Key takeaways
- National surveys measure comfort with AI in diagnosis and treatment, where many patients are uneasy. Few measure AI on the phone for office tasks.
- Patients value speed: most people who use AI for health information cite quick, immediate answers[1].
- Acceptance depends on design: say it is automated, keep it to office tasks, make a person one step away, and confirm details back.
- Test acceptance in your own practice by tracking requests for a person, hang-ups and repeat calls, by caller age group where you can.
What the surveys actually measured
Ask whether patients like AI and the answer depends on what the AI is doing. The best-known US survey on the subject, from Pew Research Center, asked more than 11,000 adults about AI used to diagnose disease and recommend treatment. Six in 10 said they would be uncomfortable if their own provider relied on it[2]. Only 38% thought it would lead to better health outcomes, while 33% said worse[2].
The relationship numbers are the starkest: 57% said AI in diagnosis and treatment would make the patient-provider relationship worse, against 13% who said better[2]. On record security, 37% expected it to get worse and 22% better[2].
None of those questions was about a phone line that books appointments or answers parking questions. That distinction matters. The fears people express, about losing the human relationship and about machines making medical decisions, are exactly the things a well-designed front office voice should not do. The survey data is a warning about scope, not a verdict on automated phone help.
The convenience side of the ledger
Other data shows patients are already using AI when it saves them time. In a 2026 KFF poll, 32% of US adults said they were turning to AI for health information and advice[1], and 65% of those users said quick, immediate advice was a major reason[1]. Physicians expect the same benefit: in the AMA's 2026 survey, patient convenience was among the areas doctors said AI is likely to help with[3].
Put the two sets of findings together and a pattern emerges. People are wary of AI replacing the clinician and comfortable with AI that removes waiting. A caller who reaches an AI voice that answers on the first ring and moves their appointment in under two minutes is getting the convenience. A caller who feels trapped in a loop when they need a nurse is getting the fear.
Age is not the barrier many assume
Practices often assume older patients will reject automated voices. Evidence from adjacent technology suggests caution with that assumption. In CDC data for 2021, telemedicine use rose with age, from 29.4% of adults aged 18 to 29 to 43.3% of those 65 and over[4]. Older patients adopt technology that solves a real problem for them, especially when someone they trust explains it.
What older callers often do need is pace and clarity: shorter prompts, a slower speaking rate, the chance to interrupt, and confirmation of every detail. Those design choices help every caller, so build them in for everyone rather than trying to detect age.
Privacy is the question callers will ask
When patients do push back on an AI voice, the objection is often about privacy rather than the voice itself. In the 2026 KFF poll, 77% of the public said they were concerned about the privacy of personal medical information provided to AI tools[1]. At the same time, 41% of adults who had used AI for physical or mental health said they had uploaded personal medical information into an AI tool or chatbot[1]. Patients worry, and many share anyway, which puts the responsibility on the practice to collect only what it needs.
Three habits reduce both the risk and the worry. Keep the AI to administrative details: name, date of birth, callback number and the type of request. Never ask callers to describe symptoms to the automated line. And be ready with a plain answer when someone asks where the call goes.
When a caller asks “Is this being recorded? Who hears it?”: “Calls to this line are handled by the practice's automated assistant and recorded so our team can follow up and check quality. Only practice staff and our phone service provider, under a confidentiality agreement, can access them. If you'd rather speak to a person, say receptionist and I'll connect you.”
Adapt the wording to what is actually true for your setup, including your recording practices, retention period and vendor agreements, and make sure front desk staff can give the same answer in person. Inconsistent answers do more damage to trust than the AI voice ever will. If you are unsure what your vendor stores, find out before go-live rather than after the first patient asks.
