What is an Agent Knowledge Base?

A knowledge base is a well-organized and relevant collection of information about your business that is uploaded to AI. The purpose is to introduce unknown information and prioritize known information.


Why is this so critical?


Your agent has a brain and is quite talented, but has no memory of your business. Every chat starts with a stranger who has to guess everything about you, your customers, and your true objectives.


When AI guesses, you get generic BS that needs heavy editing. 


When you provide your agent with information they can reference during your chat, they won't make up information to complete tasks. Instead, they'll give you more accurate, relevant, and valuable outputs, making your agent an indispensable tool for you and your team.

The 20-Minute Knowledge Base

Don't want to overthink this? We built a quick template that works for most businesses and covers the basics your agent needs.

  • The 20-Minute Knowledge Base


    1. Business Snapshot

    • Business Name
    • Location
    • Industry/Niche
    • What You Do in 1 Sentence
    • Your Unique Selling Proposition

    2. Core Offers

    • Offer Name → Price → 1-line result
    • Repeat for each major product or service

    3. Customer Profile

    • Ideal Client: (1 sentence)
    • Top 3 Pain Points
    • Top 3 Desires

    4. Proof

    • 1–2 Testimonials or Case Studies With Quick Wins

    5. Brand Style

    • Voice Description (3 words)
    • Write like
    • Don’t sound like

    ---


    This template, along with a completed sample is available for download, or to copy from a shared Google Doc.

    Get This Template

Done? Skip to Lesson 3.


Want to build a comprehensive knowledge base like we do for clients? Keep reading...

Types of Information to Include

Most of the knowledge will come directly from you, so what’s at your fingertips? The information will vary depending on the task your agent will perform. Below is a list of information we look for when starting to build a knowledge base.

  • Types of Business Information

    • Mission statement and company values
    • Brand guidelines and style guide 
    • Detailed product/service descriptions/menu
    • Technical specifications
    • Pricing information
    • Demographics and Target Market
    • Frequently Asked Questions 
    • Customer reviews and testimonials
    • Common pain points, objections, and motivations 
    • Past successful campaigns
    • Email marketing templates
    • Ad copy examples
    • Social media posts
    • Press releases 
    • Sales scripts
    • Sales history and goals
    • Common objections and responses
    • Case studies 
    • Industry reports and white papers
    • Competitor analysis
    • Market research data
    • Relevant news articles 
    • Target keywords
    • SEO best practices
    • Content gaps analysis 
    • Support ticket logs
    • Common issues and solutions
    • Customer feedback 
    • Employee handbooks
    • Training materials
    • Company policies and procedures 
    • Terms of service
    • Privacy policies
    • Industry-specific regulations 
    • Past blog posts
    • Newsletters
    • Webinar transcripts
    • Infographics 
    • Videos

Additionally, you can include articles, books, and other sourced information you want your agent to prioritize.

How to Prepare Your Data

After collecting your information, you may have PDFs, website content, spreadsheets, audio files, and other formats. Now it's time to package everything so AI can actually use it.


Why Preparing Your Data Matters


How you organize your data is crucial for performance. You want your agent to find answers at a glance, not hunt through messy files with buried information.


Well-structured data is like a neatly labeled spice rack — you can instantly grab what you need instead of hunting through jars.


The Best File Type


Your AI likes well-formatted text in Markdown (.md) files best. Don't worry - it's simple. Format your content in Google Docs, then click "File" → "Download as .md." Done.


Other formats will work, but they leave more room for error. Errors lead to inconsistent performance and make troubleshooting cumbersome.


How AI Process Different File Types

  • Markdown Syntax (Winner)

    Markdown Syntax (.md) is universally understood, very light and easy to retrieve information from. We transform all of our knowledge base documents into this format.


    If it sounds complicated, it's not. All you need to do is format your content onto a Google Doc, click "File" → "Download as .md." Done.


  • Word & Google Docs

    Work well and are easy to edit, but they're heavier files that can slow down processing. AI can read them, but markdown is cleaner and faster. If you're already working in Google Docs, just download as .md when you're done.


  • Plain Text

    Text.txt is light but lacks the formatting, leaving it to rely solely on the comprehension of text, symbols, and emojis which makes it difficult to retrieve info at a glance. Plain text is not ideal.

