this link: AI Prompt Engineering Deep Dive

5. Click on “Insert”

And you are ready.

Let me give you one more tip.

Click on “Configure Notebook” by following the screenshot below.

And click on Custom.

Then insert this prompt:

“Act as a prompt advisor and optimize the given prompts based on your sources.”

Save it.

Now ask it anything you want about prompting.

What will you get?

Your NotebookLM will analyze any prompt you share, and it will:

• Diagnose what’s wrong (vague instructions, missing context, unclear goals)

• Suggest an alternative strategy

• Execute prompt optimization

For example, I asked it to improve “Write me an article.”

Here’s what happened:

It identified the “Google search box mentality” problem (treating AI like a search engine instead of giving proper context), then rewrote the prompt using the “Interview” method from Anthropic’s course.

The result: A detailed diagnosis, optimized prompt template, and explanation of why it works better.

NotebookLM Hidden Features

You just learned how to train NotebookLM with your own sources.

That’s what 99% of users do. Upload sources. Chat with them. Done.

But Google quietly added three connections that most people don’t know about.

These connections turn NotebookLM from a simple research tool into:

• An app builder (via Gemini)

• A permanent AI assistant (via Gems)

• An automated research system (via Antigravity)

Same NotebookLM. Three hidden connections. Different power levels.

Let’s start with the first one.

1. NotebookLM + Gemini Canvas: Build Apps from Your Sources

Turn your NotebookLM into a working app builder in minutes.

How to do it?

Follow these steps.

1. Create your NotebookLM with sources

2. Open Gemini.

3. Click the “+” button

4. Select NotebookLM as a source

5. Choose your notebook

6. Prompt: “Develop app from attached Notebook.”

Here’s what it looks like.

I used Anthropic’s Prompt Engineering course Notebook earlier as the source.

Here is how the app looks:

What just happened?

Gemini Canvas turned your NotebookLM sources into a working web app in under 2 minutes.

The app uses “The Externalize Your Brain” method from Anthropic’s course.

How does it work?

• Left side: You add your context, describe the task, add examples, and desired output.

• Right side: It generates an optimized prompt for you when you click “Test,” and then you can copy it instantly.

What else can you build?

The same process works for any NotebookLM source:

• Upload cooking videos → Build a recipe optimizer app

• Upload fitness research → Create a workout plan generator

• Upload product docs → Build a customer support chatbot

• Upload meeting notes → Create a decision tracker

It is up to you now.

2. NotebookLM + Gems: Create a Permanent AI Assistant

Gems is Gemini’s version of CustomGPT or Claude Projects.

You create a specialized chatbot with:

• Custom instructions

• Specific knowledge sources

• Permanent memory across all chats

Now here’s the power move: Connect your NotebookLM as the knowledge source.

Result: Your NotebookLM becomes a reusable AI assistant that remembers everything, forever.

How to Create a Gem with NotebookLM

Reference

1. Open Gems inside the Gemini menu.

2. Add “Name” and “Description” to it.

3. In the instructions, put the prompt you want (I’ll show you the example below)

4. Choose your default tool. A new Gem chat starts with its default tool, but you can change it anytime. Here are the available tools:

   1. Deep Research

   2. Create Videos(Veo 3.1)

   3. Create Images(Nano Banana Pro)

   4. Canvas

   5. Guided Learning

5. Knowledge: What’s new here is that you can add NotebookLM as a source.

Then use this prompt as the instruction:

“You are a Prompt Advisor. When a user shares a prompt or describes what they want to achieve with an AI model, you:

Analyze their prompt for weaknesses, missing context, vague instructions, untested edge cases, hidden assumptions, and unclear output format.

Rewrite the improved version of their prompt.

Explain what you changed and why, briefly.

Use your knowledge source to apply proven prompt engineering principles and techniques in your analysis and rewrites.

If the user’s request is too vague to rewrite, ask clarifying questions first, interview them to understand the task before improving the prompt.

Your tone: Direct, practical, no fluff. You’re a senior prompt engineer doing a code review on someone’s prompt.”

Here is what it looks like:

Here is our Gem, trained using the NotebookLM we just created. You can click and test it.