Design choices callers react to
| Design choice | Likely caller reaction | What to do |
|---|---|---|
| Saying up front that it is automated | Fewer surprises and less anger later in the call | Disclose in the first sentence |
| A person one step away | Callers relax when they know they can leave | Accept operator, person, agent, receptionist or zero at any point |
| Long menus or long greetings | Impatience and hang-ups | Keep the greeting under about 15 seconds |
| Reading details back | Trust that the request was understood | Confirm name, date, time and callback number |
| Answering clinical questions | Discomfort and risk | Route every clinical question to staff |
| A text confirmation after the call | Reassurance and fewer repeat calls | Send date, time, location and a number to call |
| Interrupting or talking over the caller | Frustration | Allow barge-in and pause when the caller speaks |
Greeting scripts to test
Option A: “Thanks for calling [practice name]. I'm the practice's automated assistant. I can book or change appointments and answer office questions. To reach a person at any time, just say receptionist.”
Option B: “You've reached [practice name]. This is an automated assistant, and the team is here if you need them. How can I help today?”
Option A sets clearer expectations and suits practices with many first-time callers. Option B is shorter and warmer and suits established patient bases. Run each for two weeks and compare how often callers ask for a person within the first 20 seconds, which is a good proxy for whether the greeting landed.
How to test acceptance in your own practice
Measure before you start
Record a baseline week of hold times, abandoned calls and callbacks so you can tell whether the change helped.
Start with a narrow scope
Begin with overflow and after-hours calls, where the alternative is voicemail, so callers compare the AI with nothing rather than with a person.
Track requests for a person
Count how often callers ask for a person, and at what point in the call. Early requests suggest a greeting problem; late ones suggest a task the AI handles badly.
Listen to a sample
Each week, listen to calls that ended in a hang-up or a request for a person. Note the exact moment the caller lost patience.
Ask a short question
After a visit, ask a sample of patients how easy it was to reach the practice. Avoid asking whether they liked the AI; ask about the outcome.
Adjust and repeat
Change one thing at a time, such as the greeting or the confirmation step, and compare the next two weeks.
Example: a fictional overflow pilot
Example: a fictional internal medicine practice routes calls to an AI voice only when both front desk lines are busy and after 5 p.m. In the first two weeks, about one caller in five asks for a person, mostly in the first few seconds. Listening to those calls, the manager hears the same pattern: callers expected the receptionist and were thrown by the automated greeting.
Example, continued: the practice shortens the greeting, moves the say receptionist option to the second sentence, and adds a line on the hold message telling callers that an assistant will pick up if the team is busy. Requests for a person fall over the next two weeks, and callers who stay with the assistant mostly complete their task. The practice keeps the AI on overflow and after hours and decides to revisit daytime first-answer use after three months.
Signals that patients are not accepting it
- Repeat calls within a day about the same request, which suggest the caller did not trust the result.
- Complaints that mention being unable to reach a person.
- A rise in portal messages saying I tried to call.
- More callers pressing zero or asking for a person before the AI finishes its first sentence.
- Staff hearing at check-in that patients found the phone confusing.
Treat these as design feedback, not proof that patients reject AI. Most are fixable with shorter prompts, clearer disclosure or a faster route to staff. For the disclosure question in depth, read should you tell patients they're talking to AI.
Frequently asked questions
Do patients mind talking to an AI on the phone?
Surveys show many people are uncomfortable with AI making medical decisions, but they value quick answers. For office tasks such as booking, acceptance depends mostly on clear disclosure, a fast route to a person and short, accurate calls.
Are older patients less willing to use AI phone systems?
Not necessarily. Older adults adopted telemedicine at higher rates than younger adults in 2021 CDC data. Slower pacing, clear prompts and confirmation of details help every age group.
How can a practice tell whether callers accept its AI voice?
Track requests for a person and when they happen, hang-ups, repeat calls about the same request and complaints, and listen to a weekly sample of calls that went wrong.
References
KFF (2026). KFF Tracking Poll on Health Information and Trust: Use of AI For Health Information and Advice (fielded Feb 24 - Mar 2, 2026) (opens in a new tab) ↑
Pew Research Center (2022). 60% of Americans Would Be Uncomfortable With Provider Relying on AI in Their Own Health Care (opens in a new tab) ↑
American Medical Association (2026). More than 80% of physicians use AI professionally: AMA survey (opens in a new tab) ↑
CDC National Center for Health Statistics (2021). Telemedicine Use Among Adults: United States, 2021 (NCHS Data Brief No. 445) (opens in a new tab) ↑