AI Agent for a General Dental Office: Recall Lists and Treatment Plans

An AI agent for a general dental office can prepare recall lists, track open treatment plans and answer routine insurance questions from the practice's own records. Diagnosis, changes to a treatment plan, and any statement that commits a patient to a cost or a coverage outcome stay with the dentist and the front-office lead. The useful question is not whether an agent can write a patient message, but which tasks it can finish on its own, which it can prepare for a person to approve, and which it must never touch.

Key takeaways

• Start with preparation work: recall candidates, reminder drafts and open treatment plan lists, all reviewed before anything reaches a patient.
• Recall eligibility rules belong to the clinical team. The agent applies those rules and logs exceptions, but it does not invent them.
• Treatment-plan follow-up is a tracking task. Accepted, pending and declined plans should stay visible, while changes to the plan itself are clinical decisions.
• Insurance questions can be answered from the benefit summary on file, with the source line quoted. Any answer that promises coverage or an amount goes to a person.
• Give the agent read access by default, write access only to draft queues and task lists, and keep an audit trail for every message it prepares.

Where recall and treatment-plan work gets lost

In a general practice, the gaps are rarely dramatic. A hygiene patient is due for a cleaning, but the reminder never generates because the interval was entered differently in two systems. A treatment plan is presented at checkout, the patient says they want to think about it, and nobody records when the follow-up should happen. Benefit questions arrive by phone during a busy morning and get answered from memory. Each gap is small, but together they decide whether the schedule fills and whether accepted care actually happens.

Front-office staff usually know these gaps exist. The constraint is time during clinical hours. An agent helps when it takes the repetitive list work off the desk, produces a short queue each morning, and leaves every judgment call visible instead of buried in a spreadsheet. The same split between tracking and clinical decision appears in other health settings, as the chiropractic intake and recall guide describes.

Building a recall list an agent can prepare

Start by writing down the recall types the practice actually runs: routine cleanings, periodontal maintenance, re-checks after a procedure, and patients who asked to be contacted at a set interval. For each type, a clinician sets the rule, including the interval and the conditions that exclude a patient. The agent does not decide that a patient with a particular history needs a shorter interval. It applies the rule the clinic wrote.

• Name each recall type and write its interval rule in one sentence, approved by a clinician.
• Identify one source of truth for appointments, last visit dates and status, and state which system wins when two disagree.
• Define exclusions: treatment in progress, a visit already booked, a patient who opted out of messages, and any record flagged for clinical review.
• Have the agent produce a daily candidate list that gives the reason for each name, not only the name.
• Assign a front-office person to approve each outgoing batch, and log which messages were sent, edited or dropped.
• Review the exclusion list on a fixed schedule, because a rule that worked last season often needs adjustment after a new service is added.

Treatment-plan follow-up: track, remind, never decide

A treatment plan is a clinical document with a financial attachment, so the agent's role is narrow. It keeps the status visible: presented, accepted, scheduled, completed, declined, or waiting for the patient. It can draft a follow-up message using wording the clinic has already approved, and it can flag plans that have stayed open longer than the team's own threshold. The orthodontic aligner tracking model follows a similar boundary between monitoring a sequence and making clinical calls.

Use this decision guide when a plan generates a message or a question:

• If the plan is accepted but not scheduled, the agent may draft a scheduling reminder for staff approval.
• If the patient asks a clinical question about the plan, the agent routes it to the dentist with the relevant note attached and does not answer it.
• If the patient wants to change the sequence, reduce the scope or arrange payment differently, a staff member handles the request.
• If the plan was declined, the agent records the reason in the patient's words and sets a review date only when the clinic has a written policy for that case.
• If the record conflicts with what the patient reports, the agent flags the conflict and changes nothing.

Insurance questions: answer from the file, escalate the rest

Patients ask whether a procedure is covered, how many cleanings a plan allows in a year, or whether a deductible has been met. A useful agent can answer the factual part by reading the benefit summary the practice has on file and quoting the line it used. Benefit rules differ by plan, change at renewal, and depend on details that only the insurer or a current eligibility response can settle. The agent should state what the document says, when it was last verified, and that the office will confirm the final answer.

Illustrative scenario: a patient messages the office to ask whether a scaling and root planing visit will be covered this year. The agent finds the benefit summary from the last eligibility check, sees that the plan covers the procedure category under a coinsurance rule and a waiting condition, and drafts a reply that quotes those lines and notes the verification date. The draft says a coordinator will confirm the estimate before the visit. The agent does not state an amount. The coordinator checks the current eligibility with the insurer, then sends the estimate.

The same discipline applies to every insurance question. If the answer depends on a document the office does not hold, the agent says so and creates a task for the coordinator. Silence or a guessed answer is the failure to avoid.

Permissions, privacy and clinical control

Dental records are health information. Depending on where the practice operates, rules such as HIPAA in the United States or the LGPD in Brazil shape which data can move to a tool and who may see it. Access should follow the principle that each identity receives only the permissions its task requires. The agent reads appointment and plan status by default, writes only to draft queues and task lists, and never sends clinical content without approval. Each tool call and each draft should leave a record that a person can review later.

Patient messages are also an input risk. A message can contain text that tries to override the agent's instructions, for example a request to ignore the exclusion list. Treat inbound messages as data, not commands, and keep the recall and plan rules in the system configuration rather than in anything a patient can write. The behavioral health intake limits show why the stopping points matter as much as the tasks. The OWASP authorization guidance and the NIST AI Risk Management Framework are useful references for setting these boundaries and reviewing them over time.

How Be Human runs this as a managed service

Be Human designs the first engagement around one useful role, here recall and treatment-plan follow-up, and defines the memory and tools that role needs, the approval boundary for every outgoing message, and one observable success criterion, such as how many candidate lists reach the schedule without manual correction. Managed operations then cover monitoring, recurring evaluations of drafts against the approved rules, permission reviews when staff roles change, incident response when a rule misfires, and controlled improvements. The clinic keeps ownership of clinical rules and coverage answers, and Be Human keeps the agent working inside those boundaries.

The next step is to take the free Be Human AI diagnostic to find the first task in your office worth handing to an agent.

Frequently asked questions

Can an AI agent send recall reminders to patients without staff review? It can, but a safer setup keeps a staff approval step for each batch at the start. After the exclusion rules have been tested against real records for a period the clinic chooses, the team can decide which message types may go out automatically. Every send should be logged so the office can see who approved what.

Does an agent replace the front desk or the dental assistants? No. It handles list building, status tracking and draft preparation, which are tasks that compete with phone calls and checkout time. People still decide, speak with patients, confirm estimates and handle exceptions.

Can the agent tell a patient whether a procedure is covered? It can quote what the benefit summary on file says and show the date of the last verification. It should not promise a payment or a coverage outcome, because the insurer's determination and the specific plan terms control the answer. The office gives the final estimate.

Does the agent need access to the full patient chart? No. A useful first setup needs appointment status, recall dates, treatment plan status and the benefit summary. Clinical notes stay out of scope unless a clinician defines a specific, documented need and the access is reviewed.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.