Why Gems Give NotebookLM Permanent Memory

You created a permanent AI assistant that keeps Anthropic’s entire prompt engineering course in its memory, forever.

Unlike regular chats that forget everything when you close the tab, this Gem remembers across all conversations.

Testing Your Prompt Advisor Gem

I sent it my prompt like this:

“Write me an article.”

And here is the Gem’s answer:

What NotebookLM Gems Unlock for You

• Permanent expert memory: Train once, use forever. No need to paste the context in every conversation.

• Smart interviewing: When your prompt is vague (like “write me an article”), it asks the right questions instead of guessing.

• Source-backed advice: Every suggestion comes from Anthropic’s official techniques, not generic tips.

• Shareable expertise: Send this Gem to your team. Everyone gets the same expert-level guidance.

The key difference:

NotebookLM alone = temporary research chat.

Gem + NotebookLM = permanent AI expert you can talk to anytime.

3. NotebookLM + Antigravity: Automate Research with MCP

First, let’s learn what antigravity is.

What Is Antigravity?

Antigravity is an AI-powered IDE built for working with AI agents.

Think VS Code, but designed for AI-native development.

Key features:

• It supports multi-agent workflows and isn’t limited to Google models: Claude, Gemini, and GPT models all work.

• Open Agent Manager: You can build multiple agents simultaneously.

• Live preview: Test your apps by opening them in the browser and capturing screenshots.

It has a pretty solid free quota. You can download it from the Antigravity website.

Now we’ll connect it with NotebookLM by installing an MCP.

What is an MCP?

It is a bridge between your LLM and external apps.

It lets the model interact with those apps through pre-defined functions.

What Kind of Functions?

The NotebookLM MCP we’re about to build gives us access to 32 different actions.

Things like creating a notebook, retrieving notebook content, and much more.

Each action is handled by a specific function within the MCP.

How to install NotebookLM MCP?

No need to configure complex JSON files to install an MCP. We’ll do it with one prompt.

Paste this prompt into Antigravity:

Install and set up the NotebookLM MCP server for me:

1. Install the notebooklm-mcp-server package using uv (or pip if uv isn’t available)

2. Add it to my MCP configuration file (usually at ~/.config/opencode/opencode.json or similar)

3. Run the authentication command (notebooklm-mcp-auth) and open a browser to authorize it

4. Verify the installation is working by listing my NotebookLM notebooks

Make sure to:

• Use the correct installation method for my system

• Configure the MCP server properly in my config file

• Guide me through the browser authentication process

• Confirm everything is working at the end

Let it install the NotebookLM MCP. It will then ask you to sign in with your NotebookLM account in the browser.

Do that, and you are ready to go.

Creating Your First NotebookLM Using MCP

My baby girl will be born in a couple of months.

That is exactly why I’m building apps that organize the things new parents need to know. All I want is to help my wife.

To do that, let’s create a NotebookLM.

Let it find sources using deep research:

“Create a NotebookLM through MCP for me about how to take care of my upcoming baby girl when she is born.

This notebook should be able to find sources through the web using deep research.”

After sending the prompt, it creates a notebook first and then starts to research.

After a couple of minutes, our notebook is ready.

I checked it from the NotebookLM side, and as you can see, it’s ready to go.

After a couple of minutes, our notebook is ready.

I checked it from the NotebookLM side, and as you can see, it is ready and steady.

What’s possible with MCP?

There are 32 different NotebookLM functions available through this MCP.

Beyond creating and populating notebooks automatically, you can:

• Generate audio overviews, videos, and infographics programmatically

• Create slide decks, reports, and mind maps via automation

• Build flashcards and quizzes from your sources

• Extract data tables and sync Drive sources

• Configure custom chat settings and manage notebook workflows

A single MCP connection gives you access to all of NotebookLM’s features through code.

When to Use Each Connection?

• Gemini - Quick apps from content

• Gems- AI assistant with permanent memory

• Antigravity - Automate research & notebook creation & apps from your NotebookLM (one prompt, deep search)

Final Thoughts

NotebookLM wasn’t built to be this powerful.

But Google gave it three hidden connections that turn it into an AI command center.

Most people still use it for document summaries.

You now know how to build apps from YouTube videos, create AI assistants, and automate research with one prompt.

Same tool. Three connections. Different power levels.