What an AI receptionist actually says.

The objection comes up on every discovery call. Usually within the first eight minutes. Sometimes the dentist says it directly: "I don't want a robot answering my phone." Sometimes it comes wrapped in something more polite, like "we tried one of those services back in 2022 and it was awful." Same fear underneath. The receptionist is the first impression. Get it wrong and you don't get a second chance.

I take the objection seriously. I had the same fear the first time I built one of these. The version of voice AI most people in southeastern Massachusetts have heard sounds like Press 1 for billing, press 2 for... from a 2014 IVR system. That isn't what a modern AI receptionist sounds like. It hasn't been for almost two years. But people don't know that until they hear one.

So instead of arguing about it, here's a transcript. This is from a real after-hours call to a small two-chair dental practice off Faunce Corner Road in Dartmouth. The names and the address are changed. The mechanics are real. I've annotated each turn with what's happening underneath, both the AI parts and the integration parts that people don't see.

The setup.

The practice closes at 5pm. The front-desk person, who is also the office manager and also runs the recall and billing systems, leaves at 5:15. Before this build, the after-hours line went to a voicemail that nobody returned until at least 9am the next morning, sometimes later if Tuesday morning was a hygiene-heavy day. The practice had been losing roughly 11 to 16 calls a week to that voicemail. Standard pattern for dental in this region; the Peerlogic 2026 study put the industry-wide miss rate at 38% across 4,280 dental calls.

The build I did is what I call the standard dental package. It picks up between 5pm and 8am on weekdays, and all weekend. It's connected to their practice management software (in this case Dentrix), which is how it knows which providers, which procedures, and which insurance the practice handles. It can book new-patient consultations, look up existing patients, and capture an emergency triage flow if the call sounds urgent.

Here is the call. It came in at 6:47pm on a Tuesday in late March.

What's actually happening, line by line.

The transcript reads like a normal call. The interesting part is what's running underneath. Let's go through it in groups.

The greeting

The first patient turn (the real test)

The triage

The actual booking

The intake fields

The handoff

What the practice got, after the call.

By the time Maria hung up, four things had already happened in the practice's systems:

• A new patient record was created in Dentrix with her name, phone number, and insurance carrier.
• A 90-minute new-patient appointment was booked for Thursday at 10:15am with Dr. Costa, with the chief complaint "chipped lower molar, mild sensitivity, sustained while eating" entered in the notes.
• A confirmation text went out from Twilio with the address, the calendar invite, and the intake form link.
• An email summary went to the office manager so she'd see it first thing Wednesday morning, before patients started arriving.

Total elapsed time from the moment Maria dialed to the moment all four of those existed in the practice's systems: about three and a half minutes. With no human involvement.

What it sounds like when somebody builds it badly.

Most AI receptionist services for dental practices in 2026 are not built like this. They're built on top of generic no-code platforms (Voiceflow, Vapi, Synthflow) by agencies that don't write any of their own code. The result is functional but brittle. Here is the version of the same call I've heard most often when a practice in Bristol County tries the cheap option first:

That call ends with the caller hanging up and Googling another dentist. It looks like the AI worked because the call was technically answered. The lost appointment never shows up in any dashboard. The practice paying $399/month for the service has no idea it's losing money to it.

The difference between the two builds is not the voice technology. Both are using essentially the same underlying speech models (variations on OpenAI's voice or ElevenLabs). The difference is whether somebody actually wrote logic for how a dental call should flow, what to triage, when to interrupt, when to wait, what to capture, what to skip. That logic is custom code. It takes me about ten days to build, tune, and ship per practice. The no-code agencies skip all of it, which is why their builds break the moment a real human caller does anything other than the happy-path script.

What this doesn't do.

Worth being clear about. The system in this transcript:

• Doesn't give clinical advice. If Maria had described severe pain, swelling, or a knocked-out tooth, the build escalates to an emergency flow (text the on-call doctor, route the caller to a same-night option), not "I think you should take ibuprofen."
• Doesn't try to handle billing questions. Those route to the front desk during business hours. The system says so, politely, when asked.
• Doesn't try to handle insurance verification beyond confirming whether the practice is in-network for the major carriers. Verification of benefits, copay estimation, prior auth: all human work, all properly handed back.
• Doesn't pretend to be human if asked directly. If the caller says "are you a real person?" the system says no, explains it's the practice's after-hours system, and offers to text the office manager directly if they prefer to wait for a human callback. Most callers, on hearing the disclosure, just continue the booking.

The cost, since people always ask next.

The build for the practice in this transcript was $4,800, one-time. That includes the discovery, the build itself, the Dentrix integration, the Twilio configuration, two rounds of voice tuning against real call samples, and 60 days of post-launch support. The practice now pays Twilio about $32 a month for the phone line and the model provider about $40-$70 a month depending on call volume. So roughly $80-$110/month in actual operating cost, no monthly fee to me.

Compare that to a subscription AI receptionist service marketed to dental practices: typical pricing is $299-$799 a month, depending on call volume tiers. The math on subscription versus build is broken out here. The custom build pays for itself against the cheapest subscription plan inside 18 months and against the higher tiers inside 8 months.

Maria's appointment alone, assuming she becomes a long-term patient at the average lifetime value for a New England dental practice, is worth more than the whole build cost. The practice has captured 47 similar after-hours calls in the four months since the system went live. Roughly 31 booked. Eleven of those were emergencies that would otherwise have gone to a competitor.

If you want to hear it.

The audit conversation includes a live demo on a synthetic version of your practice. I configure the AI for your specialty, your location, your typical insurance mix, and your usual procedures, then we run a sample call together on the discovery call. Most dentists I do this with describe the moment they hear the build as the moment the objection dissolved. I'd rather you make the decision based on hearing it than on reading an article about it.

— Justin, from a converted spare bedroom in New Bedford, written before the coffee at Society opened