AI employees

AI knowledge

What the Receptionist, Concierge, Follow-Up, Recall and Scribe answer from, and what they never do. Every source has an owner and a review date, and every answer can be traced back to the source it came from.

0 selected

	Source	Used by	Status	Verified	Answers, 30 d	Updated

The AI employees answer only from these sources and the practice profile. They never use them to give medical advice, and they say so when asked.

Recent changes

who changed what

One canonical field per fact. Hours, prices and locations live here and nowhere else, so the AI employees never have two versions to choose between. Fields synced from Appointments are read-only here.

Will not answer

fixed for every practice

These cannot be turned off. The fallback wording is yours to edit.

Emergency script

v3 · approved by Dr. Johnson, Sep 1

Trigger phrases (clinician-approved)

What the AI says, then does

Escalation rules

When	Goes to	Hours and fallback	Times, 30 d	On

Every hand-off carries the full transcript. Warm transfers ring for 25 seconds, then fall back.

What the AI may collect

minimum necessary

Call recordings kept 30 days, transcripts 6 years with the record, chat logs 6 years.

Disclosure

by law and by default

Says it is an AI at the start of every call and chat“You’re talking with Genesis Health’s AI receptionist.” Cannot be turned off.
“AI assistant” label stays visible in chatPersistent, not just in the first message
No clinical titles in the personaNever “doctor”, “nurse” or a clinician’s name (California AB 489)
Patients can ask for a person at any timeAlways routed, never argued with

AI-drafted messages to patients

A clinician reviews before anything is sentDefault. The message is labelled “Written with AI, reviewed by” the clinician; exempt from California’s disclaimer rule Send with an AI disclaimer and a way to reach a personOnly for administrative messages; clinical content always waits for review

By patient location

Questions they couldn’t answer

Question	Asked, 30 d	Last asked	Employee	Channel	Status

When the AI cannot answer it takes a message instead of guessing, and the question lands here. Nothing is added to the knowledge until a person answers it.

Suggestions to review

Suggestions are drafted by the AI from real conversations. Nothing is published until a named person accepts it; the approver is logged.

Test suite

Question	Employee	Expected	Last result

Every source or guardrail change marks the suite “Not run since change”. Test conversations never count in reporting.

Ask as a patient

ReceptionistConciergeFollow-UpRecall

Ask anything a patient might ask. You’ll see the answer, the source it came from and any guardrail that fired.

Preview conversations are not counted in reporting.

Prototype. Sources and conversations are samples. Time is fixed at Thu, Oct 8, 10:52 AM so the sample lines up across screens.