AI receptionists for medical practices: the complete guide

By Umer Farooq, founder of Code Khalaq · Published 7 September 2026 · 14 min read

What does an AI receptionist actually do on a patient call?

An AI receptionist is software that answers your practice's phone line, holds a natural conversation with the caller, and completes the routine tasks a front desk handles by phone. In a typical call it greets the patient, works out why they are calling, and then does one of four things: books or reschedules an appointment, answers a practical question such as hours or parking, takes a structured message with a callback promise, or transfers the call to a person.

The transfer rule is the part that separates a well-built system from a demo. A good AI receptionist is designed to recognise the moments it should not handle: symptoms, medication questions, billing disputes, and callers who are distressed. Those go to your staff during office hours and to a message queue or on-call line after hours. It never improvises clinical advice.

What it collects is deliberately small. Booking an appointment needs a name, a date of birth, a callback number, and the type of visit. It does not need a medical history. Systems built on this "minimum necessary" principle are easier to trust and easier to keep compliant, a point we return to in the HIPAA section.

The practical difference for your team is that the phone stops interrupting them. Front desk staff are usually doing three things at once: checking a patient in, answering a question from the back office, and hearing the phone ring. Something has to give, and it is almost always the phone.

Why are practices looking at this now?

Because the numbers on missed calls are worse than most owners assume. Industry analysis puts average missed-call rates at 23 to 35% across medical practices, with solo practices missing 30% or more. One study of 7,000 calls across 22 medical practices found that 42% of inbound calls went unanswered.

The second number is the one that hurts. Research from AnswerNet found that 85% of patients who cannot reach a practice on the first call do not call back. New patients in particular simply call the next practice on their list. Missed calls cluster at lunch, after 5pm, and during the busiest in-office hours, which is exactly when the caller is most likely to be someone new.

Put those together with what a patient is worth over time and the cost becomes visible. For dental practices, a new patient's lifetime value is commonly estimated at around $10,000, with published estimates ranging from $6,700 to $22,000. Missing even a couple of new-patient calls a month compounds into six figures of lost lifetime value a year.

AI receptionist vs answering service: which fits your practice?

Human answering services have been the default overflow option for decades, so this is the comparison most owners start with. The two differ on four things: what happens on the call, hours, cost model, and system access.

Neither option is a clinician, and both should escalate anything about symptoms or medications. The real gap is the message-only model. If a caller wanted an appointment and could not book one, the message your staff reads tomorrow morning is often about a patient who has already gone elsewhere.

How much does an AI receptionist cost compared with an answering service?

It depends on call volume, and that dependency is the whole story. Answering services commonly bill per minute of operator time, including hold and message taking, usually rounded up in increments. Quiet practices pay little and busy ones pay a lot, and the bill moves every month.

AI receptionists typically charge a flat monthly subscription with a stated minutes allowance. That is easier to budget and to audit, and it usually works out cheaper as volume grows. The things to check are the one-time setup fee, the overage rate once you exceed bundled minutes, and any integration cost for your scheduling system. Because pricing is flat and published there are fewer variables, but always ask for the full rate sheet in writing.

A third option is hiring. A part-time front desk hire covers office hours only, still misses lunch and peak times, and costs more than either service once you include onboarding and cover. Most practices we talk to are not choosing between AI and a person. They are choosing what to do with the calls their person cannot reach.

How does an AI receptionist book into your PMS or EHR?

This is the section to read twice, because integration depth is where most disappointing rollouts go wrong. There are two ways a phone agent can "book an appointment", and they behave very differently.

Direct integration means the vendor has built an API connection to your specific system, whether that is Dentrix, Open Dental, athenahealth, or another PMS or EHR. The agent reads live availability at the moment of the call and writes the appointment into your schedule with the right provider and appointment type. When your front desk fills a slot, the agent sees it immediately.

Calendar-layer booking means the agent works from a synced copy of your schedule that refreshes on a delay. It is quicker to set up and can be a reasonable start for a solo practice with simple scheduling. The risk is stale availability: a slot your staff just filled still looks open, and the agent books it. That is where almost every double-booking complaint comes from.

If your practice management system is old or has no API, you still have options. Vendors with engineering capacity can build a bridge to the system's database, a sync utility on your office server, or a fast confirmation queue for your staff. Ask the vendor to scope your specific system before you commit, and ask to watch a live booking rather than a slide.

What does HIPAA require from an AI receptionist?

When a patient says "this is Maria Lopez, I need to reschedule my root canal", that sentence is protected health information. It connects an identifiable person to their care. Any company that receives, stores, or transmits PHI on your behalf is a business associate, and HIPAA requires a signed business associate agreement, a BAA, before that vendor touches patient information. No BAA, no deal.

