The practice phone rings while the doctor is with a patient and the receptionist is checking someone in at the counter. Nobody answers. In most cases the caller does not try again: they ring the next number in the search results. That is what pushes practices to ask whether an artificial intelligence system could answer the phone in place of a person.
The honest answer is: partly yes, and precisely where it matters most. But the line between what a virtual receptionist may do and what it must never do, in healthcare, is sharp and not up for negotiation. This article describes it plainly, because it is the only solid basis for a decision.
What a virtual receptionist is
A virtual receptionist — often searched for as an "AI receptionist" — is a system that answers calls, and frequently messages, with a synthetic voice. It understands natural language, holds a simple conversation and carries out concrete actions: taking a request for an appointment, giving a piece of information, capturing a contact, handing the call to a person.
It is not the touch-tone menu — "press 1 for appointments" — that routes without understanding anything, and it is not a chat widget on the website. It is a voice counterpart that handles the predictable part of the conversation and recognises when to stop.
None of these systems is indistinguishable from a person in a complex conversation: on straightforward requests they work well, with an anxious caller telling three stories at once much less so. The difference between a successful deployment and a disastrous one lies in the scope assigned to the system, not in the technology.
What it can do
The tasks it handles reliably are logistical and repetitive — not by coincidence, the ones that eat most of a receptionist's day.
- Answer out of hours. Evenings, the lunch break, weekends, public holidays: the use case with the clearest value, because today nobody picks those calls up at all.
- Answer when the line is busy. The second and third simultaneous caller hear an engaged tone.
- Take a request for an appointment. Name, contact details, the kind of appointment in general terms, rough availability. The front desk then calls back, or the system offers open slots if it is connected to the diary.
- Give fixed, verifiable information. Address, opening hours, how to reach the practice, what to bring, whether a referral is needed.
- Record cancellations. One received at nine in the evening lets the slot be reallocated the next morning.
- Filter, route and leave a record. Every call becomes a record with the request, the contact and its status, so the queue is ready when the practice opens.
The value sits here: requests that dissolve into nothing become written contacts somebody can work on.
What it must never do
This list matters more than the previous one, and it admits no exceptions "just for the simple cases".
No clinical triage. An automated system does not judge how serious a symptom is, does not decide whether a patient should be seen today or in a fortnight, and does not set clinical priorities. Triage is a professional act.
No medical advice. Nothing on medicines, doses, stopping a treatment, or interpreting symptoms or test results. Not even the "general" version, which is still advice.
No handling of emergencies. The system has to recognise signs of urgency and react in exactly one way: point immediately to the local emergency number and the correct channels. This is the first behaviour to test before it goes live, not the last.
No clinical data captured by voice. The receptionist records that a person is asking for an appointment, not why. The right question is "what type of appointment do you need", never "what are your symptoms".
No pretending. The system has to state that it is an automated assistant. Passing itself off as a person is unfair to the patient, and in the European Union transparency about interacting with an AI system is also becoming a legal obligation under the AI Act.
No total replacement of the front desk. There has to be a fast route to a person at all times, asked for by the patient or triggered when the conversation leaves the expected path. A system that cannot hand over is a wall.
That is exactly the scope we built Claudia AI, the virtual receptionist for medical practices around: logistics yes, clinical never. It is not commercial caution, it is the condition under which a tool like this belongs in a practice at all.
Privacy and compliance: what to check
A virtual receptionist processes personal data on behalf of the practice. Under the GDPR the vendor is therefore a processor, and a written agreement under Article 28 is required. Before signing, these are the questions to ask.
- Where is the data processed and stored? Servers inside the European Union is the answer that makes life simpler. Where there are transfers outside the EU, the safeguards relied on have to be identified and documented.
- Are calls recorded? If so, the caller has to be told at the start, a legal basis defined and a retention period set. Many practices keep only the structured transcript of the logistical data, which is lighter and easier to defend.
- Does the data feed the vendor's model training? The acceptable answer is no, and it belongs in the contract.
- Who can access the conversations, and what happens if the service goes down. Roles and access logs on one side; a fallback route on the other, because calls must land on a real number, not in a void.
The privacy notice and the record of processing activities have to be updated accordingly. Where the platform is hosted outside Europe, the questions to put to the vendor are covered in the article on GoHighLevel and GDPR in a European medical practice.
How to introduce it without causing damage
The most common mistake is switching the system on for every call from day one. A sensible rollout is gradual: out of hours only to begin with, where the risk is minimal because the current alternative is nothing at all; then overflow, so the system answers only when the line is busy or after a set number of rings; then a selective widening, for instance logistical information at peak times.
Before each step, test by hand: call in and try the ordinary request, the confused one, the angry one and, above all, the urgent one. If the simulated emergency does not produce an immediate referral to the emergency number, the system does not go live.
Then decide what happens afterwards: who reviews the requests, within what time they are called back, where the contacts end up. A system that gathers twenty requests nobody returns produces twenty more disappointed patients than before. Collected requests belong in the same flow as every other channel, described in the article on patient communication across email, SMS and WhatsApp.
What it costs, and when it makes sense
Pricing is usually a subscription plus a usage component on minutes, with a setup cost on top. Figures vary widely and change constantly, so ask for them on the same parameters from at least two vendors: monthly fee, cost per minute, minutes included, what happens above the threshold, setup, contract term.
The sensible comparison is not "AI versus receptionist", which is a false choice, but the cost of the system against the value of the calls that currently go unanswered — a number to measure first, as described in the article on what a missed call really costs a medical practice. No serious vendor can promise how many extra patients will arrive.
Frequently asked questions
Can a virtual receptionist replace the receptionist?
No, and that is not the right goal. It covers the calls nobody picks up today and gives time back to the person for the work that needs judgement and presence.
Can it tell a patient whether a symptom is urgent?
No. Triage is a professional act and stays outside the scope. The system has to recognise signals of urgency and refer immediately to the emergency number, without assessing anything.
How do we know whether our practice needs one?
Start from real numbers: calls received, calls missed, and when they are lost. If you would like us to look at them with you, request an assessment with no commitment. If you would rather get a general picture first, there is the free guide to medical marketing.