Best Practices

Organize by topic, not by file type

Create separate knowledge bases for distinct areas — for example, "Product FAQs", "Pricing & Plans", and "Return Policy" rather than one large "Everything" library. Focused libraries produce more relevant search results because the AI retrieves from a smaller, more targeted pool of content.

Keep individual documents focused

A 10-page document covering one topic retrieves better than a 100-page document covering ten. If you have a large policy manual, consider splitting it into individual topic files before uploading.

Match your similarity threshold to your content

The default threshold (0.7) is a good starting point for well-written, topic-specific documents. If the AI frequently says it doesn't have information when it should, lower the threshold slightly (e.g. 0.6) to allow broader matches. If responses feel off-topic, raise it.

Keep documents current

The AI answers based on what's in the knowledge base — not what's on your live website. When you update a policy, price list, or product description, re-upload the relevant document so responses stay accurate.

Use the retrieval limit thoughtfully

The default retrieval limit (5 chunks) works for most use cases. For workflows covering a wide range of topics, increasing it gives the AI more context. For very focused chatbots (e.g. a single product), a lower limit reduces noise.

Test after changes

After adding or removing documents, send a few test messages through your demo to verify the AI is pulling the right content before sharing the chatbot with customers.