Do Patients Trust AI in Healthcare? Evidence Reveals Insights

There's a headline that keeps circulating in healthcare trade press — some version of "patients still want people." A recent piece in the American Chiropractic Association's publication framed it that way, and it resonates because it feels true. But "patients want people" is too blunt to act on. It's not that patients reject automation — most use a self-checkout, book flights without a human, and refill prescriptions through an app. The real finding from the research is more specific, and far more useful if you run a medical practice: patients are comfortable handing routine logistics to a machine, and they want a person reachable the moment something matters clinically.

Get that distinction right and AI is an asset patients appreciate. Get it wrong — by burying the path to a human, or pointing automation at clinical judgment it shouldn't touch — and you erode the trust your practice runs on. Here's what the evidence actually shows, and what it means for the one place most practices deploy AI first: the phone.

What patients accept — and what they don't

The cleanest single data point comes from a 2024 study in BMC Medical Ethics, a mixed-methods survey of 600 U.S. adults. It asked about AI across a spectrum of healthcare tasks, and the split was stark. 84.2% of respondents were comfortable with AI handling administrative work like scheduling appointments and follow-ups. Comfort for intake data entry sat around 61%. But when the tasks turned clinical, acceptance dropped sharply: roughly half were comfortable with AI assisting a diagnosis, about 45% with AI recommending medication, and only a third with AI administering treatment. As the authors put it, respondents "were more comfortable with the use of AI in health-related tasks that were not associated with doctor-patient relationships, such as scheduling patient appointments or follow-ups."

That gradient is the whole story. The same person who happily lets software book their appointment gets uneasy when software starts making medical decisions. And the discomfort on the clinical side is not a fringe view. The Pew Research Center surveyed a nationally representative sample of 11,004 U.S. adults and found that "six-in-ten U.S. adults say they would feel uncomfortable if their own health care provider relied on artificial intelligence to do things like diagnose disease and recommend treatments." Only 39% said they'd be comfortable with it.

Put the two studies together and the message to a practice is clear: there is broad permission to automate the administrative layer, and real resistance to automating clinical judgment. The patients are drawing exactly the line a careful practice should already be drawing.

The pattern isn't "anti-AI" — it's "keep a human reachable"

It would be easy to misread the clinical caution as technophobia. It isn't. When researchers look closely, what patients consistently want is not the absence of AI but the presence of a human — especially for anything sensitive, urgent, or emotionally charged.

A 2024 study in JMIR Human Factors (888 respondents) found that willingness to use a health chatbot tracked the stakes of the conversation. For emotionally sensitive topics like mental health, the preference for a person over a bot was overwhelming — on the order of 81% favoring a person versus a low single-digit share preferring a bot — while openness to automation rose for lower-stakes, routine intake. A 2025 qualitative meta-synthesis in JMIR AI, reviewing primary-care patients across multiple studies, reached a compatible conclusion: patients saw clear value in AI for "clinical documentation, practice operations… and triage, with the potential to reduce administrative tasks," while doubting that AI could or should replace a physician's judgment.

And when patients are asked to rank arrangements head-to-head, the ordering is remarkably stable. A 2024 experiment in Frontiers in Psychology (1,183 participants) found that "people prefer a human doctor, followed by a human doctor with an AI system, and an AI system alone came in last place." The human-plus-AI blend beat AI alone by a wide margin — patients don't object to AI in the loop, they object to AI instead of the loop. Trust did most of the explaining.

Even the trade-press piece that started this — written, it's worth noting, by someone affiliated with a human-receptionist service, so read its numbers as marketing rather than research — lands on the same human-in-the-loop instinct: its most defensible claim is that patients want a human to review what the automation does. That's not an argument against AI. It's an argument for designing AI so a person is always within reach.

What the evidence implies for design

The design takeaway follows directly from the data: let automation own the routine, high-volume work — scheduling, refills, reminders — and keep the path to a person effortless for everything else. A fully autonomous "AI receptionist that handles everything" is the configuration patients trust least, because it does the one thing they're most wary of (clinical judgment by machine) while removing the one thing they most want (a reachable person).

We've argued the mechanics of that design elsewhere, and won't rehearse them here — the architectural case in autonomous AI voice agents vs. hybrid AI, the operational question of which patient calls can be automated and which still need a human, and the patient-experience case in why AI front desks fail patients. This piece stays on the demand side: not what's technically possible or operationally efficient, but what patients themselves, in study after study, say they actually want.

What "automate the routine, preserve the human" looks like in practice

This is the principle CallMyDoc is built around, and it's worth being concrete about how it differs from an "AI receptionist."

When a patient calls, the system identifies them, captures what they need, and routes by what the patient selects — not by any attempt to interpret their symptoms. A refill, an appointment change, a billing question, a form request: these route and, where possible, resolve automatically, because that's the routine administrative work patients are happy to have handled fast. Crucially, the system never assesses a caller's medical condition or decides clinical urgency on its own. If a patient indicates their need is urgent, the call is flagged and a real person — the on-call provider or staff — is notified, with every interaction documented in the EHR. The standing instruction for a true emergency is always the same and stated up front: hang up and call 911.

That design is why patients don't experience it as "fighting a robot." The routine gets faster, the human stays one step away, and nothing about a clinical decision is left to a machine. It's also why this works as after-hours coverage: the routine after-hours calls get handled without waking anyone, while the calls a patient marks urgent reach a provider quickly — a median of about 11 minutes via the mobile app, versus the 45–90 minutes typical of a traditional answering service. Across more than 27 million patient calls in 40 states, roughly 47% of inbound calls are fully automatable — which is exactly the routine-administrative share the research says patients are glad to automate. The other half, the part that needs a person, still gets one.

(One scope note for accuracy: AI-driven self-scheduling that writes back to the calendar is available today for athenahealth practices; Veradigm and Altera TouchWorks practices get the call-automation and routing described here. We'd rather state the boundary than blur it.)

An honest caveat about the evidence

It's worth being precise about what the research does and doesn't establish, because precision is part of trustworthiness. No single, definitive randomized trial has proven the rule "patients accept AI best when it handles routine work and preserves human access." That principle is a synthesis — it emerges consistently across the surveys and studies above, but it's assembled from multiple bodies of evidence rather than handed down by one landmark experiment. The individual findings are solid and peer-reviewed; the unifying principle is an inference drawn from their convergence.

We think that's still the right thing to design around. When independent studies using different methods, populations, and questions keep pointing the same direction — administrative AI welcome, clinical AI cautious, human access non-negotiable — that convergence is itself a kind of evidence. And it aligns with the simpler truth practices see every day: patients don't mind technology that saves them time, as long as a person is there when they actually need one.

The takeaway for your practice

"Patients want people" is half the lesson. The fuller version — the one you can build a phone strategy on — is that patients want the routine handled effortlessly and a person reachable for everything else. Practices that automate the administrative load while keeping the human path obvious get the efficiency without the trust cost. Practices that deploy a do-everything autonomous bot get the efficiency and pay for it in the exact currency the research says matters most: patient confidence.

If you want to see what "automate the routine, preserve the human" looks like on your own phone line — which calls resolve themselves, which reach a person, and how every one is documented — Book a demo →. We'll walk through how it would handle a real day of your call volume, and where the human stays in the loop by design.