Setting up the AI Receptionist

The AI Receptionist answers your inbound calls. Existing patients can book, reschedule, and cancel visits over the phone; callers who aren't patients yet are handled the way you choose — as a lead for follow-up, or as a brand-new patient registered during the call.

Turn the line on

Go to Settings → AI Receptionist.

The Receptionist line toggle enables "call intake to provision a dedicated receptionist number". Switching it on assigns a phone number to your practice.

Provisioning status

Provisioning status shows "current phone number and assistant connection details". When everything is working it reads Active — "the receptionist is ready to answer inbound calls" — with an active badge.

Below that:

• Phone number — the provisioned number, e.g. +1 (415) 918-5486.
• Last ready — when the assistant last confirmed it was available.

This is the number to publish or forward your main line to. It isn't your existing practice number; provisioning creates a new one. Most practices forward their published number to it rather than reprinting stationery.

New callers

New callers is where you "choose what the receptionist does when a caller is not an existing patient". Two modes:

• Capture a lead (the default) — "the receptionist takes the caller's name and contact details and records a lead for your team to follow up in the CRM pipeline." Nothing is added to your patient list; the caller lands in Leads for a human to pick up.
• Register as a patient — "the receptionist collects the caller's details, creates a patient record, and can book a visit on the same call. If your insurance feature is enabled, it also collects insurance details."

In register mode, the receptionist works through the caller's details one question at a time — name with spelling confirmed, date of birth read back, phone, email, and gender. Only the name and date of birth are required; a caller can decline the rest. When your insurance feature is on, it also asks whether the caller plans to use insurance and, if so, captures what's on their card — insurance company, member ID, group number if the card lists one, plan name, the subscriber's address, and whose policy it is.

Two safeguards keep the patient list clean. The receptionist only registers someone after they explicitly confirm they've never been a patient at your clinic — otherwise it re-checks the records with corrected details or falls back to taking a lead. And if the details it collects match an existing record after all, no duplicate is created; the receptionist continues the call with the existing patient.

Switching modes takes effect from the next call — no re-provisioning needed.

Custom instructions

Custom instructions add "behaviour your receptionist should follow on every call", and are "used to tailor how the receptionist greets, qualifies, and routes callers". The field takes up to 20,000 characters.

The placeholder shows the intended style — specific, checkable rules:

• "Always confirm caller name and date of birth."
• "Never schedule appointments before 9am or after 5pm local time."

Write constraints rather than personality. "Be friendly" gives the model nothing to act on; "never quote a price for cosmetic procedures, take a message instead" does. The rules worth writing are the ones where getting it wrong would matter — clinical boundaries, scheduling limits, what not to promise.

Be explicit about what the receptionist must not do. An AI answering healthcare calls should not give clinical advice, confirm whether someone is a patient, or discuss anything from a record. Write those boundaries in rather than assuming them.

Knowledge base

Knowledge base lets you "upload PDFs and docs your receptionist can reference". Click or drag files into the upload area, up to 25MB.

The empty state suggests what to include: "Drop intake protocols, FAQs, or pricing sheets above. Your receptionist will reference them on calls."

This is how the receptionist answers practice-specific questions — your hours, what a service costs, what to bring to a first appointment — without you writing it all into the instructions field.

Keep uploads current. A stale pricing sheet doesn't sit quietly in a folder; it gets quoted to callers. Anything you upload should be something you'd be comfortable having read aloud to a patient.

What happens to calls

Existing patients can book, reschedule, or cancel visits directly on the call — appointments appear in Visits like any other booking.

Where new callers end up depends on the mode. In Capture a lead, they arrive in the Leads section of the sidebar, and Settings → Leads controls lead capture and qualification. In Register as a patient, they appear in your Patients list right away — often already with a first visit booked — and callers who decline to register are still recorded as leads so nobody falls through.