What Belongs in an AI Receptionist Knowledge Base?

An AI receptionist can only give a dependable answer when the business information behind it is clear and current. A pile of old PDFs and website pages is not yet a usable knowledge base. The goal is a small set of approved answers, decision rules, and handoff instructions that match how the business actually operates.

Start with questions callers really ask

Review a sample of calls and messages. Group the repeated questions by topic: hours, service areas, appointment types, pricing policies, preparation, cancellations, and how to reach a person. Write an answer for each frequent question in the words your team would use with a customer.

Do not include every internal detail. The agent needs the answer it is allowed to share and the condition under which that answer applies. For example, “We serve these postal codes” is more useful than a long sales document about expansion plans.

Add boundaries beside answers

Each answer should have an owner, a last review date, and a rule for uncertainty. If a caller asks for a price that depends on an inspection, the approved answer should explain the process for getting a quote. It should not invent a figure. If availability changes, the agent should check the live calendar rather than quote a static list of slots.

The knowledge base should also say when to stop answering and route to a person. Complaints, unusual requests, and questions outside the approved information need a clear handoff destination. A useful “I need to connect you with the team” is better than a polished guess.

Keep facts and actions distinct

An answer such as opening hours is reference information. Booking an appointment is an action that depends on live system state. The agent may need both, but they should not be maintained as one paragraph of instructions. Keep policies in one place and action rules in another so updates are easier to verify.

For each action, write down the required inputs, the system that confirms success, and the fallback if that system fails. A booking is confirmed by the calendar, not by the agent's intention to book.

Make updates part of the operating routine

Assign a person to review the content whenever hours, coverage, services, or policies change. Keep a short change log, then test the most important questions after each update. If staff repeatedly correct the same answer, the source information or its rule probably needs attention.

Before launch, ask team members to test the knowledge base with real phrasing, including vague and interrupted questions. Note whether the agent answers correctly, asks a useful follow-up, or hands off. That review creates a more reliable system than simply adding more documents.

For the full call flow around this information, see What Is an AI Voice Agent for Business?. Ontix can help turn approved answers and handoff rules into an AI voice agent.