An AI receptionist for a medical practice can safely do the administrative half of the phone work: answer when the desk is busy, book and change appointments, pass on requests for prescriptions and referrals, and take messages. It must never give clinical advice, read out results, or decide how urgent a symptom is. Those jobs stay with your doctors and nurses, and the AI is built around that line.
This guide is for practice owners and managers. It looks at what a medical practice phone line carries, the rules the AI must work to, and the privacy questions to settle before it takes a single call. The same technology serves dental practices, but the risks are different here, and we cover that in our guide to an AI receptionist for dental practices.
What a medical practice phone line carries
A medical front desk takes a mix of calls that no other kind of business gets. Some are routine. Some are sensitive. A few could be urgent in a way the caller does not realise.
- Appointment bookings and changes. The largest group, and the most suitable for automation.
- Repeat prescription requests. Administrative to receive, clinical to approve.
- Referral and certificate requests. The same: the request is admin, the decision is not.
- Test result calls. Sensitive, and tied to identity.
- Questions about symptoms. Clinical, whatever they sound like.
- Billing and fee questions. Answerable from your published information, apart from account-specific matters.
- Carers, family and other agencies. A third party asking about a patient needs careful handling.
- Genuine emergencies. Rare, but the call that matters most.
What the AI can handle
The tasks that suit it are the ones where the answer is on your own page or in your own diary.
- Booking. Matching the appointment type to the right length, the right doctor and the right clinic location, and respecting rules such as “new patients see these doctors” or “long consults for these topics”.
- Changing and cancelling. Releasing the slot and offering the next one.
- Practice information. Hours, location, parking, accessibility, billing policy, how telehealth works, what to bring.
- Taking requests. A repeat prescription, a referral or a certificate request is written up as a task for the right person, with the patient’s details and the request in their words. The AI does not say it will be approved.
- Taking messages. Any message a clinician or the manager needs to see, with the urgency flagged.
That list would clear a large part of a front desk’s phone queue, and leave staff with the face-to-face patients in front of them.
What always goes to a person
Write this list with the principal doctor and keep it short, specific and in the AI’s rules. These categories are our suggested starting point. Your clinicians decide the final list.
- Any symptom described as new, worsening, severe or frightening.
- Any question starting with “should I”, “is it normal”, “can I take” or “do I need to”.
- Any request for test results, or for the outcome of a visit.
- Anything involving a child or a vulnerable person where the caller is worried.
- Mental health distress of any kind.
- A complaint, or a caller who is angry or crying.
- Any caller who says something that sounds like an emergency.
For an emergency, the AI tells the caller to hang up and call 000, then notifies the practice. It does not ask questions to work out how bad it is, and it does not tell the caller to wait. The wording should be written by your clinicians and tested on a recording.
Triage rules: who decides urgency
Some practices let reception staff apply a written urgency protocol. An AI receptionist is a different case, because it cannot see the caller, notice the tone of voice the way a person does, or recognise a symptom described in unusual words. For that reason, we suggest the AI does not triage at all.
What it can do is recognise that the call needs a person, and move it. The difference matters. Triage is “this is urgent, so book today”. Escalation is “I cannot decide this, so I am passing you to someone who can.” The second is the only thing the AI should say. Animal clinics have a different mix of calls, covered in our guide to an AI receptionist for veterinary clinics.
Practical design points:
- Offer a live transfer during opening hours, and a defined after-hours route.
- If nobody answers the transfer, take an urgent message to a named person and tell the caller what happens next.
- Never end a call that has been escalated without a clear next step for the caller.
- Test the route with different accents, background noise and half-sentences. People rarely describe symptoms neatly.
What an AI receptionist for a medical practice may say
An AI receptionist for a medical practice needs a firm rule on identity. A voice on a phone is not proof of who someone is. Before the AI confirms that a person is a patient, or discusses an appointment, it should ask for details you can check, such as full name and date of birth, and match them to the record.
Even after a match, restrict what is said. A common approach is that the AI confirms or books an appointment, and says nothing about the reason for a visit, a diagnosis or a result. A call from someone who says they are a partner, parent or employer asking about another patient goes to staff, not to the AI.
Privacy and consent
Medical calls contain health information. Before choosing a service, settle where the data goes. The Australian Health Practitioner Regulation Agency (Ahpra) says that practitioners remain responsible when they use AI, including administrative uses, and that they must protect confidentiality and privacy under privacy and health record laws. It also says practitioners should know whether patient data is used to train an AI tool, and whether identifiable data could end up in its learning database. It adds that public generative AI tools can make personal information publicly accessible, and that patients should be told about AI use.
Turn that into questions for any provider:
- Where are recordings, transcripts and summaries stored, in which country, and who can see them?
- Are they used to train any model? Can that be switched off in writing?
- How long are they kept, and can you delete one?
- What does the AI say at the start of the call about being automated and about recording?
- If the provider changes, what happens to the data?
Rules on recording differ across Australian states and other countries. Start with our guide to call recording laws in Australia, and confirm the position for your practice with your adviser. Our guide to AI and the Privacy Act covers the wider privacy questions.
Setting it up in a sensible order
- Record one week of calls to your desk, or ask the team to log them, and sort by type.
- Write the escalation list with the doctors, plus the after-hours route.
- Write the information the AI can use, and what it must refuse to discuss.
- Connect it to your appointment book, and test with made-up patients.
- Go live in business hours first, with a person reading the call summaries daily.
- Add after-hours cover once the daytime calls are behaving.
Watch for one failure above all: a caller who needed a person and did not get one. Read the escalated and the not-escalated calls every week for the first month.
When a simpler option is enough
If your problem is a small number of missed calls, an online booking page, a clear voicemail message and a text back may cost less and carry less risk. An AI is worth considering when volume is high enough that the desk cannot answer, or when calls arrive outside hours. Count your missed calls first.
CentriWeb builds AI phone receptionists for health practices that answer from your own information, book into your real appointment book and hand over to a person when they should. Check our guide on how an AI voice agent works before you talk to any provider.