Surveys show many customers distrust AI in service, mostly because they fear losing a route to a person. Waiting on hold also drives hang-ups. Here is how to weigh both risks.
+ Gartner found 64% of customers would rather companies did not use AI for service, and the top fear is losing access to a person.
+ A 2026 Gartner survey found 87% of customers say an option to reach a human is essential when a company uses generative AI.
+ Long holds and unanswered rings are a separate, well-documented hang-up risk that an agent answering 24/7 can reduce.
+ Fast answers, an honest greeting, a short path to booking and an easy route to a person are the design choices that matter most.
+ Clinical or urgent calls should never be handled by the agent, and Velaire Health routes them to a human.
You want every call answered, but you worry a patient will hear an automated voice, sigh and hang up. Here is what the research says about that worry, and what it does not.
After hours. A patient calls to book.
The desk is dark. The call is gone.
Same call, answered fast.
The same call, with and without an answer.
You've probably lost a booking to a phone that rang out, and now you're weighing an AI phone agent to answer it. The worry is the obvious one: a patient hears an automated voice and hangs up. That worry deserves a straight answer, not a pitch. Here is what the caller research shows, where it runs out, and which design choices decide whether a caller stays on the line.
Some will, and you should plan for it. Surveys show a large share of customers would rather not deal with AI in service, mostly because they fear being blocked from a person. But a phone that rings with nobody there also loses callers. The real question is which risk is bigger at your practice, and what you can control.
The research splits in two. Attitude surveys ask people how they feel about AI in customer service. Behavior data shows how many callers give up when they wait. Neither measures what you want to know: how often a patient hangs up on an AI agent at a med spa or dental practice.
We could not find a published, independent hang-up rate for AI-answered practice calls. If a vendor quotes one, ask who measured it, how, and on which calls.
If someone gives you a precise hang-up rate for AI calls, ask for the method before you trust the number.
What the research does show is why callers distrust automation, and what happens when nobody answers. Side by side, the risk gets easier to weigh.
The biggest fear is not the voice itself. It is being stuck. In Gartner's 2024 survey, customers say their top concern about AI in service is that it will get harder to reach a real person. Next come worries about jobs and about wrong answers. Most of those fears point to something a practice can design around.
The clearest data comes from Gartner. In a July 2024 release, Gartner reported that 64% of customers would prefer companies did not use AI in their customer service [1]. The survey ran in December 2023 with 5,728 customers, and 53% said they would consider switching to a competitor if they found out a company was going to use AI for service [1]. The top concern was that AI would make it harder to reach a person, followed by AI displacing jobs and AI giving wrong answers [1].
What customers say about AI in service
64%
Would prefer companies did not use AI in service
Gartner, December 2023 survey, released July 2024
87%
Say an option to reach a human is essential
Gartner, February and March 2026 survey
50%
Say service is easier when companies use GenAI
Gartner, same 2026 survey
Gartner surveys about two years apart. The picture shifted, but the demand for a route to a person stayed.
About two years later the picture had shifted, though not reversed. A Gartner survey of 3,566 B2B and B2C customers, run in February and March 2026, found that 50% say their service interactions are easier when companies use generative AI. The same survey found 87% say it is essential to offer an option to reach a human agent [2]. Gartner also asked customers who were unwilling to use AI what would change their mind. The most common answer was the ability to switch to a human agent if needed [2].
The same release notes that 58% of customers who use generative AI have used it to complete a task on their behalf, and that customers increasingly expect AI to help them take action, such as booking an appointment [2]. That is close to what a practice asks of a phone agent.
Read together, these numbers say people are not refusing to talk to software. They are refusing to be trapped by it.
These are general customer service surveys, not patient surveys, and they measure stated preference, not behavior on a live call.
Often, yes. A caller who reaches nobody, or waits too long, leaves for a reason that has nothing to do with AI. Vendor benchmarks for healthcare suggest patience runs out within about a minute, and a federal review found even a large health system missing its own answer targets. Your own call data will show which loss is bigger.
Patient Prism, a vendor that sells call analytics to healthcare practices, has compiled abandonment benchmarks from several industry sources. Its summary says that more than 60% of patients in a queue abandon after 1 minute of waiting, and more than 90% abandon by 5 minutes [3]. It also reports an average hold time on connected healthcare calls of 4.4 minutes, against a 50 second answer target it attributes to HFMA [3].
Queued callers who abandon, by wait
After 1 minutemore than 60%
By 5 minutesmore than 90%
Directional figures compiled by Patient Prism, a call analytics vendor, from other industry sources.
