Building knowledge bases that serve as the Retrieval-Augmented Generation (RAG) for Generative AI has been a fascinating journey. A lot of the Gen AI tools I’ve used, from Anthropic to OpenAI, BoodleBox, a myriad of web-based models to Google’s Gemini and NotebookLM have relied on different approaches. A set of rules has emerged for most of the Gen AI tools, but Google Gems and NotebookLM obviously have changed the ballgame. I hope you’ll offer critiques of the organization.

Listen to Audio explanation of my thinking about this blog entry

Let’s review how things have worked in the past, and still do for many of these tools. I want to briefly cover these before jumping into the ways that Gemini, Gem, and NotebookLM make things better through the use of Google Drive and permanence/updating of docs attached.

The Update Rule v1: Delete, then Replace

When working with a ChatGPT Project or Custom GPT, you usually provide information that limit “hallucinations” or the made up content. The more detailed the content, the stricter the instructions to adhere to that provided content ensures the Gen AI produces content that is predictable to your specifications. That’s why RAG is so important.

ChatGPT Plus (the version I have available to me) has certain limits for both Projects (which allow a level of permanence, and having multiple chats in a Project allows all chats to benefit from shared chats and creations in that Project) and Custom GPTs. I prefer ChatGPT Projects since I can have multiple chats informed by information distilled into text documents.

The updating rule for ChatGPT Projects and Custom GPTs is simple:

To update the knowledge base, remove obsolete, discrete module of content and update it.

How do you remove it? You delete the file. You don’t edit or modify it online. Instead, you prepare it offline, then add it to the Project or Custom GPT.

As far as I know, this process is true for Claude Projects, BoodleBox Knowledge Bases, and local AI chatbots. To replace the information the Gen AI refers to, you delete the content, then replace it with a new version. You are “plugging in” a new module with updated information.

The Google Gemini Improvement

When using Google Gemini, you can add lots of sources and connect them easily to a Gem. These sources, if Google Docs, can be updated.

The Update Rule v2: Modify and Refresh

The rule for these?

Connect the Google Docs to either a NotebookLMs or a Gem. When you need to update content in the Google Doc, update the Google Doc by modifying the content of the document.

This is different than deleting a file in ChatGPT Projects/GPT, then uploading a replacement. This means that, over time, you are able to get fine-tuned sources for your knowledge base. And, they are all conveniently stored in Google Drive.

This is a huge improvement over juggling text files (and other file formats).

Testing Google’s Live-Update Approach

As you can see from the diagram above (which captures all the text info shared in this blog entry above into an infographic), you can see that live documents are powerful. A simpler way to think of this:

This is where you have multiple documents in Google Drive that have been added as sources to a NotebookLM. You can have a variety of NotebookLMs, in turn, feeding into a Google Gem. Here’s another way to see it:

Or, of course, this:

For me, this last one captures the idea or concept I’m trying to get at best. My next step? Trying this out with a real project.

Assembling the Gem Project on Gen AI

So, I’ve been collecting tons of information into a variety of “content” or “topic libraries.” These are knowledge bases that I can apply to different situations. For example, I’ll be working with various associations and I want to generate content that is predictably about applying Gen AI in their unique situations. These topic libraries of information are ones I can add to each.

Here’s a sample outline of what’s possible:

Building Your AI Marketing Research Gem: A Step-by-Step Guide

Begin with the End in Mind

Your Goal: A custom Gem that serves as your unified AI Marketing Research Assistant, capable of answering questions across three specialized knowledge domains—marketing applications, organizational adoption, and ROI measurement.

Phase 1: Organize Your Source Documents in Google Drive

Step 1: Create a Dedicated Folder Structure

📁 AI Marketing for XYZ Organization Research Hub (Gem)
   ├── 📁 AI in Marketing (NotebookLM)
   │   ├── Case studies
   │   ├── Tool comparisons
   │   ├── Campaign examples
   │   └── Trend reports
   ├── 📁 AI Adoption Strategies (NotebookLM)
   │   ├── Change management frameworks
   │   ├── Implementation guides
   │   └── Organizational readiness assessments
   └── 📁 AI ROI (NotebookLM)
       ├── Metrics and KPIs
       ├── Cost-benefit analyses
       └── Success measurement frameworks

The information for the NotebookLMs above would come from a variety of transcribed podcasts and other content gathered. It is worth pointing out that I would have another NotebookLM outlined above focused on what the Association or organization might be about, etc. As much information as I could gather would go into it. I have not reflected that last NotebookLM about the Organization/Association below.

