What to Put in Your AI Knowledge Base (And What to Leave Out)

Your AI chatbot is only as smart as what you put inside it. A business that uploads a well-structured knowledge base gets an AI that answers questions accurately, sounds like the brand, and handles 80% of inquiries without human help. A business that uploads the wrong content — or nothing at all — gets a bot that confuses customers and damages trust. The difference is entirely in what you choose to include.

This guide walks you through exactly what belongs in your AI knowledge base, what to leave out, and how to write entries so the AI can use them effectively.

What to Include in Your AI Knowledge Base

The goal of your knowledge base is simple: give the AI everything it needs to answer the questions your customers actually ask. Start by pulling your top 20 questions that your team answers every single day. These are your highest-value entries.

1. Frequently Asked Questions

Think about the questions your support team handles repeatedly. Compile a list of the top 20 and write clear, direct answers for each. These are the backbone of any effective knowledge base. Examples include: "What payment methods do you accept?", "How do I track my order?", and "Can I return a product after 30 days?"

2. Product and Service Descriptions

Include accurate descriptions of everything you sell or offer. Be specific: list exact specifications, available sizes and colors, weight, material, pricing, and any limitations. Vague entries like "we sell clothing" produce vague answers. Specific entries like "The Classic Linen Shirt is available in sizes S–XXL in white, beige, and navy. Price: 149 AED" produce accurate, useful answers.

3. Company Policies

Your return, refund, exchange, and shipping policies should be written out in full. Customers ask about these constantly, and the AI needs exact details to answer correctly. Include conditions, timeframes, exceptions, and the steps involved. Do not reference external documents — write the policy text directly.

4. Business Hours and Contact Methods

Include your operating hours for every day of the week, including public holidays if relevant. Add your WhatsApp number, email address, phone number, and physical address if applicable. This avoids customers asking the AI for information it cannot find.

5. Troubleshooting Guides

For businesses selling products or services with technical components, add step-by-step troubleshooting guides for the most common issues. "My order shows delivered but I did not receive it" or "I cannot log into my account" are examples of issues the AI can resolve if the guide is available in the knowledge base.

What NOT to Include

Just as important as what you add is what you leave out. The wrong content does not just waste space — it actively degrades the quality of AI responses.

• Internal HR and finance documents: Employee handbooks, salary information, and internal processes should never enter your customer-facing knowledge base.
• Legal fine print: Terms and conditions written in dense legal language confuse AI models. If a policy is important, summarize it in plain language and add that summary instead.
• Outdated prices or discontinued products: Stale information is worse than no information. If the AI quotes an expired price, you have a customer service problem. Review your knowledge base every time prices or products change.
• Competitor comparisons: Never include competitive analysis in your knowledge base. The AI may surface competitor names in unintended ways.
• Internal jargon and abbreviations: Your internal shorthand means nothing to customers. Write entries in the plain language your customers use.

How to Write Knowledge Base Entries the AI Can Use

Even the right information, written poorly, produces poor AI responses. Follow these principles when writing each entry:

1. Use clear, complete sentences. Bullet points and abbreviations work in human notes but perform poorly as AI training data. Write full sentences that state the complete fact.
2. One fact per statement. Combine too many ideas in one sentence and the AI may extract only part of the meaning. Keep each statement focused on a single piece of information.
3. No ambiguity. Avoid language like "usually", "sometimes", or "it depends" unless you follow immediately with what it depends on. The AI will mirror your vagueness back to customers.
4. Consistent terminology. If your product is called "the Classic Linen Shirt", use that name everywhere. Mixing "linen shirt", "classic shirt", and "the shirt" in different entries makes the AI less precise.

How Often to Update Your Knowledge Base

A knowledge base is not a set-and-forget document. It needs to stay current to stay accurate. At minimum, review and update your knowledge base whenever:

• Prices change
• New products or services are added
• Policies are updated
• Business hours change for holidays or seasonal periods
• You receive a cluster of similar complaints the AI could not handle

A quarterly review is a good baseline for most businesses. High-volume businesses with frequent product changes should review monthly or on every significant catalog update.

How to Test the Quality of Your Knowledge Base

Before going live, test your AI with your top 10 customer questions. Ask each one exactly as a customer would phrase it, including informal language and typos. Evaluate every response:

• Is the answer accurate?
• Is it complete, or does it leave out important details?
• Does it sound like your brand?
• Does it answer what was actually asked, not just what the question contained?

For any response that falls short, go back to the relevant knowledge base entry and improve it. Add missing details, clarify ambiguous language, and retest. This loop — ask, evaluate, improve — should continue for the first two weeks after launch.

Once your AI is live, the RevoplyAI dashboard shows you every conversation, including cases where the AI said it could not find an answer. Those gaps are your roadmap for what to add next. Over time, your knowledge base becomes more complete and your AI handles a higher percentage of conversations without escalation.

The after-hours benefit is one of the first things businesses notice: a well-fed knowledge base means overnight inquiries are answered correctly without anyone on your team being awake.

Watch: How to Add Content to Your Knowledge Base

These short tutorials show exactly how to add your business information to RevoplyAI — uploading PDFs and Word files, training from a YouTube video, or typing directly in the text editor.

Frequently Asked Questions

How many entries does a knowledge base need to work well?

There is no fixed number, but most businesses see strong results with 30 to 60 well-written entries covering their most common questions, core products, and key policies. Quality matters far more than quantity. Ten accurate, clearly written entries outperform 100 vague or outdated ones.

Can the AI handle questions not in the knowledge base?

A well-configured AI agent will tell the customer it does not have that information and offer to escalate to a human agent, rather than guessing or making something up. This is the correct behavior. Seeing these gaps in your conversation history tells you exactly what to add next.

Should I upload PDF catalogs or write entries manually?

Both can work, but manually written entries consistently produce better AI responses. PDFs formatted for print often contain layouts, headers, and tables that are hard for AI to parse cleanly. If you use PDFs, keep them text-heavy and well-structured. For anything critical — pricing, policies, FAQs — write the entries directly in plain text.