AI agents are now booking medical appointments for patients, starting with Zocdoc. Booking is the first health task patients will let an agent do.

If you run a clinic, keep your openings and your insurance list accurate. Get the booking right, and patients will likely let an agent do more. Get it wrong, and nothing else it offers will matter.

AI in Practice, Issue 2 · AI agents. The opportunity, the risks and the safeguards that earn a patient's trust.

Two launches at the end of October made this real.

On Oct. 28, Zocdoc opened its booking system to outside AI agents. A day later, Ro, the online clinic, launched Robin. Robin works from the information Ro holds on each patient.

An agent can only book as well as the data it can see. That is where trust is won or lost first.

The opportunity

A patient asks an assistant for a skin doctor nearby who takes their insurance. The agent checks real openings and books one, at any hour.

Zocdoc's guide for agents says they can search doctors by specialty, location and insurance. They can check live availability, book, and cancel or reschedule for the patient.

The demand is arriving. Zocdoc says agent traffic is up nearly 30 times since July. It has not shared a starting number, so we cannot tell how big that is.

Zocdoc's CEO, Oliver Kharraz, MD, told Healthcare Dive that if patients want to hand booking to an AI agent, “we should give them the frictionless access to care.”

For a clinic, this is a front desk that never closes. Keep your openings and insurance list accurate, and any agent can book you.

The risks

Kharraz named the main one. Agents could send patients to doctors who do not exist, to appointments that are not open, or to doctors who do not take their insurance.

He blames scattered directories and old insurance data. “The consequences are real. And it's the patient who pays the price,” he told Digital Health News.

Those are his warnings about the whole field. The reporting I read gave no error rate for agent bookings.

The cause is ordinary. Picture a directory that still lists a plan the clinic dropped. Now picture a fast agent repeating it to every patient who asks.

A second risk is rules with no visible policing. Healthcare Dive reported that Zocdoc did not say how it will make sure agents respect its data agreement.

The safeguards

Zocdoc's rules for agents set a useful floor. An agent must say it is an AI, name the company behind it, and say it acts for the patient. It must get clear consent before it books, changes or cancels anything.

It may use patient data only for the task at hand. It may never train on that data or sell it. It may not make up availability, credentials or insurance.

Agents that break the rules get slowed down or cut off.

Those rules cover outside agents arriving at one platform. A clinic's own agent talks to the same patients, so it needs standards too.

I run a company that makes AI agents for clinics. I think any booking agent should meet three more:

It answers only from information the clinic has approved.
It books only under the clinic's rules, or hands staff a complete request.
Staff can review every call afterward.

The first stops the agent guessing about insurance, hours or who is available. The second keeps the rules in the clinic's hands, and a complete handoff means the patient never has to start over.

The third gives the owner something to check. It also gives the patient someone to hold to account.

The quiet work sits behind all three. An agent that answers only from approved information is only as good as that information.

Six questions to ask about your own data, and about any agent that books from it

Where do the hours, calendar and insurance list live, and who updates them when something changes?
When did a patient last catch an error before staff did?
Does the agent say it is an AI at the start, and ask consent before it books, changes or cancels?
Where do its answers come from, and who at the clinic approved them?
What happens when a request falls outside the clinic's rules, and can staff review every conversation afterward?
Is patient data used to train models, or sold?

Why records are the harder step

Booking needs a name, a time and an insurance plan. Records hold diagnoses, medications and lab results.

I have only seen press coverage of Robin, not Ro's own announcement. RuntimeWire's summary of Business Insider's report says Robin draws on information from doctors, prescriptions and smartphones.

The report said Ro has published no performance data and no safety evaluation. It has not said how Robin handles medication changes. Business Insider had not seen Ro's privacy policy for Robin.

Outside expert Ateev Mehrotra said he wants evidence first. He would rather see an agent that lives inside an official health record system, where strict privacy rules apply.

Patients draw a similar line. In EY's April 2026 survey of 1,000 Americans, 60% were comfortable with AI booking appointments. Only 47% were comfortable with AI helping decide treatment, and 44% with AI reassuring them about a health worry.

KFF polled 1,343 US adults from Feb. 24 to Mar. 2, 2026. It found 77% worried about the privacy of medical information given to AI tools.

It also found that 41% of people who use AI for health have already uploaded personal medical information, such as test results or doctor's notes.

In Canada, provincial health privacy laws, such as Ontario's PHIPA, already set the rules for clinics. The same questions apply to any agent that touches a patient's file.

To me, the gap between 60% and 47% says trust gets built on booking first. Records come later. A clinic that gets booking right earns the standing to ask for more.

Before an agent reads health records, I would want three things. The patient opts in for that specific use. The agent reads only what the task needs. And a clinician signs off on anything clinical.

If an AI agent booked your next appointment, what would you want afterward to trust it: a text confirmation, a transcript of the conversation, or a call from the clinic's staff?

Sources

Want to hear how a clinic's own booking agent sounds? Talk to Aida, no signup.

This is the web edition of AI in Practice, a plain guide to AI agents for clinics and service businesses. Know someone who would find this useful? Send them this page. They can sign up at attainmentlabs.com/newsletter.

Inbound

Outbound