ai-receptionist,buyer-guides
Training your AI receptionist on your FAQs
Your AI receptionist is only as good as the information you feed it. Most platforms let you upload a list of FAQs, and most business owners upload a list that’s too short, too vague, or copied from a website that was written for Google instead of callers. Training the AI on your FAQs is the highest-leverage setup work you’ll do, and it takes less time than you’d think if you do it in the right order.
Collect the real questions first
Start with the questions your callers actually ask, not the ones your website lists. The website FAQ was written to rank; the phone FAQ should reflect what people say out loud. If you’ve answered phones yourself, you already know the top ten. Write them down exactly as callers phrase them.
If you haven’t been the one answering, ask whoever has. One good prompt: “What do you find yourself saying five times a day?” That sentence will produce your core list in about ten minutes.
Typical phone FAQs for a service business look like this:
- What are your hours?
- Where are you located / do you come to my area?
- How much does [common service] cost?
- How soon can you get me in?
- Do you offer free estimates?
- What forms of payment do you take?
- Do you guarantee your work?
- I’m an existing customer and I need to reschedule.
Notice these are short and blunt. Callers on the phone don’t ask “What is your cancellation policy and what fees apply?” They say “What if I need to cancel?” Write the questions the way they’re asked.
Write answers for the ear, not the eye
This is where most FAQ training goes sideways. A website answer can be three paragraphs. A phone answer has about twenty seconds before the caller interrupts. Every answer you train should fit in two or three spoken sentences, then offer the next step.
Compare:
Website version: “Our service area covers the greater metro region including all of Davidson County and portions of Williamson, Rutherford, and Sumner counties. For addresses outside these areas, a trip fee may apply depending on distance.”
Phone version: “We cover the whole metro area and the surrounding counties. Where are you located? I can confirm we come to you.”
The phone version is shorter, ends with a question that moves things forward, and doesn’t make the caller do geography. Train every answer in this shape: the fact, then the next step.
Mark what the AI may not say
Equally important as the answers: the topics where the AI should stop and hand off. Some questions look like FAQs but carry risk. Price questions where the real answer is “it depends” should give the published range and then offer an estimate booking, not a firm quote. Medical, legal, and safety advice should never come from the AI, no matter how routine the question sounds. Insurance coverage questions should give the general policy and route specifics to staff.
For each of these, write the deflection line the AI should use. For example: “I can give you our standard range, but a firm price needs an estimate. Want me to set one up?” or “That’s something one of our team needs to answer for you. Can I connect you, or have them call you back?” These lines should be part of the training, alongside the answers.
This is also where your escalation rules connect to the FAQ work. The FAQ list defines what the AI knows; the escalation rules define where its knowledge stops.
The one-line test for every answer
After you write each answer, check it against this: could a new employee say this on their first day without getting in trouble? If the answer requires judgment, nuance, or access to something the AI can’t see, it doesn’t belong in the FAQ list. It belongs in the escalation list.
A common trap is training the AI on pricing that changes. If your prices move with material costs, seasons, or job complexity, don’t train exact numbers. Train ranges, train the factors that move the price, and train the booking step. Stale exact prices in training data are worse than no prices at all, because the AI states them with total confidence.
Update on a schedule, not on frustration
FAQs rot. You change your hours, add a service, drop a payment method, and forget to tell the AI. The fix is a standing habit: review the FAQ list once a month against your call transcripts. Look for the questions the AI answered badly or deflected unnecessarily, and fix the training. Ten minutes a month beats a quarterly overhaul.
Also review after any business change: new services, price changes, holiday schedule shifts, new staff the AI mentions by name. If your holiday scripts change, the FAQ answers about hours need to change with them.
How many FAQs do you actually need?
Start with fifteen to twenty. That covers the daily questions for most small businesses. You’ll add more from the transcripts in the first month. Don’t try to anticipate everything on day one; a tight list of real questions beats a long list of imagined ones, and the transcripts will show you what you missed.
The whole exercise takes an afternoon: collect the real questions, write ear-friendly answers, mark the off-limits topics, and set a monthly review. That afternoon is the difference between an AI receptionist that answers like someone who works there and one that sounds like it read your website once.
I write the agents that run on clinic phone lines and inboxes: the conversation engine and the booking logic behind them, plus the integrations with Pabau, Fresha and Phorest. Everything here comes out of systems we have actually shipped, not a content plan.