Step 2: Prepare Your Documents

For each folder, ensure your Google Docs are:

• Clearly titled — Use descriptive names (e.g., “2025 AI Email Marketing Benchmarks” not “Notes 3”)
• Well-structured — Use headings (H1, H2, H3) so NotebookLM can parse sections effectively
• Source-attributed — Include where the research came from for credibility
• Date-stamped — Add “Last Updated: [Date]” at the top for version tracking

Document types to include:

Phase 2: Build Your Four NotebookLMs

Step 3: Create NotebookLM #1 — AI in Marketing

1. Go to notebooklm.google.com
2. Click New Notebook
3. Name it: AI in Marketing Research
4. Add sources by clicking Add Source → Google Drive
5. Select all documents from your 📁 AI in Marketing folder
6. Let NotebookLM index and generate its initial summary
7. Test with a query: “What are the most effective AI tools for email marketing personalization?”

Step 4: Create NotebookLM #2 — AI Adoption Strategies

1. Create a new notebook named: AI Adoption Strategies for Organizations
2. Add sources from your 📁 AI Adoption Strategies folder
3. Test with: “What are the key barriers to AI adoption in mid-size companies?”

Step 5: Create NotebookLM #3 — AI ROI

1. Create a new notebook named: AI Return on Investment (ROI)
2. Add sources from your 📁 AI ROI folder
3. Test with: “How do I measure the ROI of an AI-powered content generation tool?”

Phase 3: Create Your Unified Gem

Step 6: Access Gem Creation

1. Go to gemini.google.com
2. Click on Gem manager (left sidebar)
3. Click New Gem

Step 7: Configure Your Gem’s Identity

Name: AI Marketing Research Assistant

Instructions (paste this as your Gem’s system prompt):

You are my AI Marketing Research Assistant, specializing in three interconnected domains:
1. **AI in Marketing** — You know about AI tools, platforms, and techniques used in marketing campaigns, personalization, content creation, analytics, and customer engagement.
2. **AI Adoption Strategies** — You understand how organizations implement AI, including change management, stakeholder buy-in, training, pilot programs, and scaling strategies.
3. **AI ROI** — You can discuss metrics, KPIs, cost-benefit analysis, and frameworks for measuring the return on investment of AI marketing initiatives.
When answering questions:
- Draw connections across all three domains when relevant
- Cite specific research or examples when available
- Provide actionable recommendations, not just theory
- Flag when information may be outdated and suggest verification
- Ask clarifying questions if my query spans multiple areas
My goal is to make informed decisions about AI marketing investments for organizations.

Step 8: Connect Your NotebookLMs as Context

When chatting with your Gem, you can reference your NotebookLM research by:

1. Opening a relevant NotebookLM
2. Copying key synthesized insights or summaries
3. Pasting them into your Gem conversation as context

Pro tip: Create a “Master Summary” Google Doc that contains the key findings from each NotebookLM, then attach that doc when chatting with your Gem for persistent context.

Phase 4: Maintain Your Living Knowledge Base

Step 9: Establish an Update Workflow

Step 10: Test Your Complete System

Run these queries through your Gem to verify everything works:

What AI marketing tools offer the best ROI for small marketing teams, and what adoption challenges should I anticipate?

Create a 90-day implementation plan for adopting AI-powered content generation, including success metrics I should track.

Compare the ROI of AI chatbots vs. AI email personalization based on my research.

Quick Reference: The Complete Flow

Google Drive Docs (live, editable)
        ↓
   NotebookLM #1: AI in Marketing
   NotebookLM #2: AI Adoption Strategies  
   NotebookLM #3: AI ROI
   NotebookLM #4: Organization Specific Info
        ↓
   Gem: AI Marketing Research Assistant
        ↓
   You: Unified answers across all research