Voice AI Use Case
Build a Voice AI Knowledge Base From a Client Website
By Marnix Geerkens. Published 2026-05-12. Updated 2026-05-28.
In short
A knowledge base gives a GoHighLevel Voice AI agent real facts to answer from, pulled from the client website (services, prices, hours, policies). Without it, the agent guesses and gets things wrong. You build the base by gathering the key pages, turning them into clean question-and-answer pairs, and adding them to the agent so it answers from your facts. A good base is the single biggest fix for an agent that makes things up.
- A knowledge base is the set of real facts the agent answers from.
- Pull it from the client website: services, prices, hours, and policies.
- It is the best fix for an agent that guesses or invents answers.
What is a Voice AI knowledge base?
It is the collection of facts the agent can pull from when a caller asks a question. Think of it as the agent reference book: services, prices, hours, location, and the answers to common questions.
When the agent has a knowledge base, it answers from your facts. When it does not, it falls back on the model guessing, which is how you get wrong prices and made-up policies on a call.
Why pull the knowledge base from the client website?
The website is where a client has already written down the truth about their business: what they offer, what it costs, when they are open, and the rules. It is the cleanest source you have.
Reusing the website also keeps the agent in sync with what the client tells the public, so the phone says the same thing the site says.
How do you build the knowledge base?
Gather the key pages
Collect the pages that hold facts: services, pricing, about, hours and location, and any FAQ. Skip the fluff and the marketing copy.
Turn the content into question and answer pairs
Rewrite each fact as a short question and a short answer, the way a caller would ask it. "How much is a deep cleaning?" with the real price. "Are you open Saturdays?" with the real hours.
Add the base to the agent
Load the question and answer pairs into the agent knowledge base or paste them into the prompt for a small set. Tell the agent to answer from this source first.
Add a "do not know" rule
Tell the agent: "If the answer is not in your knowledge base, say you do not have that and offer to take a message." This stops it from guessing.
How do you keep the knowledge base accurate?
Set a habit to update it when the client changes prices, hours, or services. A stale base is worse than none, because the agent confidently says the wrong thing.
Review the agent call logs now and then to spot questions it could not answer, then add those to the base.
What are the common mistakes?
Dumping whole marketing pages into the base. The agent does better with clean question and answer pairs than with long sales copy.
No "do not know" rule, so the agent guesses when a question is not covered.
Letting prices and hours go stale. Update the base whenever the client changes them.
Forgetting to review what callers actually ask. The call logs tell you the gaps.
Frequently asked questions
What is a knowledge base for a Voice AI agent?
It is the set of real facts the agent answers from: services, prices, hours, location, and common questions. With a knowledge base, the agent answers from your facts. Without one, it guesses, which is how wrong answers reach callers.
How do I build a knowledge base from a client website?
Gather the pages that hold facts (services, pricing, hours, FAQ), rewrite each fact as a short question and answer pair, add the pairs to the agent, and tell it to answer from this source first. Add a rule to say "I do not know" for anything not covered.
How do I stop the agent from making up answers?
Give it a knowledge base of real facts and add a clear rule: if the answer is not in the base, say so and offer to take a message. An agent with no facts to draw on is the one that invents answers.
How often should I update the knowledge base?
Update it whenever the client changes prices, hours, or services. A stale base is worse than none because the agent will confidently state the wrong thing. Review call logs to catch questions the agent missed.
Does this work for Conversation AI too?
Yes. The same knowledge base feeds both Voice AI on calls and Conversation AI in chat and SMS. Build it once and both the phone agent and the chat bot answer from the same facts.