There are now more than two dozen AI receptionists marketing themselves to dental practices. They demo well. They all promise to "handle every call, 24/7." Most of them, on closer inspection, are either generic call-answering bots with a dental skin, or scheduling widgets that hand off to a human the moment something interesting happens. This guide is the checklist we wish every practice owner had before they signed a contract.
What an AI receptionist actually has to do
A dental front desk is not a phone tree. It is a scheduling engine bolted to a clinical workflow bolted to an insurance verification queue. Anything that calls itself an "AI receptionist" but only does one of those three things is not replacing your receptionist — it is augmenting your hold music. A real AI receptionist for dental should be able to:
- Answer the phone in a natural voice, in your practice's brand and tone.
- Identify whether the caller is a new patient or existing — by matching name, date of birth, and phone number against your chart database.
- Honor provider, operatory, and appointment-type rules when offering a slot.
- Write the booking back into your practice management system (PMS) — not just take a message for staff to enter later.
- Triage emergencies to a real human when the situation requires it.
- Handle the call in the caller's language — or at minimum, in Spanish.
Anything less than that is a fancy answering service, and you should price it like one — not like a front-desk hire replacement.
The seven criteria that actually matter
1. PMS integration depth (write-back vs. message-taking)
This is the single most important question and the one most vendors will dance around. Ask: does your AI write the appointment directly into my PMS, in real time, or does it generate a message that my staff has to process the next morning?
Message-taking AI is fine — it's a glorified voicemail-with-transcript. But it is not a receptionist. It does not save your team time; it shifts the data-entry work to the morning. A real AI receptionist talks to Dentrix, Eaglesoft, Open Dental, Curve, Denticon, or CareStack the same way a human would, opens the chart, checks the provider's column, and drops the appointment in. By the time you walk in the next morning, the chart is ready and the operatory is blocked.
If the vendor cannot demo a live booking flow into your PMS in under five minutes, treat that as a hard no.
2. Scheduling rule fidelity
A dental schedule has more rules than a calendar can hold. Provider hours, operatory assignment, hygiene vs. doctor slots, age-based appointment types, time-block requirements for crowns and restorative, family booking, same-provider rescheduling rules, recall intervals — and that's before you get to insurance restrictions. A receptionist who has worked at your practice for two years has all of this in their head. The question is whether the AI can hold it too.
During the demo, give the vendor a real edge case from your office. Something like: "A new patient calls in for cleaning. They mention their five-year-old needs to be seen too. Walk me through what your AI does." If the answer involves "we'll route that to your team," that's a real limitation — family booking is one of the highest-frequency call types in dental.
3. After-hours and overflow behavior
Most missed-call revenue lives in two windows: the lunch hour and after-hours. The first is overflow — your team is on the phone, at the counter, or eating. The second is everything between 5 PM and 9 AM, plus weekends and holidays. Ask the vendor specifically how their system behaves in both. Does it pick up simultaneously with your line ringing once, after three rings, or only when forwarded? Can it stay on the line with the patient for as long as the call takes, even if that's eight minutes for a new-patient intake? Does it handle emergency calls differently after-hours than during business hours?
A weak answer here means the AI will look great in the demo and quietly miss the same calls your front desk does.
4. HIPAA, BAA, and where your data lives
Any AI handling patient calls is a Business Associate under HIPAA. You need a signed BAA. Period. If the sales team has to "check with legal" about whether they can provide one, the answer is no. Ask:
- Do you sign a BAA at the standard tier, or only at enterprise?
- Where are call recordings stored, and for how long?
- Is PII redacted from transcripts before they're used for model training?
- What happens to my data on cancellation?
You don't need the vendor to be a security-first organization. You need them to answer these four questions without flinching.
5. Language coverage
English-only AI receptionists rule out a meaningful share of the calls coming into a US dental practice — in some markets, more than 30% of inbound calls. Spanish is non-negotiable. Languages that matter beyond Spanish depend heavily on your market: Vietnamese, Mandarin, Portuguese, Russian, French, Tagalog. Ask whether the AI detects the caller's language automatically or whether the patient has to press 2.
6. Pricing transparency
Headline price is rarely the real price. Build a total-cost-of-ownership figure that includes the monthly base, per-minute overages, PMS integration fees, onboarding charges, contract minimums, and any add-on costs for SMS, after-hours, or outbound calls. Most "$300/month" plans we've seen come in at $450–600/month all-in after the first ninety days.
Flat-rate, unlimited-call pricing exists in this market. Reward vendors who offer it.
7. Time to live
Setup time is a feature. The best AI receptionists are configured to your practice in a few days — your hours, services, providers, scheduling rules, the way you want callers handled. The worst take six weeks and require multiple training sessions with your team. Ask for an honest timeline, and ask what the vendor's job is in week one vs. what your office has to do. A forward-deployed engineer who tunes the system after launch is worth more than a slick onboarding portal.
Red flags that should kill a deal
- "We integrate with your PMS" — but the integration is a webhook that emails appointment requests to your front desk. That is not integration. That is a contact form.
- Per-minute pricing that uncaps overage — a busy month will cost you double.
- Multi-year contracts with auto-renew clauses — if the product is good, the vendor doesn't need to lock you in.
- No real demo of an actual call — if you can't hear the AI talking to a caller for at least 60 seconds on the demo, you don't know what you're buying.
- Vague answers about HIPAA — covered above. Walk.
How to pilot one without disrupting your office
The lowest-risk pilot is to start with overflow and after-hours coverage only. Forward calls to the AI when your line is busy or your office is closed — and only those. Your team's day doesn't change. The AI's only job for the first thirty days is to catch the calls you were already losing. Run that for a month, look at the bookings the AI closed, and decide whether you want to expand it to full daytime coverage.
Most practices that go this route find they recover three to five new-patient bookings in the first month — enough to pay for the service several times over. Whether you keep the AI on after that is a question of how much of the work you actually want to hand over.
The honest TL;DR
Pick an AI receptionist that (1) writes into your PMS, (2) honors your scheduling rules, (3) covers after-hours and overflow as the headline use case, (4) signs a BAA at the standard tier, (5) speaks Spanish, (6) tells you exactly what it costs, and (7) is live in your office in a week. Anything else is marketing.
Lila is built to all seven of these criteria. If you want to hear what a real call sounds like, the demo on our homepage is an unedited recording. We're happy to walk through any of the questions above on a call — book a demo and bring your toughest scheduling edge case.