Your front desk knows the answer. AI agents need it too.

AI assistants have started booking beauty and wellness appointments. The calendar is the easy part. The hard part is everything a front desk knows: which service, with which provider, whether a consultation comes first, the deposit, and where the business line ends and the clinical one begins.

DIRECTIONA future direction. No present-tense capability claim.

Explore with AI

The request

“Find a med spa near me that does this treatment, with an appointment after 5 this week.” Then the follow-ups: does it need a consultation first, what's the deposit, which provider. Every one of those is a business fact. None of them is medical advice. A good agent keeps it that way.

Most of what clients ask a front desk is a business fact.

The calendar is opening up

Google's AI Mode books beauty and wellness appointments through partners such as Booksy, Fresha and Vagaro, and Google says it will call businesses for users in categories including beauty. Fresha announced bookings in ChatGPT and Claude in August 2026, medspas included. The scheduling platforms ship AI receptionists of their own.

The rails assume the services, rules and openings they read are right.

Agents can reach the calendar. They can't fix what's wrong around it.

What the front desk knows that the web doesn't

The service menu on the site and the one in the booking platform rarely match exactly: names, lengths, prices. Whether a first visit must be a consultation is often stated only halfway through booking. Providers join, move locations, or stop offering a service, and the team page catches up later.

Med spas are mostly single locations, and the groups average nine. In a group, each of those differences repeats per location.

The facts exist. They just don't agree with each other in public.

The clinical line

A filler injection is a medical procedure in the FDA's words, and AmSpa's guidance puts a good faith exam by a qualified clinician before a first treatment. So the boundary is clear. An agent can say a service requires a consultation. It can't say who is a candidate, whether it's safe for them, what the side effects are, or what results to expect.

We built that line into the model, not the prompt: clinical questions return the consultation path, and the consultation request has no free-text field, so it can't carry health information.

Business facts, yes. Clinical judgment, never.

Openings without guessing

Availability lives in the scheduling platform, by provider and room, and it changes by the hour. The rule we built: an opening is stated only from a reading under 15 minutes old. Older is “ask the location.” Missing is unknown, never “fully booked.” And an opening is never a hold.

An opening has a time on it, or it isn't stated.

Controlled actions

Booking, rescheduling and deposits run through the business's own platform. Lehvel's first action is a consultation request routed to a person at the right location, with a receipt that commits nothing. Anything more needs the business's opt-in, its platform's own interface and a gate in front of it.

Start with a request a person answers.

The broader thesis

Med spas show that not every vertical's hard problem is a product feed. Here it's operational truth: services, providers, rules and openings, and the line around them. The same core handles it, with a different set of facts on top.

01 · Question
What does an AI agent need from a med spa to answer a client and take the right next step, without crossing into clinical judgment?
02 · Why it matters
Before modeling med spas we read how their front desk actually works: the scheduling and EMR platforms, their own AI receptionists, Google's booking partners, the marketplaces opening calendars to AI assistants, and the rules around health claims and health data. The aim was to find where outside assistants run out of trustworthy answers, and where they must stop.
03 · State
DIRECTION A future direction. No present-tense capability claim.
04 · What we built or tested
  • A tested med spa model in Lehvel's repo, on the shared core for businesses customers visit or book: services by location with their stated requirements and terms, providers only as the business lists them, openings only from a reading under 15 minutes old, a clinical boundary that sends candidacy, safety and results questions to a consultation, and a consultation request with no free text. No business's record is loaded.
  • Not live: any agent-facing tool or consultation request for a med spa.
05 · Evidence
U.S. med spa locations (AmSpa, 2024 report)10,488 in 2023, up from 8,899 in 2022; 81% are single-location.
Repeat clients (AmSpa, same report)73% of patients are repeat patients, spending $527 per visit on average.
Booking in AI Mode (Google, Nov 2025)AI Mode helps book beauty and wellness appointments through partners such as Booksy, Fresha and Vagaro.
Calls on a user's behalf (Google I/O, May 2026)Google will call businesses for users in select categories, including beauty.
Bookings in ChatGPT and Claude (Fresha's announcement, Aug 2026)Fresha opened bookings to ChatGPT and Claude through an MCP server, including medspas.
Fillers (FDA)Getting filler injected “should be considered a medical procedure, not a cosmetic treatment.”
06 · What we learned
  • Booking is arriving through the platforms and marketplaces. A med spa doesn't need another booking system; it needs its facts to be right in all of them.
  • The front desk's knowledge is the gap: service names that differ between site and booking menu, consultation rules buried in the flow, providers who moved, deposits stated only at checkout.
  • The scheduling platforms ship their own AI front desks. Those answer on the business's channels. Outside assistants still read whatever is public.
  • There's a hard line in this vertical. Business facts can be answered. Candidacy, safety, side effects and results belong to a qualified provider, and the right answer is how to request a consultation.
  • A consultation request should carry no health information at all. The simplest way to guarantee that is to give it nowhere to go.
  • Unknown is not “no.” A service with no stated consultation rule is “ask the location,” never “no consultation needed.”
07 · Limitations
  • AmSpa's own figures conflict (10,488 locations in its report recap, “more than 11,000” in a release); we use the recap and its date.
  • Most scheduling platforms' APIs are partner-gated or enterprise-only; what they expose was not read directly.
  • Whether HIPAA applies depends on how each practice bills. We didn't, and won't, decide that for any business.
  • No business's record is loaded, so the model hasn't met a real scheduling export.
08 · What changes next
Build it with one multi-location med spa first: services and requirements by location, providers as listed, and a consultation request that lands with the right front desk.
Negative results are published as results. A state on this page is a claim about what is true today, not about what is planned; a limitation is a limitation.
These tools belong to the people who made them. Lehvel reviewed and, where noted, tested them. Nothing here is a Lehvel product, and no performance figure attributed to a vendor has been reproduced by us.

Explore with AI