Guide
What to approve before an AI receptionist answers your customers
October 30, 2026 · Written by the New Odyssey team, with AI help.
You are under a sink, on a roof or in a customer's kitchen. Your phone buzzes with a question you could answer in ten seconds, if you could pick it up.
An AI receptionist can take those questions for you. The risk is the thing every owner worries about: it says something you wouldn't. The fix is not a smarter assistant. It is a short list of things you approve first.
Decide what it may answer
Write down the questions customers ask most. For a home-service or trade business that is usually:
- What you do, and what you don't.
- Which areas you cover.
- Your hours, and how fast you usually reply.
- How you work out a price: a call-out fee, a quote after a visit, a range.
- How to book.
For each one, write the answer in the words you would use. These are the answers the assistant gives. If it isn't written down, it should not be said.
Decide what it must never guess
Some answers change by the job, the day or your diary. Prices for odd jobs, appointment times and whether you are free this afternoon are the usual three. An assistant should not make these up. It should say it doesn't know, and take a message or a booking request that you confirm.
Decide what counts as urgent
Say what should reach you at once: a burst pipe, a gas smell, a customer who is already waiting. Decide how you want to hear, and make sure anyone who asks for a person can reach you quickly, with their details and what they said.
An AI receptionist is not an emergency service. For a real emergency, the right answer sends the customer to the emergency number, not to a chat.
Be open about who is answering
People should know they are talking to an AI before the conversation starts, and know if it is being transcribed. Tell customers up front. It builds more trust than hoping nobody notices.
Test it like a customer
Before it goes live, talk to it yourself with the questions customers really ask, including the awkward ones: a same-day request, a job you don't do, a price you can't give without seeing it. If an answer is wrong, the fix is almost always in what you approved, not in the assistant.
Where Onda fits
Onda answers customers by voice or chat on your website or a link you share, using only the information you've approved. It never guesses prices, times or availability, says so when it doesn't know, and offers to take a message. People are always told it is an AI and that the conversation is transcribed, and it emails you at once for anything urgent or when someone asks for a person. Audio is never recorded. Phone answering is coming soon; today it answers on your website's "Talk to us" button and a shared link.
Questions
What should an AI receptionist never answer?
Anything you haven't approved. It should not guess prices, times or availability, and it should say so and offer to take a message when it doesn't know.
How do I test it before customers use it?
Talk to it yourself, with the questions customers really ask, including the awkward ones: prices for odd jobs, same-day requests, and things you don't do.