The phone rings while you're halfway through a job, or with a patient, or on the M1 with no hands free. It goes to voicemail, and a fair share of callers hang up without leaving a message. That gap is what AI receptionists are built to fill. They're also oversold, so here's a plain look at what they do well, where they should hand over to a person, and what you need to sort out on disclosure and privacy before one answers your phone.
What an AI receptionist is
An AI receptionist is software that answers your business phone line and holds a normal spoken conversation with the caller. It listens, understands what they're asking in their own words, and responds with a natural voice. Behind the scenes it's connected to the things a human receptionist would use: your calendar, your customer records, your list of services and your phone system.
That's quite different from the old phone menu. "Press 1 for accounts, press 2 for bookings" makes the caller do the sorting. An AI receptionist lets them say "my hot water system's leaking and I'm in Ringwood, can someone come out this week?" and works out from there.
It is not a person, and it shouldn't pretend to be one. Hold onto that thought; it shapes almost everything below.
What AI receptionists can do well
Most of a front desk's phone work is repetitive and well defined, and that's the sweet spot. A well set-up AI receptionist can:
- Answer common questions with answers you've approved: opening hours, service areas, what you do and don't do, how payment works, where to park.
- Book, move and cancel appointments straight into the real calendar, so there's no "I'll get someone to call you back to confirm".
- Capture the details you need: name, number, address, what the job or appointment is about, and anything specific to your business, like the make of the appliance or whether the caller is a new patient.
- Log every call against the customer record with a short summary, so the next person to speak with that customer knows what was said.
- Transfer the call to the right person when it needs a human, or take a detailed message when nobody's free.
- Cover after-hours and overflow, which is often where the most enquiries are lost.
The details it captures are only valuable if they reach the people who act on them. If new enquiries need sorting into "book now" and "follow up later", our guide on how AI lead qualification turns enquiries into booked jobs covers the questions and scoring that sit behind a good phone conversation.
What they can't do, or shouldn't
An AI receptionist is a receptionist. It isn't your plumber, your physio, your accountant or your lawyer, and it should be set up so it never acts like one.
| Situation | What the AI should do |
|---|---|
| Caller wants advice (clinical, legal, financial, safety) | Decline politely, capture the question, book a time with a qualified person |
| Caller is upset or making a complaint | Acknowledge it, take the details, transfer or arrange a call from a named person |
| Caller describes an emergency | Tell them to call Triple Zero (000) if anyone is in danger, then follow your urgent-job process |
| Caller asks for a human | Transfer, or take a message with a clear callback time, without arguing |
| Caller wants a firm price for a complex job | Give only price ranges you've approved, and book an assessment |
Here's my view, and some vendors won't like it. An AI receptionist that makes it hard to reach a human is worse than voicemail. If a caller says "can I just talk to someone?" and the software tries to keep them in the conversation, you've saved a few minutes of staff time and lost the customer's goodwill. Build the escape hatch in from day one.
They also have ordinary technical limits. A bad mobile line, heavy wind on a building site or two people talking at once can trip up speech recognition, just as it trips up people. Good systems ask the caller to repeat themselves or confirm key details back, such as the phone number and address, rather than guessing.
Telling callers they're speaking with an AI
This is the easy one to get right, and the most damaging to get wrong. People reasonably expect to know whether they're talking to a person or software.
The government's business.gov.au guidance on using AI in your business puts it plainly: a customer probably doesn't care that you used AI to help write a social media post, but they want to know when they're talking to an AI rather than a human for customer support. The OAIC's guidance on privacy and commercially available AI products says public-facing AI tools such as chatbots should be clearly identified as such, and that privacy policies should explain how a business uses AI.
In practice, the receptionist says it at the start of the call in one short, friendly line, something like "Hi, you've reached the team at [business name]. I'm an AI assistant, and I can book appointments, take details or put you through to someone." That's enough. It doesn't need a disclaimer speech.
In the receptionists we build, that disclosure is always on and can't be switched off. I'd encourage you to treat it as non-negotiable whatever system you use.
Call recording, transcripts and privacy
An AI receptionist works by turning speech into text, so most systems produce a transcript of every call, and many keep an audio recording too. That raises legal questions you should look at before going live. What follows is general information, not legal advice.
Recording a call. In Australia, recording and intercepting phone calls is covered by Commonwealth law, including the Telecommunications (Interception and Access) Act 1979, and states and territories have their own listening or surveillance devices laws, which aren't identical. We won't summarise those rules here because the details matter and they differ by jurisdiction. The common, sensible practice is to tell callers at the start that the call is recorded or transcribed, and why. If you're unsure what applies to you, check the legislation that covers your state or territory or ask a lawyer.
Privacy law. A transcript with a caller's name, number, address and the reason they called is personal information. Whether the Privacy Act 1988 applies to your business depends on a few things. The OAIC's page on privacy obligations for small businesses explains that most businesses with an annual turnover of $3 million or less aren't covered, but some are regardless of turnover, including health service providers. That catches a lot of clinics, so if you run one, assume the Australian Privacy Principles apply and check the detail.
Either way, decide these things before launch:
- What is recorded or transcribed, and what callers are told about it.
- How long recordings and transcripts are kept, and who in your team can access them.
- Which AI providers process the calls, where the data is hosted, and whether it's used to train anyone's models.
- How a customer can ask to see or correct what you hold about them.
Ask any supplier those questions in writing. Our clients are told which providers process their data and where it's hosted, and it's never used to train those providers' models. We cover the wider topic in AI, privacy and customer data for Australian businesses.
How to test an AI receptionist before going live
Never switch one on for real customers until your own team has tried to break it. A demo shows you the happy path; testing shows you everything else. Have three or four people make calls like these:
- The easy booking. A standard appointment request. Check it lands in the right calendar, at the right time, with the right details.
- The vague caller. "Yeah, it's making a noise." Does it ask sensible follow-up questions or just guess?
- The noisy line. Call from a car, a windy street or a busy café. Does it confirm the phone number back correctly?
- The advice seeker. Ask it whether you should be worried about a symptom, a leak or a tax bill. It should decline and book you with a person.
- The upset customer. Complain about a past job. It should stay calm, take details and hand over.
- The "put me through" request. Ask for a human early. It should transfer or take a message, first time.
- The failed transfer. Make sure nobody answers the transfer. What happens to the caller then?
- The off-script question. Ask for a price you haven't approved, or about a service you don't offer. It should not make something up.
After each call, read the transcript and the summary it logged. The summary is what your staff will actually rely on, so it needs to be accurate and short. Then fix the instructions, and run the same calls again. Keep the list; it becomes your regression test every time you change what the receptionist knows.
Rolling it out without annoying your regulars
Start narrow. After-hours and overflow calls are the obvious first step: the AI only answers when nobody else can, so the downside is small and you see real calls quickly. Once the transcripts look right for a few weeks, widen it.
Tell your regular customers it's coming. A line on your website and in your next email is plenty. And keep a person reading a sample of transcripts every week, at least early on; it's the cheapest way to catch a wrong answer before it becomes a habit.
The receptionist also needs to talk to your other systems properly. If it books into one calendar while staff use another, or logs calls somewhere nobody looks, you've made more admin, not less. That's why we connect the phone system, calendar and CRM as part of the job, which our system integrations work covers, and why the AI receptionist we build writes straight into the same customer records your team uses.
Done carefully, an AI receptionist means fewer missed calls and fewer scribbled messages, and your staff get to spend their phone time on the conversations that need a person. If you'd like to work out whether it suits your business, book a call with our team and we'll go through how your calls come in today.
— Craig Winder, Founder
