Before hiring an AI receptionist, a dental practice should ask six things: where patient information goes, how bookings reach your schedule, what happens on emergencies and insurance questions, how outbound reminders stay legal, what the contract lets you take with you, and how pricing behaves when call volume spikes. A vendor who answers these in specific, checkable terms is worth a trial. A vendor who answers with adjectives is not.
Where does patient information go, and will they sign a BAA?
Ask this first, because dental calls contain health information the moment a caller says "my crown fell off" and gives a name. Under, a vendor that handles protected health information for a covered entity is generally a business associate, and the covered entity needs a written agreement with it. [HHS publishes sample business associate agreement provisions](https://www.hhs.gov/ /for-professionals/covered-entities/sample-business-associate-agreement-provisions/index.html) you can use as a checklist.
Be direct about it: NovaReps does not claim compliance, and you should not accept that claim from any vendor on a sales page alone. Ask for the agreement in writing, ask which subprocessors (the speech, language-model, and telephony providers) touch call audio and transcripts, and have your compliance lead or counsel review it. If a practice requires a BAA and the vendor can't provide one, the conversation ends there.
Then ask the practical questions: can you export call records, can you request deletion, and is there an audit trail showing who accessed personally identifiable information? Those are mechanisms you can verify during a trial, unlike a compliance badge.
How do bookings actually reach our schedule?
"Integrates with your calendar" can mean very different things. The honest version of the question is: does the agent read live availability during the call, or does it collect a request that someone re-enters later?
NovaReps books against connected Google or Microsoft calendars in real time and confirms a slot before the caller hangs up. That works well if your front desk schedules from a calendar. It is not the same as writing directly into a dental practice management system, and you should ask any vendor, including us, exactly how the appointment lands in the system your hygienists and doctors work from. If the answer involves a second system your staff must reconcile, price that labor into the decision.
Also ask about the rules the agent applies. Can you set different appointment lengths for a new-patient exam versus a cleaning versus a filling? Can you block time for emergencies? A receptionist that books a 30-minute slot for a 90-minute procedure creates more work than it saves.
What happens with emergencies, insurance questions, and anything clinical?
This is where AI receptionists differ most in practice. Good setups define, in writing, what the agent will not handle. Test these with real scenarios before signing:
- A caller says they have facial swelling or are bleeding after an extraction. Does the agent offer a callback, route to the on-call number, or tell them to seek urgent care, per your protocol?
- A caller asks "do you take my insurance?" Can the agent answer from a list you supply, or does it guess?
- A caller asks whether a symptom is serious. The correct behavior is to decline clinical advice and hand off.
Call the demo line yourself and try to make it fail. Ask it something you would never want an AI to answer. The way it declines tells you more than any feature list. For the general handoff pattern, see how an AI receptionist works.
How is this different from the answering service we already use?
Answering services take messages; most can't check a calendar mid-call. If you're comparing, ask the AI vendor to show a call where an appointment was booked and confirmed end to end, then ask your current service whether it can do the same. AI receptionist vs answering service: what's actually different walks through that comparison in detail.
Also ask what happens when the AI can't resolve a call. A message that lands in your inbox with a transcript is acceptable. A dropped caller is not.
How do reminders and missed-call text-back stay legal?
Reminders and missed-call text-back are where AI receptionists often earn their keep in a dental office, since no-shows and unanswered calls both cost chair time. They also involve outbound contact, which has rules. The FCC has ruled that AI-generated voices count as "artificial" under the TCPA, which means consent matters for automated calls.
Ask how consent is captured and recorded, what the calling-hours and opt-out controls are, and whether there is a safety layer that blocks calls you shouldn't be making. NovaReps ships an outbound safety stack for exactly this reason, and you can read how it works in what an outbound safety governor is. Whatever vendor you choose, make them explain the mechanism rather than say "we're compliant."
Who is accountable, and whose brand is on it?
Many dental practices buy this through a marketing agency or IT provider rather than directly, because the agency already manages their website and phones. That's a legitimate model: agencies can resell a white-label platform under their own name. But it adds a question for you: who do you call at 8 a.m. Monday when the agent mishears a patient's name? Ask who configures the agent, who reviews call transcripts, and what the response time is for fixes.
What does the contract let us take with us?
Ask what you keep if you leave: call recordings, transcripts, the agent's prompt and knowledge base, and your phone numbers. Don't assume is available; confirm it in writing with whichever vendor you pick, because we don't make that promise on your behalf.
Also ask about the underlying providers. Platforms that let you bring your own API keys for the language model or voice, as NovaReps does with BYOK, can reduce lock-in and make costs more transparent. Platforms that bundle everything hide that.
How does pricing behave when the phones get busy?
Monday mornings and the days after a holiday are your peak. Ask whether pricing is per minute, per call, or a flat tier, and what happens to calls above a concurrency limit. A cheap plan that sends callers to busy signals at 9 a.m. defeats the purpose. Get the full cost for a realistic month, including overages, before signing. For our ranges and structure, see how much an AI receptionist costs and the pricing page.
Is there a short test that tells us if it's ready?
Run a one-week pilot with a defined script of ten calls: three routine bookings, two reschedules, one insurance question, one emergency, one clinical question, one angry caller, one caller who hangs up midway. Score each on whether the outcome matched your protocol. If you're comparing platforms, how NovaReps compares to Retell shows what to check on the developer-platform side versus a packaged receptionist.
If a vendor won't let you run that test, you have your answer.