Med Spa AI Agents: Consents and Treatment Follow-Up Boundaries
A practical guide for med spa operators on using an AI agent for consent paperwork and follow-up, with clear limits between admin work and clinical decisions.

An AI agent for a med spa should handle the administrative work around treatments: sending the approved consent form, checking that a current signed form exists before a visit, scheduling follow-up check-ins, and routing patient messages to the right licensed person. It should not judge candidacy, interpret symptoms, choose or adjust treatments, or decide whether a patient is cleared for care. Writing that boundary down before the first deployment is what makes the agent safe to run and easy for your team to trust.
Key takeaways
- Clinical judgment stays with licensed staff. Candidacy, dosing, contraindications and any assessment of a reaction remain with the practitioner, and the agent never answers those questions.
- Write two lists before setup: the administrative tasks the agent may complete alone, and the triggers that force a handoff to a person.
- Build company memory around consent versions, approved aftercare sheets and follow-up protocols, with each item tied to its owner, version and effective date.
- Start with one workflow and one success criterion you can observe, such as no treatment booked without a matching signed consent version on file before the appointment.
- Review permissions, logs and handoff quality on a fixed schedule, because forms, staff and clinic policies change.
Where administrative work ends and clinical judgment begins
Most med spa front desks already draw this line informally. A coordinator can send paperwork and confirm an appointment, but a nurse or practitioner decides whether a patient should proceed. An agent needs the same line written as rules, because it will receive questions at any hour and it cannot read the room. The safest approach is to default to handoff: if a message could be clinical, the agent treats it as clinical.
Use this decision guide as the starting draft for your practitioner to edit:
- The agent may send the current consent form for a booked service and remind the patient to complete it.
- The agent may check whether a signed form for the current version exists and flag gaps to staff.
- The agent may send aftercare instructions that the practitioner has already approved, word for word.
- The agent may schedule, move or confirm follow-up visits in the clinic calendar within rules staff set.
- The agent must hand off any question about whether a treatment is appropriate, whether a symptom is expected, or how a medication or pregnancy status affects care.
- The agent must hand off any request to change consent wording, grant an exception, issue a refund, or discuss a complaint.
- The agent must hand off any message describing an unexpected reaction and must not offer reassurance or instructions beyond the approved text.
What company memory should hold for consents and follow-up
Company memory is the set of approved documents and facts the agent is allowed to rely on. For consents and follow-up, that means specific documents with version history, not general knowledge the model happens to have. When a patient asks what to expect after a service, the agent should retrieve the approved aftercare sheet for that service and cite it, rather than generating an answer. Retrieval of this kind is described in general terms in the Google Cloud architecture material linked in the sources, and the principle is simple: the answer comes from your document, and the document has an owner.
Store these items, each with an owner and an effective date:
- Consent forms by service and version, with the date each version became active and the practitioner who approved it.
- Aftercare sheets by service, including the exact contact instructions for symptoms that need a call.
- Follow-up protocols that state the check-in timing, the channel, and the person who signs off on any change.
- Escalation contacts with working hours and a backup for each shift.
- Records of which consent version each patient signed, kept in the system of record and linked to the appointment, not copied into the agent's notes.
Keep the memory small and current. Old consent versions should stay in the archive with their dates so you can explain what a patient signed, but the agent should only offer the active version.
Memory that compounds is memory someone maintains: each correction from a review becomes a dated update to a document.
A step-by-step plan for the first engagement
A first engagement should be narrow enough to finish and measure. The sequence below assumes one clinic, one workflow and a supervising practitioner who can approve the boundary.
- Step 1: Choose one workflow. Pick either the pre-visit consent check or the post-treatment check-in, not both.
- Step 2: Have the supervising practitioner approve the handoff list in writing, including who receives each type of handoff and how fast they must respond.
- Step 3: Collect the active consent forms, aftercare sheets and protocols for the services in scope, and record the version and effective date of each.
- Step 4: Define one success criterion you can observe, for example that every booked treatment in scope has a matching signed consent version on file before the appointment time.