  • PDFs

    PDFs can be tricky for AI to process. While AI can read text-based PDFs, the formatting often gets jumbled - tables break apart, columns merge together, and headers get separated from their content.


    Some PDFs are actually saved as images (like scanned documents), which makes it more difficult. If you can't highlight and copy text from the PDF, it's an image file.


    Best practice: Even with text-based PDFs, copy the content into Google Docs where you can clean up the formatting and ensure your agent reads it properly.

  • PowerPoint / Slides

    Common for business presentations, but AI struggles with slide layouts, bullet animations, and design elements. The text is there, but buried in formatting.

  • Spreadsheets

    CSV and JSON are both excellent formats for structured, tabular data, making them easy for large language models (LLMs) to process and understand.


    Excel (.xlsx) can be processed, but the complexity may vary. It is better to convert to CSV for more straightforward data.

  • Website Links

    Not all LLMs can access the Internet, and they tend to be inconsistent when asked to follow links or browse the Web. 


    Best practice: For better control over your knowledge base, copy the content, reformat it in a Google Doc and upload to your knowledge base as a .md file.

  • Image Links

    Images are large and more difficult to process quickly. However, in some scenarios, they are irreplaceable in their purpose. You know what they say.


    Best practices when using images (.jpg, .png):


    Clearly title the files. EX: Photo_Restaurant_Main_Dining_Room_View_From_Bar_Seat


    Include information about the images, or at least acknowledge that there are images in the knowledge base.


    Keep the pictures focused, avoiding excessive clutter that detracts from their purpose.


    Include a short description as a text overlay on the image.

  • Video Files

    Not all LLMs accept video files like MP4. The files are also large and complex, making them difficult to process quickly. Your best move is to transcribe the video into text.

Action Steps

Create Your Knowledge Base

This organized foundation you're creating will serve you far beyond your first agent. You'll use it to build specialized agents, deploy advanced AI tools, and stay ahead as AI becomes essential for business.

1. Collect 

Copy text files, transcribe audio files, and gather PowerPoint content into Google Documents.

2. Clean 

Remove irrelevant content like ads, navigation menus, and off-topic information that could confuse your agent.

3. Prepare

Add clear headings, bullet points, and bold text to make information easy for AI to scan and retrieve.

4. Finish

Title each document clearly, download as .md files, and save everything in a organized folder.

Bonus: Real Agent Build

I'll play right alongside you—building a personalized meal planner for my pet. Sounds silly? Maybe. But it perfectly demonstrates the transformation  from useless AI chats, to a genuinely helpful tool.

  • Example Agent

    Roman’s Meal Planner


    Info from me: I included relevant information about my pet like age, breed, weight, diet, exercise, etc. I also included a picture.


    Info from Internet: I chatted with AI to find the top people in the field and to locate their articles and blogs.


    Note: AI already knows about these articles, it just doesn't know I want the agent to prioitize that information above all.

☑️  Lesson Checkpoint


YES,  I have at the information that my agent will need to know on a Google Document.

YES, I add clear titles and downloaded my Google Docs as a Markdown file (.md)

YES, I'm ready and excited to deploy my first agent in the next few minutes!

Lesson 3:

  • What are the best practices when adding content to a document?

    Front-load key information

    LLMs prioritize early content. Important context up top improves understanding and response quality.


    Prioritize value

    LLMs process a limited number of tokens per request. Dense, relevant content avoids wasting that limit.


    Consistent formatting

    Clean, predictable formatting helps LLMs extract structure and meaning (e.g., headings, bullets, tables).


    Redundancy for essential info

    Repeating critical points (like names, definitions, rules) ensures they’re retained and reinforced across context windows.


    Remove outdated or irrelevant data

    Old or conflicting info can confuse the model or lead to inaccurate answers.

  • What if I'm not technical?

    Perfect. This is for business owners, not programmers. If you can copy, paste, and upload files, you're qualified to build an agent.

  • Should I build multiple agents?

    One first. Master the process, see the results, then build your second in 30 minutes. Most businesses end up with 3-5 chat agents before building other AI solutions.