How to train an AI support agent on your knowledge base (the right way)
Supportweave Team · June 10, 2026 · 6 min read
An AI agent is only as good as what it knows. The difference between a bot that delights customers and one that invents policies comes down to how you feed and ground it.
1. Bring the right sources
- Crawl your whole site, not just the homepage
- Upload canonical docs: policies, manuals, pricing
- Add FAQs for the questions you already get daily
- Keep it fresh with scheduled re-crawls
2. Ground every answer
Retrieval should fuse semantic search with keyword search, with your FAQs boosted so the answers people ask for most surface first. Answers should cite their sources, and when nothing relevant is found, the bot should say 'I don't know' instead of guessing. An optional LLM reranking pass can re-sort the top results for relevance when you want to trade a little latency for extra precision.
3. Close the loop
Track the questions your bot couldn't answer and turn them into new FAQs. A knowledge base is a living thing; the gaps your customers surface are your roadmap.
Supportweave does the heavy lifting: hybrid retrieval with FAQ boosting, sources, calibrated 'I don't know', and a knowledge-gap loop that turns misses into FAQs in one click — plus optional LLM reranking when you want to tune for precision.