Home › Resources › Blog › AI Dental Receptionist Benchmark: What Happened on the Same New-Patient Call
BLOG · 4 MIN READ

AI Dental Receptionist Benchmark: What Happened on the Same New-Patient Call

The useful question is not “Which bot has the biggest feature list?” It is “What happens when a patient changes direction, mentions pain, or asks for a person?”

DentalTechHub reviewed one supplied recording per vendor. Nine entrypoints accepted the patient scenario. DialZara, MyfrontdeskAI, and RingCentral were vendor-information lines, so their calls are described without pretending they were configured office deployments. 1

Dead-air share in the 12 test recordings

Call rhythm, not a score

Across the supplied recordings, dead-air share ranged from 25% to 39%, while the reported speaking rate ranged from 166 to 302 words per minute. 2

Those numbers are descriptive. Silence may reflect a lookup, and speed may feel efficient or rushed. The chart is a listening guide, not a leaderboard. The three vendor-information entrypoints are also marked separately because their conversations had different jobs.

Disclosure varied

Some scenario calls identified the assistant as AI at the opening. The mconsent and Sally transcripts did so. In other calls, identity became clear only after the caller asked. 3

The RingCentral information-line transcript did not clearly answer the first direct identity question, while Dentobot described the call as a demo but sidestepped the later identity question before returning to scheduling. 3

For a practice, disclosure should be a configured policy—not an improvised response. Test the exact opening, the response to a direct question, and any language-specific variations.

Booking confidence needs a receipt

A convincing appointment flow should distinguish among checking, holding, requesting, and confirming. In the RevenueWell scenario, the assistant required more registration information before it would search availability. In the Flossy scenario, the original appointment was represented as booked, but the requested reschedule ended with office follow-up rather than a clear confirmation of the new slot. 4

Neither pattern is automatically wrong. A practice may deliberately require registration fields, and a callback can be safer than a false confirmation. The important question is whether the caller and staff receive the same, accurate status.

Human handoff was not one behavior

The recordings showed several handoff patterns: an announced transfer, a callback offer, and a message for the team. Dentobot announced a transfer but the transcript ended before completion. Dentina offered a message. Aron offered to collect callback details. 5

A buyer should test completion, not just intent. Confirm that the destination rings, the message contains the right context, the callback enters a staff queue, and the patient hears a realistic expectation.

What we did not turn into rankings

We did not combine pause share, disclosure, booking behavior, and handoff into one vendor score. These fields have different meanings, and the calls were not equivalent enough to support a universal ordering.

We also did not treat a vendor statement on its own demo line as independent proof of integration or clinical safety. Those claims belong in follow-up research and a configured test environment.

Practical interpretation

Use this benchmark to decide what to reproduce in your own demo:

  1. Call as a new patient.
  2. Change the requested day after the assistant proposes a slot.
  3. Ask an insurance question that requires qualification.
  4. Introduce an urgent-sounding symptom.
  5. Ask directly whether the caller is speaking with AI.
  6. Ask for a person and verify the handoff reaches its destination.
  7. Check the practice management system and staff queue afterward.

Read how the test was structured, review the common gotchas, and use the practice-fit guide before browsing the marketplace category.

Frequently asked questions

Which AI dental receptionist won?

This pilot does not name a universal winner. One recording per vendor and mixed entrypoints do not support that conclusion.

What does dead air mean?

It is the measured share of a recording without speech. It does not reveal why the silence occurred or how a patient experienced it.

Can a demo prove integration quality?

No. Verify a configured workflow end to end, including the practice management system, notifications, and staff follow-up.

Sources and methodology

  1. Entrypoint normalization
  2. Mechanical call-rhythm range and caveat
  3. Observed AI disclosure behavior
  4. Observed booking and intake behavior
  5. Observed handoff behavior
Next Resource →