Best Practices for Building Your Knowledge Base

Why Your Knowledge Base Matters

The knowledge base is the single most important factor in how well your AI performs. Think of it as the AI's brain — the more relevant, accurate information you feed it, the smarter and more helpful it becomes. A sparse knowledge base leads to vague, generic responses. A well-stocked one leads to precise, helpful answers that feel like they came from a human expert.

What to Upload

Start with the essentials: your return policy, shipping policy, warranty information, and FAQ document. These cover the majority of customer questions and immediately improve the AI's accuracy.

Next, add product-specific documentation: sizing guides, care instructions, compatibility charts, or installation guides. Any information that helps a customer make a purchasing decision or use your product correctly belongs in the knowledge base.

Don't forget your brand voice guidelines. Upload a document that describes your brand's tone, vocabulary, and values. The AI can reference this when crafting responses, ensuring consistency across all customer interactions.

Document Format Best Practices

The AI processes your documents best when they are clean and well-structured. Use clear headings (H1, H2, H3) to organize information. Write in complete sentences rather than bullet-point fragments — the AI understands narrative context better than isolated data points.

For product data, CSV files work well. Include columns for product name, description, price, category, stock status, and any unique attributes (size, color, material). The AI uses this structured data to answer factual questions like "Do you have this in blue?" or "What's the cheapest option?"

Avoid uploading scanned PDFs with images of text. The AI can't extract text from image-based PDFs. Use text-based PDFs or plain text files whenever possible. If you only have scanned documents, run them through OCR (optical character recognition) software first.

Website Crawl Strategy

The website crawl is the fastest way to build your knowledge base, but a thoughtful crawl strategy makes a big difference. Start by crawling your key pages: product pages, FAQ page, About page, and any policy pages (returns, shipping, privacy).

Avoid crawling blog archives or news sections unless they contain evergreen product information. The AI doesn't need to know about last year's holiday sale — in fact, outdated information can cause it to give incorrect answers.

After the initial crawl, set a recurring crawl schedule (weekly is ideal for most stores). This ensures your AI stays current as you add products and update policies.

Testing and Iterating

Don't just upload and forget. Test your AI regularly by asking it questions you know customers will ask. Check the Conversations tab to see what customers are actually asking, and add any missing information to your knowledge base.

If the AI gives an incorrect or incomplete answer, investigate why. Usually, it's because the relevant information wasn't in the knowledge base, or it was present but poorly structured. Fix the source document and re-upload — the AI will immediately improve.

This feedback loop is the secret to a great AI experience. The stores with the best-performing chatbots are the ones that treat their knowledge base as a living document, constantly updating and refining it based on real customer interactions.