A recall list is a starting point for patient communication. Turning it into a useful conversation takes current records, appropriate timing, a respectful contact plan, and a clear scheduling outcome. This playbook outlines an administrative workflow for a supervised voice AI pilot. Care intervals and treatment decisions stay with the dental team.
Start with the right recall list
The practice decides who is due and what comes next.
ADA practice-management guidance describes using practice software and staff follow-up to support recare appointments [1]. For an AI-assisted workflow, a staff member should first verify the cohort: current patient status, the clinician-approved interval, existing appointments, contact preferences, and exceptions. Do not treat every patient as due at a universal six-month interval.
Keep the outreach purpose narrow. A routine recall invitation should not become an unsolicited discussion of treatment, insurance benefits, or financial eligibility. Establish the permitted contact channel, frequency, opt-out process, and calling windows before outreach. Have the practice review the legal requirements that apply to its calls. A record in the practice system is not, by itself, permission for every kind of automated contact.
- 01Staff approve cohort→
- 02Check preferences→
- 03Invite a conversation→
- 04Record the response→
Remove avoidable confusion
Exclude patients already scheduled, records requiring staff review, and numbers flagged as wrong or restricted before the next contact attempt.
Use approved wording
Introduce the practice and AI assistant. Keep voicemail limited. HHS guidance highlights discretion in messages and confidential communication requests [2].
Respect a no
Make it easy to decline, request a person, or change contact preferences. Confirm how staff review and propagate that preference across relevant systems.
The AI should work from a staff-approved recall cohort; it should not decide when a patient clinically needs care.
Give each answer a clear path
Confirm, reschedule, decline, or ask a person.
A useful appointment conversation reduces uncertainty. Read back the agreed date, time, location, and visit type after a verified booking. If the connection can only submit a request, explain that staff will confirm availability. Set limits for provider changes, linked family appointments, interpreter requests, longer visits, and other exceptions the office already handles manually.
The workflow below is a proposed design, not an observed outcome. Use fictitious patient records for testing. Try an unavailable slot, a patient interrupting the readback, an ambiguous date, a parent calling for a child, and someone asking about pain or medication. Staff determine representative-access rules and clinical escalation scripts; the assistant should not answer treatment questions from general model knowledge.
- 01Understand preference→
- 02Check permitted slots→
- 03Verify or mark pending→
- 04Read back the next step→
Ready to schedule
Offer only approved visit types and available slots from an assessed connection. Keep an appointment request distinct from a confirmed appointment.
Needs another time
Use the office's rescheduling rules. If no suitable option exists, capture a callback request rather than repeatedly offering unsuitable times.
Needs the dental team
Route clinical questions, uncertainty about treatment, or a request for a human to the approved team. Record only the permitted context.
A conversation can end usefully without a booking: a corrected number, respected preference, or well-owned staff handoff is meaningful work.
Follow the whole appointment journey
Bookings and attended visits are different outcomes.
Report a sequence of distinct counts: approved records, eligible contact attempts, reached patients, scheduling requests, verified appointments, and attended visits. Use unique patients or episodes where appropriate, so repeated calls do not inflate success. Track declines and wrong numbers separately from unreachable patients. Define a consistent follow-up window before comparing cohorts.
Illustrative example: 200 approved records produce 120 reached patients and 36 verified appointments. The booking rate among reached patients is 30%; among all approved records it is 18%. Neither number is an attendance rate or an incremental lift. To estimate incremental value, compare with a suitable baseline, account for appointments that would have happened anyway, and use contribution after variable care costs rather than treating gross production as profit.
- 01Approved records→
- 02Patients reached→
- 03Verified appointments→
- 04Attendance reviewed→
Patient experience
Review requests for a person, complaints, preference changes, and repeated contacts. A higher dial count is not a better patient experience.
Office workload
Measure staff review, cleanup, callback handling, and corrections alongside time released. Separate implementation effort from ongoing operations.
Scheduling quality
Audit duplicate bookings, incorrect visit types, wrong locations, and unconfirmed requests. Have a staff owner reconcile exceptions.
Judge a recall pilot on accurate scheduling, respectful contact, staff workload, and eventual attendance together.
Research, with its limits visible.
These whitepapers synthesize the sources below with original workflow recommendations by Vatsal Patel. They are not systematic reviews, clinical studies, or measured Voicesis outcomes. Calculations are illustrative. Capabilities, system access, contracts, and safeguards must be confirmed for each deployment. This is operational education, not clinical or legal advice.
- American Dental AssociationRecare Appointments ↗
Practice-management guidance on recare tracking and follow-up; not evidence for AI recall effectiveness.
- U.S. Department of Health and Human ServicesMay providers leave messages for patients? ↗
Patient message privacy and confidential communication requests.
Sources reviewed September 24, 2026. External sources may change. No endorsement by the cited organizations is implied.
Keep a copy · Download PDF ↓