Treat those figures as directional. They are a vendor's compilation of other studies, mostly from other vendors. The shape of the curve is the point: callers give up fast, and a busy or empty front desk is where many of them are lost.
An independent benchmark comes from the VA Office of Inspector General, whose October 2025 report reviewed the Veterans Health Administration's clinical contact centers [4]. VHA set two standards: no more than 5% of calls abandoned, and at least 80% of primary care scheduling and pharmacy calls answered within 30 seconds [4]. In fiscal year 2024, scheduling services had an estimated 11% of calls abandoned and did not meet the timeliness standard [4]. Medical facilities that kept their own local phone queues left about 222,000 of 2.3 million calls unanswered, a rate of about 10% [4].
A 2022 study of 6 Veterans Affairs medical centers chosen for above-average primary care access records the standard those sites were held to: an average speed of answer of 30 seconds or less, and a call abandonment rate under 5% [5].
Those are large federal systems, not a med spa, so do not read their rates as yours. They do show that a 30 second answer target for scheduling calls is a real standard, and that a staffed operation can still miss it.
So the comparison is this. An agent that answers 24/7 removes the hold queue and the after-hours ring-out, and adds a new risk: some callers will not like software. The trade is only worth it if the second risk is clearly smaller.
The alternative to an AI agent is rarely a person on the first ring. It is usually hold music, voicemail or nothing.
Four choices matter most. Answer quickly. Say plainly what the caller is talking to. Get to the booking without a long menu. Make a person easy to reach. Each one maps to a fear or a failure the research names, and each one is something you can check in a demo before a single patient hears the agent.
Answer fast. The abandonment curve punishes delay. Whatever answers should pick up quickly at midday on a Monday and in the middle of the night on a Sunday.
Say what it is. A caller who discovers halfway through that the "receptionist" is software may feel tricked, and trust is exactly what is at stake. A short greeting that says the caller is talking to an automated agent, what it can do, and how to reach a person sets expectations early. Ask any vendor, including Velaire Health, to play you the exact greeting before launch.
Get to the point. Most booking calls are simple. The caller wants a time. Velaire Health checks live availability, books the appointment into the practice calendar and texts the patient the time. A short path to a confirmed slot gives the caller a reason to stay.
Keep a person one step away. Gartner's data says the route to a human is the thing most likely to change an unwilling customer's mind [2]. Gartner also warns against making AI a mandatory first step for every issue: customers forced through several failed AI exchanges before reaching a person are less likely to use the tool again [2]. The route needs to be short, spoken plainly, and work every time.
Two versions of the same call
Loses the caller
Keeps the caller
Pickup
Rings out or goes to hold
Answered quickly, day or night
Greeting
Voice pretends to be a person
Plain greeting that says what it is
Path
Long menu before anything useful
Goes straight to the booking
Exit
No clear way to reach the team
A person is one request away
Each difference maps to a fear or a failure the research names.
Because it is the condition of trust. The surveys show people accept automation when they know they can leave it. Remove that exit and even a capable agent feels like a wall. Keep it, and many callers will let the agent book them, since that is often the faster path to what they wanted.
Gartner's 2024 release put the fear directly: many customers worry that generative AI will become another obstacle between them and an agent [1]. Its 2026 survey found customers push back when AI acts as a barrier to reaching a human [2]. The agent should never argue with a caller who wants a person, never loop them through the same question, and never hide the option.
Be honest about what the agent will not do well. A caller who wants to chat, or has a billing question that needs judgment, is better served by your team. Gartner's guidance points the same way: automation works best when it attempts resolution only when confidence is high, with a clear path to human support [2].
In practice, the route to a person looks different by time of day. During office hours it may be a transfer to the front desk. After hours it may be a clear message and a callback. Either can work if the caller hears exactly what will happen next.
A useful test is to call your own line and ask for a person in three different ways: "Can I talk to someone?", "Operator", and "I want a real person." If any of those fails, fix it before launch.
Anything clinical or urgent. A caller with a symptom, a reaction, a worry about a recent treatment or any urgent question should go to a human, not to a booking flow. This is a safety rule first and a trust rule second. Software that sounds like it is giving clinical answers puts the patient and the practice at risk.
Velaire Health never diagnoses and gives no medical advice. A caller with a clinical or urgent question is never booked by the agent; the call routes to a human. That carve-out should be configured and tested before go-live, using the kinds of phrases your patients actually use, not just textbook terms.