Three things follow from that. First, "HIPAA certified" is a red flag phrase. HHS does not certify products or vendors and no such certification exists. A careful vendor says so, signs a BAA, and explains its actual safeguards: encryption in transit and at rest, access controls, audit logs, and a clear retention policy. Second, call recordings and transcripts are usually PHI themselves, so ask whether they are stored, where, for how long, and whether you can shorten retention. Third, your practice stays the covered entity. A BAA shares responsibility; it does not transfer it. If a vendor has an incident they report it to you, and you are responsible for notifying affected patients.

The questions to put in writing to any vendor: Do you store audio and transcripts, and for how long? Where is data stored and which subcontractors touch it? Is everything encrypted in transit and at rest? Will you sign a BAA, and at which pricing tier? Is patient data used to train your models? Do you keep audit logs? What happens when a caller asks a clinical question?

Can an AI receptionist reduce no-shows?

Partly, and only as one piece of a system. No-shows fall when three things are in place: a confirmation request early enough to matter, an effortless way to reschedule, and a plan for refilling the slot when someone cancels. An AI receptionist contributes to all three because it can make and take those calls at scale, but the design of the reminder ladder matters more than the technology.

A practical pattern is three touches: a confirmation request about 48 hours out, a detailed reminder at 24 hours, and a short nudge about two hours before. At least one should ask for a reply, so silence flags the appointment for follow-up while there is still time to backfill from a waitlist. Fees rarely prevent no-shows because they act after the patient has already decided to miss; if you keep one, pair it with reminders so nobody is surprised by it.

Measure it consistently. Divide appointments missed without notice by total scheduled appointments in the period, decide up front whether late cancellations count, and apply the same rule every month. There is no universal benchmark; your own trend, segmented by provider and appointment type, is the useful number.

What is different for dental practices?

Dental front desks run on a different rhythm: recall reminders, hygiene appointments booked six months out, treatment plan conversations that should never be automated, and practice management systems such as Dentrix and Open Dental that a vendor either integrates with directly or does not. The lifetime value of a new dental patient, commonly estimated around $10,000, also makes each missed new-patient call more expensive than in many medical settings.

What can be automated: call answering around the clock, booking directly into the PMS, recall reminders, confirmations, no-show recovery, and basic new-patient intake. What should stay human: treatment plans, billing disputes, clinical concerns, and upset patients. Most practices redeploy the freed-up time rather than cut roles, because the phone was the thing the team could never keep up with.

Judge results on three numbers with baselines recorded before you start: answered-call rate, which should improve within a week or two; booked-appointment rate, which becomes meaningful once booking is switched on; and no-show rate, which takes a couple of months to trend clearly.

How do you evaluate an AI receptionist vendor?

Most vendor demos are built to impress in five minutes. Your evaluation should be built to find out what happens in the sixth. The single most important question is what happens when the AI cannot handle a call. The safest systems transfer to your staff during office hours, take a structured message with a callback promise after hours, and never attempt clinical conversations. A vendor who claims the AI handles everything has told you what to expect.

Beyond that, the questions cluster into four groups. Handoffs: when does it transfer, to whom, and what does the caller hear? Booking: does it write into your actual PMS or EHR, and can you watch a live booking land in your real calendar? Compliance: will they sign a BAA, at which tier, and how are recordings handled? Commercials: what is a billable minute, what is the overage rate, what is the setup fee, and what does a fair pilot look like?

Get every answer in writing. A vendor who welcomes the list is one you can probably work with. One who waves it off has shown you their support process before you signed anything.

What does a safe rollout look like?

Do not switch your main number on day one. Run a test line first for a few weeks: script scenarios from real calls, have staff phone in at peak times, and review every transcript. Then forward after-hours calls, which are the lowest-risk and highest-value segment. Then add overflow calls during office hours. Switch the main line only once bookings are landing correctly in your real calendar and your staff trust the handoffs.

1. Baseline. Record answered-call rate, booked-appointment rate, and no-show rate for the previous month.
2. Test line. Two to three weeks of scripted calls and transcript review. Fix the scripts, not the staff.
3. After-hours. Forward evenings and weekends. Review the morning message queue daily.
4. Overflow. Forward calls that ring more than a set number of times during office hours.
5. Main line. Only when bookings, transfers, and messages have been clean for at least two weeks.
6. Review. Compare the three numbers against baseline monthly, and add the AI receptionist to your next security risk analysis.

That sequence is how we onboard practices onto Pulse, and it is how we would recommend rolling out any vendor. The goal is not to prove the technology works. It is to find the edge cases with your own calls before a patient does.