An AI receptionist is only as good as what you've told it. Put in vague, incomplete information and it'll give vague, unhelpful answers. Put in clear, specific information about your actual business and it'll handle enquiries like someone who really knows what they're talking about. This guide walks through what to include, how to write it, and what to leave out.
The garbage in, garbage out problem
Most people set up an AI receptionist in 20 minutes, write a few sentences about their business, and then wonder why the responses aren't that great. The issue isn't the AI — it's the source material. If you'd give the same brief to a new human receptionist, they'd also be giving customers vague answers on their first day.
The more specific and complete your knowledge base, the better the AI performs. It's worth spending an hour getting this right rather than patching it every time a customer gets a confusing response.
What to include
Services with real descriptions
Don't just list your services. Describe them briefly. Instead of "Remedial massage," write "Remedial massage — 45 or 60 minute sessions targeting muscle tension and injury recovery. Often used for neck and back pain, sports recovery, and chronic pain management." That gives the AI enough to answer follow-up questions accurately.
Pricing ranges
You don't have to commit to exact prices — in fact, for project-based work, it often makes sense not to. But giving a ballpark ("initial consultations from $90" or "general service call from $150 + parts") is far more useful than nothing. Customers asking about price aren't trying to lock you in; they're trying to know if you're in the right ballpark before they go further.
Hours and location
Straightforward, but be specific. If you're closed on Sundays but available for emergencies, say that. If parking is limited at your clinic, mention it — that's a real question patients often have. If you service specific suburbs but not others, be explicit about the boundaries.
Booking process
How does a new customer book with you? Walk through it step by step. Do they call, fill out a form, use an online booking system like Calendly, HotDoc, Cliniko, or Fresha? Is there an initial consultation before a treatment plan? Do they need to fill out health forms before arriving? The booking process is one of the most common things customers ask about, and if the AI can walk them through it confidently, you remove a lot of friction.
FAQs written the way customers ask them
This is where most knowledge bases fall short. Business owners write FAQs the way they'd phrase them, not the way customers ask. Instead of "What is your cancellation policy?" write "Can I cancel or reschedule my appointment?" That's what people actually type. Read through your actual enquiry history if you have it, or just think about the questions you hear most often on the phone.
What to leave out
Anything you're not comfortable committing to in writing. If your prices are genuinely case-by-case and you don't want a customer holding you to a quote the AI gave them, don't put a number in the knowledge base — tell the AI to say pricing varies and offer to call back with an assessment.
Complex clinical or legal situations that require professional judgement shouldn't be handled by an AI. If you're a healthcare provider, you don't want the AI making treatment recommendations. It should provide information and direct people to book an appointment.
Internal business information that customers don't need to know — staff schedules, supplier relationships, internal processes — doesn't belong in a customer-facing knowledge base.
The update habit
Your AI needs to reflect your current reality. If you add a new service, change your hours, or update your pricing, the knowledge base needs to follow. Set a monthly reminder to review it — it only takes 10 minutes, and it stops the AI from giving outdated information to customers.
Testing your own setup
Before you go live, have a conversation with your own widget as a stranger would. Ask the questions a first-time customer would ask. Notice where the answers feel vague or off. Those gaps are where you need more information in your knowledge base. Do this when you first set it up, and again after any significant update.
If a question catches the AI off-guard and it gives a generic answer, that's a signal to add a specific answer to your FAQs. Each iteration makes it more useful.
Think of the knowledge base as your AI's training manual. A new staff member with a good manual performs far better than one who was just pointed at the front desk.