The agent's job is the front office: answering, checking availability, booking and confirming by text. It is not a nurse line and should never sound like one.
If you want the wider picture of where front-desk calls go missing and what each one is worth, our post on the real cost of a missed call covers the operational side.
Compare the hang-ups you already have with the ones an agent might add. If calls ring out after hours or wait on hold at peak times, you are losing callers today. An agent trades that loss for a trust risk, and good design shrinks that risk. Measure both, before and after launch, on the same basis.
Your phone system or carrier can usually show missed, abandoned and after-hours calls. A simple way to frame them: take the share of calls that went unanswered or abandoned last month, and set it beside the VHA standard of no more than 5% abandoned. If you are well above that line, the hold queue deserves attention first. Then set up the agent so it can be judged on the same terms.
Caller worry
Design choice that answers it
What Velaire Health does
"Nobody will pick up"
Answer fast at all hours
Answers calls 24/7
"I'll be stuck with a bot"
Clear, short route to a person
Ask to hear the route before launch
"It won't get my booking right"
Check real availability, confirm in writing
Checks live availability, books into the practice calendar, texts the patient the time
"It will try to answer medical questions"
Hard carve-out for clinical topics
Routes clinical or urgent calls to a human; never diagnoses
"Setup will disrupt the office"
Short setup, no new equipment
Under 48 hour setup, no new hardware
Before you switch it on
✓How fast is the call picked up, at peak hours and overnight?
✓Does the greeting say the caller is talking to an automated agent?
✓What exact words route a caller to a person?
✓Which topics are always routed to a human, never booked?
✓Does the patient get a text confirming the time?
✓Can you review call outcomes, including hang-ups?
Questions to put to any vendor, and to your own setup, before a patient hears the agent.
Judge the agent on booked appointments and caller hang-ups together, not on either one alone.
A few weeks of review after launch will tell you more than any survey. Listen to a sample of short calls and find the moment callers dropped. If the problem is the greeting, change the greeting. If callers ask for a person and struggle to reach one, fix the route. Our post on why after-hours calls are different covers how to size the gap you are trying to close.
The honest summary: some patients will hang up on an AI, mostly from feeling trapped, and you can design against that. Weigh it against the callers you already lose to silence and hold music, then decide with your own call data in front of you.
FAQs
Questions about this
Some will. Surveys show many people would prefer a human, and the biggest fear is being trapped with no way to reach one. We found no published, independent hang-up rate for AI-answered calls at medical practices, so treat any vendor claim of a precise rate with care and ask where it came from.
We think so. A caller who finds out mid-call that they are not talking to a person may feel misled, and trust is the thing at stake. A short, plain greeting that says what the agent can do, and how to reach a person, sets expectations before the caller has to guess.
Gartner found the ability to switch to a human was the most common thing that would change an unwilling customer's mind about AI. Your setup should make that route short and obvious, and it should work every time. Before launch, call your own line and ask for a person in a few different ways, then fix any phrasing that fails.
No. Velaire Health never diagnoses and gives no medical advice. A caller with a clinical or urgent question is not booked by the agent and is routed to a human instead. That limit protects the patient and the practice, and it should be configured and tested before the agent takes a single live call.
Both lose the booking. The difference is the cause. A caller who hangs up on hold left because nobody answered. A caller who hangs up on an agent left because of trust or friction. The second is a design problem you can work on: speed, honesty about what the agent is, and a clear route to a person.
Builds AI front-office systems for medical and aesthetic practices. Posts here start from published sources rather than vendor claims, and every number links back to where it came from.
What published research can and cannot tell you about replying to new inquiries, and a target your front desk can hold for live calls, callbacks and web forms.
Clinical and urgent calls are never booked by Velaire Health. They go to a person. Here is why that line exists, and how to write the handoff so an urgent caller never waits in a queue.
Voicemail looks like a safety net and behaves like an unmanaged work queue. Here is why the ownership problem is structural, what the alternatives actually change, and the number that settles it.
One is capturing discretionary demand that decides in the moment. The other is refilling a chair from a patient list it already owns. Same complaint, opposite problem, and the fix differs at every layer.
A missed call at 2pm and a missed call at 8pm are not the same problem. One is a delay you recover from. The other is usually just gone, and nobody is there to see it happen.
Vendor benchmarks disagree on when dental patients call. Here is how to read your own call log, map your peak hours by day, and see where the desk overflows.
The plain-language explainer: what it does, what it deliberately doesn't, and how it compares with voicemail, an answering service, a phone tree, and hiring.
