How a knowledge base powers an AI assistant
When you add sources, the system splits them into passages and indexes them for search. At answer time, the assistant retrieves the passages relevant to the question and writes a reply from them, a technique called retrieval-augmented generation. The quality of answers is capped by the quality of the sources.
The best knowledge bases are specific: real prices or price ranges, service areas by town, what is and isn’t included, cancellation rules, and what to do in urgent situations. Contradictions between an old PDF and the current website are a common source of wrong answers, so remove or refresh outdated sources.
Example: an interior design studio
An interior design studio adds its services page, a PDF describing its three design packages, and written answers about consultation fees and travel radius. A visitor asks whether the studio does single-room refreshes outside the city. The assistant answers from the package PDF and the travel-radius answer, then offers a consultation slot. This scenario is illustrative.
Knowledge in InsertChat
InsertChat’s knowledge base accepts website pages by crawl or sitemap; PDF, Word, PowerPoint, text, and ebook files; CSV and Excel spreadsheets; scanned PDFs and images read with text recognition; audio and video files, which are transcribed; and questions and answers you write. One knowledge base serves calls, chat, texts, email, and WhatsApp.
Website sources can refresh on a schedule that depends on your plan, and Knowledge gaps lists real customer questions the assistant couldn’t answer well. Start with how to build your assistant’s knowledge base, or see the interior design setup.