- Step 5: Set permissions narrowly. The agent gets read access to the schedule and consent status, write access to reminders and calendar holds, and no access to clinical notes beyond what the workflow requires.
- Step 6: Run in shadow mode first. The agent drafts messages and staff approve or edit them for a period the team agrees on before anything is sent automatically.
- Step 7: Review misroutes, missed handoffs and document gaps every week. Update memory with dated changes, and widen scope only after the criterion holds.
Illustrative scenario: the morning after a filler appointment
Illustrative scenario, not a client example: a patient messages the clinic the morning after a filler appointment asking whether the bruising is normal and whether she can take a pain reliever she already has at home. The agent recognizes both questions as clinical. It does not answer either one. It sends a short acknowledgment that a team member will reply, retrieves the approved aftercare sheet for that service, and includes only the contact instructions printed on it. It then routes the thread to the on-duty nurse with the appointment, the consent version the patient signed, and the patient's exact words. It logs the handoff with a timestamp.
The nurse replies to the patient directly. Only after the nurse marks the case as cleared does the agent schedule the standard check-in message for the next day. If the nurse does not respond within the agreed window, the agent escalates to the backup contact rather than sending a reassurance message on its own. The value here is not speed of the clinical answer. It is that the patient gets a timely, accurate route to a person, and the clinic can later show what happened.
How Be Human's managed approach handles this
Be Human designs and operates the agent in this kind of setup. A first engagement defines the one role, the memory and tools it needs, the approval boundaries and the success criterion, and the clinic's supervising practitioner signs off on the clinical handoffs. After launch, Be Human runs monitoring, recurring evaluations of routing and document use, permission reviews, integration maintenance and incident response, and it proposes controlled changes that the clinic approves. Clinical rules remain the clinic's responsibility. The service is remote and available in English and Portuguese.
Controls, permissions and review
The boundary is only as good as the controls behind it. Do not rely on the prompt alone to stop the agent from doing something it should not do. Authorization should be enforced by the systems the agent connects to: each tool gets its own scoped credential, and the calendar or messaging connector refuses actions outside the granted permissions. The OWASP authorization guidance in the sources is a useful checklist for this. Patient messages can also contain text that tries to change the agent's instructions, so treat any inbound message as data to be read, never as a command, as the OWASP guidance on prompt injection describes.
Keep an audit trail. For each action, record which tool was used, which document version the agent relied on, who approved any message, and when a handoff was made. Observability concepts in the OpenTelemetry primer apply here: you want traces that let a reviewer reconstruct one interaction end to end.
Set a review cadence and keep it. Weekly, review handoffs and any messages that were sent without approval. Monthly, review permissions, document versions and the success criterion. Whenever a consent form or aftercare sheet changes, test the agent against the new version before it goes live. Data protection rules for health information vary by location and change over time, so confirm your obligations with qualified counsel before you connect patient records. The NIST AI Risk Management Framework gives a structure for mapping these risks to owners and controls.
If you want to find the first task in your clinic that is worth handing to an agent, take the free Be Human AI diagnostic.
Frequently asked questions
Can an AI agent answer patient questions about aftercare? It can share aftercare instructions that the practitioner has already approved, retrieved from the current document and quoted without changes. Any question about whether a symptom is expected, or how a medication affects recovery, belongs to licensed staff and must be routed to them with the relevant context attached.
What is the first thing to define before setting up the agent? Define the handoff list in writing, with the supervising practitioner's approval and named backup contacts for each shift. The agent's permissions and success criterion should follow from that list, not the other way around.
How does the agent know which consent version a patient signed? It should read that fact from your system of record, where the signed form is linked to the appointment and its version. The agent should not infer the version from memory or from the patient's message, and if the record is missing, it should flag the gap to staff.
How do we know the agent is still behaving correctly after launch? Use recurring evaluations built on a fixed set of realistic scenarios, including edge cases like unexpected reactions and requests for exceptions, and review the logs of handoffs and approvals. Re-run those checks whenever a consent form, aftercare sheet, protocol or permission changes.
Your next step
See where an AI agent fits in your company first.
Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.
Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.


