Patients want to know when AI books their visit. Designing a disclosure-first scheduling agent
Most US adults want to be told when AI schedules their appointment. Scripts, opt-out and escalation rules for a clinic booking agent that says what it is.
The short answer
Design the agent to say it is AI in its first sentence, offer a person at any point, and hand over whenever a patient mentions symptoms. In Pew's June 2026 survey of US adults, 56% wanted to know when AI schedules appointments. Disclosure costs one sentence; discovering it later costs trust.
Key takeaways
- In Pew's survey of 3,488 US adults (22 to 28 June 2026), 72% said it is extremely or very important to be told if a provider uses AI.
- Scheduling ranked lowest of the tasks Pew asked about, yet 56% still wanted to know when AI books their appointment.
- In the same survey, 46% were not sure whether AI had been used in their care, which is a trust gap a clear disclosure closes.
- A disclosure-first agent names itself in the first sentence, offers a person at any point and hands over when a patient mentions symptoms.
- Log every disclosure, opt-out and handover, so you can show what the agent did and how patients responded.
In this article
What patients told Pew
Most US adults want to be told when AI is part of their care, including the admin. In Pew Research Center's survey of 3,488 US adults, fielded 22 to 28 June 2026, 72% said it is extremely or very important that providers tell them when they use AI.1
The wish holds even for routine tasks. Pew asked about six uses. Analysing scans and making a diagnosis topped the list at 81% each, and scheduling an appointment came last. Even so, 56% wanted to know when AI schedules their appointment.1
Patients are also unsure what is already happening. In the same survey, 46% said they were not sure whether AI had been used in their care, and 63% wanted more say over how it is used.1 Our view: a booking agent is the easiest place for a clinic to start being clear, because the stakes are low and the patient is already talking to it.
US adults who want to be told when AI is used, by task
Why disclose for something as small as booking?
Because patients find out anyway, and finding out later feels like being misled. A booking call that sounded human, followed by a reminder text signed by an assistant, invites the question of what else was automated.
Patients also bring their own AI habits. In KFF's poll of 1,343 US adults (24 February to 2 March 2026), 32% had used AI for health information or advice in the past year.2 Many patients know what a chatbot sounds like. Pretending otherwise rarely works, and it spends trust you need for later conversations about care.
There is a practical reason too. Rules on AI disclosure differ by state and by country, and they are changing. In the EU, Article 50 of the AI Act requires AI systems that talk directly with people to tell them they are dealing with an AI, unless that is obvious.3 Our note on EU AI Act Article 50 covers those rules for clinics serving patients in Europe. Disclosing by default means you are not redesigning the agent every time a rule changes. Check the rules that apply to you with counsel; this post is design guidance, not legal advice.
The disclosure script
Put the disclosure in the first sentence, in plain words, before any question. Say what the agent is, what it can do and how to reach a person. Here is an illustrative script for a fictional practice.
Three rules keep the script honest. Use the word "AI" or "automated", not a human name with no label. Do not ask for anything sensitive before the disclosure. Repeat the label on every message the agent sends later, such as reminders and confirmations, so the record matches what the patient heard.
Opt-out without penalty
Every patient should be able to reach a person at any point, with one word or one key. The opt-out only works if it costs nothing: the same slots, the same priority and no extra wait beyond what staffing requires.
Our view: an opt-out that drops patients into a long hold queue is not an opt-out. If the clinic cannot staff a callback within a stated time, say so in the script and offer a callback instead. Tell patients exactly when someone will call.
Store the preference. A patient who opts out once should be routed to a person next time without asking again, unless they choose the assistant.
A disclosure-first booking flow
- Disclose
First sentence: an AI assistant, what it can do, how to reach a person.
- Identify
Verify the patient with the minimum details the booking needs.
- Screen for handover
Any symptom, urgency or clinical question goes to a person at once.
- Book
Offer slots from the scheduling system; the agent writes only to the calendar.
- Confirm and label
Confirmation names the assistant and repeats the way to reach staff.
When must the agent hand over?
Hand over the moment a patient says anything clinical. A booking agent is not a triage service, and it should not try to judge whether a symptom can wait.
- Symptoms or urgency. Any mention of pain, bleeding, breathing trouble or "is this serious" goes to a person, with emergency guidance first where the words suggest it.
- Medication questions. Dose, side effects or refills go to clinical staff.
- Billing and insurance disputes. These need someone who can make a decision.
- Distress or confusion. Repeated misunderstanding, frustration or a request for a person.
Our escalation design for support agents sets out how to pass context to the person, so the patient does not repeat themselves. The same pattern applies here: the handover carries the transcript and the reason.
Limit what the agent can touch. It should read open slots and write appointments, and nothing in the clinical record. Our least-privilege checklist was written for ERPs and CRMs, but the principle carries over. Any vendor that handles patient data also needs the agreements your privacy rules require.
What to log, and what to measure
Log enough to show what the agent said and did on every call or chat. If a patient later asks whether they spoke with AI, you should be able to answer from the record.
| Log | Why it matters |
|---|---|
| Disclosure delivered, with timestamp | Proves the patient was told before booking |
| Opt-outs and the turn they happened | Shows where the script loses patients |
| Handover reason | Checks the clinical screen works |
| Booking completed, changed or abandoned | Measures whether the agent does the job |
| Complaints mentioning the assistant | Early sign of a trust problem |
Our recommended minimum log for a booking agent.
Review the opt-out rate in the first month. A high rate early on is normal and tells you where the script is unclear. A rate that does not fall after fixes suggests your patients prefer a person for this task, which is a result worth knowing before you automate more.
Patients also ask AI about clinics before they ever call. Our guide to what a US clinic should check in AI answers covers that side, and our healthcare page covers both. For practices in Australia, our note on what to publish within health advertising rules does the same.
