On the left is what a leading assistant said on November 16, 2026. On the right is the same request when the hospital has agentic booking turned on. The doctor and times on the right are made up.
On insurance: I can't confirm in-network status for your specific plan, and this is worth verifying carefully.
I can't book it for you. I don't have access to [Hospital]'s scheduling system, so the appointment has to come from you by phone or online. Phone: [number].
The hospital's own surgeon is listed third, coverage is never confirmed, and the patient leaves with a phone number.
[Hospital]'s knee team can likely help. A few quick questions so I match you with the right doctor: Does your knee lock, catch or give way? Have you done physical therapy, and for how long? Can I share your MRI report and insurance details with [Hospital]?
Dr. Maya Ellison, knee and sports medicine surgeon, [Hospital] Orthopedics, 1.2 miles
The assistant asks what a scheduler would ask, insurance is confirmed, and the slot is real. No phone call.
New patient · 58 · right knee · UHC Medicare Advantage. Checked against payer criteria for knee arthroscopy with partial meniscectomy.
Every check shows its evidence, so your team can see why each patient was routed. A patient with two weeks of knee pain and no MRI is booked with a non-surgical sports medicine doctor at your hospital instead, who can refer within your system later.
Every booking an AI assistant attempts runs through the same five steps your schedulers and nurses handle on the phone. Today the assistant fails at most of them. Substrate gives it a way to finish each one under your rules.
The "today" column comes from a real AI assistant transcript (Nov 16, 2026; hospital, doctor and phone details hidden). Triage criteria shown on this page are typical of published payer policies for knee arthroscopy.
Your doctors, locations, specialties, accepted plans and access rules are published as structured pages that ChatGPT, Claude, Gemini and Perplexity can read and cite. We track which assistants send patients to you and which send them elsewhere.
A read-only, unauthenticated endpoint lets any assistant search your doctors, plans and open appointment windows. It carries no patient data. It uses the same model as our public payer MCP, which assistants already query at mcp.substrateai.com/payers.
When the patient agrees to share, their assistant hands over symptoms, imaging, therapy notes and insurance. Substrate checks them against payer criteria and each doctor's fit rules, verifies eligibility, and shows every check with its evidence.
The appointment is written to your scheduling system with the packet attached. If the patient would rather talk to a person, your front desk gets the request and calls back while the patient is still interested. Your team sees new patients found, triaged and booked by service line each week.
It is when a patient's AI assistant finds a doctor, checks coverage and books the appointment for them. Substrate is the hospital side of that exchange: it gives assistants accurate data about your doctors and plans, asks your triage questions, verifies insurance and books into your schedule under rules your team sets.
Any assistant that can read the web or call an MCP server. Your data is published as structured pages and as an MCP endpoint, and booking runs on standard scheduling resources (FHIR R4 Slot and Appointment) so no single assistant or EHR is required.
The public layer carries no patient data. Clinical and insurance details move only after the patient agrees to share them in their assistant, travel over an authenticated handoff, and are processed under a BAA with HIPAA and SOC 2 controls and a full audit trail.
Insurance is checked before a slot is offered, and each doctor's fit rules decide who books with whom. A patient who isn't ready for a surgeon gets the right first visit instead, such as non-surgical sports medicine, so they still start their care with you.
A marketplace books the patient into its own inventory and charges per booking, and the relationship sits with the marketplace. A portal assistant helps once the patient is already logged in to your portal. Substrate opens your own schedule to assistants outside your walls, on your rules, so the patient stays yours.
With one service line. We need read access to your provider directory and scheduling templates, plus the questions your schedulers already ask for that service line. We start by showing you what ChatGPT, Claude and Gemini say about you today.
Book 30 minutes. Pick one service line and we'll run the questions your patients ask in ChatGPT, Claude and Gemini, show you who they get sent to, and walk through what triage, insurance checks and booking would look like on your schedule.