AI & automation

AI receptionist for clinics and professional offices: what it can and can't handle

What an AI receptionist can safely do for a clinic or professional office (booking, FAQs, after-hours calls), where it must hand off, and the privacy rules.

October 21, 20267 min read

An AI receptionist can reliably handle the repetitive part of a clinic's or professional office's phone and chat traffic: booking, rescheduling and cancelling appointments, answering standard questions about hours, location, fees and preparation, taking messages, and routing callers after hours. It should not give clinical or legal advice, triage symptoms, handle emergencies beyond directing people to 911, or collect more personal and health information than the booking requires. Set up with those limits and with a clear path to a human, it can take a real load off your front desk. Set up without them, it creates privacy and liability risks your office is responsible for.

What an AI receptionist actually is

The term covers a few different tools:

Voice agents that answer your phone line, speak naturally, and can book into your scheduling software.
Chat assistants on your website or messaging channels that answer questions and take booking requests.
Hybrid services where AI answers first and a human receptionist team takes over for complex calls.

Under the hood, most use a large language model, guided by your office's information (services, hours, policies) and connected to your calendar or practice management software.

What it handles well

These are tasks where the answer is the same every time, or where the AI is following a clear script:

Booking, rescheduling and cancelling within rules you set: appointment types, durations, which practitioner, buffer times, new versus returning patients or clients.
Frequently asked questions: hours, parking, accessibility, what to bring, how to prepare for a procedure (using text you approve), accepted insurance, direct billing, cancellation policy, fees for standard services.
Message taking: name, callback number, reason for the call in a few words, urgency level according to your rules.
After-hours coverage: answering every call rather than sending people to voicemail, booking where possible, and giving your emergency instructions.
Overflow: picking up when your staff are on another line.
Reminders and confirmations, by text or call, with the option to confirm or reschedule.
Basic intake for professional offices: for a law or accounting firm, collecting the type of matter, the person's contact details and a conflict-check name list for staff to review.
Both official languages, if the system is set up and tested in French and English.

The benefit is mostly about missed calls. Every unanswered call to a clinic is a patient who may book elsewhere, and every call your receptionist takes while a patient stands at the desk is attention split two ways.

What it should not handle

This is the part vendors don't always stress.

Clinical or professional advice

An AI receptionist should never answer "Should I be worried about this symptom?", "Is this medication safe with that one?" or "Do I have a case?" Even if the model's answer sounds reasonable, it isn't licensed, it can be wrong, and your office carries the responsibility. Configure it to say clearly that it can't advise and to offer an appointment or a callback.

Triage and emergencies

Don't let the AI decide whether a situation is urgent based on symptoms. Program a fixed response for anything that sounds like an emergency: "If this is an emergency, hang up and call 911." Then, if your office has an urgent line or on-call process, give that instruction word for word.

Sensitive or complex conversations

Bad news, billing disputes, complaints, distressed callers, family members asking about someone else's care, lawyers calling about a file: these go to a person. Set clear keywords and a "talk to a human" option that always works.

Anything it can't verify

The AI shouldn't confirm test results, share information from a file, or confirm to a caller that someone is a patient or client. Identity verification over the phone is hard, and it's where privacy breaches happen.

The privacy rules you're working under

Health information is among the most sensitive personal information under Canadian law, and professional files are confidential by duty. Which law applies depends on where you are and what kind of practice you run:

PIPEDA applies to private-sector organizations in most provinces, built on ten fair information principles including consent, limiting collection, safeguards and openness.
Alberta, British Columbia and Quebec have their own private-sector privacy laws.
Ontario, New Brunswick, Nova Scotia, and Newfoundland and Labrador have health information laws recognized as substantially similar to PIPEDA. In Ontario, for example, that's the Personal Health Information Protection Act (PHIPA). If you run a clinic, check which health privacy law covers you before choosing a vendor.
Quebec's Law 25 adds specific obligations: a privacy impact assessment and a written agreement before personal information is sent outside Quebec, informing people when a decision about them is made exclusively by automated processing, and reporting confidentiality incidents that present a risk of serious injury. Penalties can reach $10 million or 2% of worldwide turnover.

The Office of the Privacy Commissioner of Canada's principles for generative AI apply directly to this use case. Among them:

Limit collection to what's needed, and don't retain prompts and transcripts longer than necessary.
Be open: tell people when they're dealing with an AI system.
Stay accountable: using an automated tool doesn't move responsibility away from your organization.
Take extra care with sensitive information. The OPC notes that in contexts like health care, consent alone may be inadequate.

Practical privacy checklist

Disclose at the start of every call or chat that the person is speaking with an automated assistant, and how to reach a person.
Collect the minimum. For booking, you usually need name, phone, date of birth (to match a record), and a short reason. Not a full symptom history.
Know where the data goes. Ask the vendor where calls, transcripts and recordings are stored and processed, which subprocessors are involved, and whether your data trains their models. If data leaves Canada, or leaves Quebec for a Quebec practice, factor that into your assessment and your privacy notice.
Set retention limits for recordings and transcripts, and make sure you can delete them.
Sign a proper agreement with the vendor covering confidentiality, security, breach notification and data return.
Update your privacy policy to describe the assistant and what it collects.
Check your regulator's guidance. Colleges and law societies often have rules on confidentiality and on supervising technology and staff; they apply to an AI receptionist the same way they apply to a human one.

What it costs

Pricing models vary. A few published examples, in US dollars:

Per-minute plans: Rosie lists plans from $49 per month for 250 minutes up to $299 per month for 2,000 minutes.
Per-call plans: Smith.ai's AI receptionist pricing lists a Pro plan at $150 per month with per-call rates around $2, and a higher tier at $500 per month.

Those are the software subscription. For a clinic or professional office, the larger cost is often the setup: connecting to your scheduling or practice management system, writing and testing the scripts, configuring French and English, and building the handoff rules. Integration with older or closed practice software can be the deciding factor; some systems simply don't allow outside booking.

How to roll it out safely

Start with after-hours and overflow. Your team still answers during the day; the AI covers the gaps. You'll see real transcripts without risking your core service.
Write the "never" list first: clinical advice, test results, emergencies, complaints, third-party information. Test every item on it.
Test with your staff playing difficult callers: confused patients, angry clients, someone who speaks mostly French, someone describing chest pain.
Review transcripts weekly for the first month. Fix answers that are wrong, vague or too long.
Measure: calls answered, bookings made, handoffs to staff, complaints. If handoffs are high, the scope is wrong.
Expand gradually to daytime overflow, reminders, and website chat once the basics are solid.

When not to use one

If your booking rules are so complex that staff themselves need judgment for most appointments.
If your practice software can't connect, and the AI would only take messages; a simple voicemail-to-text may do the job for less.
If your patients or clients are mostly elderly or vulnerable callers who are likely to find an automated system distressing. Offer an always-available human option, or don't use it on that line.

Where PNE fits

Our AI growth systems work includes AI chat and AI receptionist setups for service businesses: we map your call types, write the scripts and the "never" list, connect booking where your software allows it, configure French and English, and build the handoffs to your team. For clinics and professional offices, we keep the scope to administrative tasks and design the data collection around the privacy rules above.