Guide: Best Practices for an AI Agent's Knowledge Base

The quality of your AI Agent's answers is directly proportional to the quality of its knowledge base. The RAG (Retrieval-Augmented Generation) system is powerful, but it relies on well-structured, clear information to be effective.

Here are the best practices for creating and managing your knowledge base documents.

1. One Topic, One Document

Don't put all your company information into a single, massive file. It's much more effective to create separate, focused documents for each topic. This helps the AI quickly find the most relevant context.

• Good: shipping_policy.txt, return_policy.txt, product_specs_model_x.txt
• Bad: all_company_info.txt

2. Use Clear, Descriptive Headings

Structure your documents logically using headings. If you are using Markdown (.md) files, use markdown syntax (## for headings, ### for subheadings).

Example (faq.md):

## Shipping Information

### How long does shipping take?
Standard shipping takes 3-5 business days. Express shipping takes 1-2 business days.

### What are the shipping costs?
Standard shipping is $5. Orders over $50 qualify for free standard shipping.

## Return Policy

### What is your return policy?
We accept returns within 30 days of purchase for a full refund. The item must be in its original condition.

This structure helps the AI understand the hierarchy and context of the information.

3. Write in Plain Language (Q&A Format is Best)

Write the content as if you were explaining it directly to a customer. A question-and-answer format is ideal because it closely mimics how the user will interact with the agent.

• Good: "How do I reset my password? -> To reset your password, click the 'Forgot Password' link on the login page..."
• Bad: "Password resets are a user-initiated process via the authentication portal."

4. Be Explicit and Thorough

The AI can only know what you tell it. Don't assume it can infer information.

• If your store is closed on Sundays, explicitly write "We are closed on Sundays."
• If a product comes in three colors, list them: "The SuperWidget comes in Red, Green, and Blue."

5. Keep it Updated

A knowledge base is a living resource. If your return policy changes or your pricing is updated, you must update the corresponding document in your agent's knowledge base.

• Process:
  1. Edit the file on your local computer.
  2. In the AI Agent's settings, delete the old version of the file.
  3. Upload the new, updated version.

By following these best practices, you can build a robust and accurate knowledge base that empowers your AI Agents to provide truly helpful, instant support to your customers.