The 15-Minute Rule
New data for healthtechs on what people will accept from AI support
Founder's noteThe clock starts when someone asks
We keep running research on how end-consumers feel about AI because we believe that ensuring they have a great experience is key to making our customers (we call them subscribers) successful. This report extends our research into support in healthcare settings, where people are often worried, in need of care, short on time, or some combination of the three!
Healthtechs are some of the companies where support is often the first part of their service a patient or member deals with. AI can be a double-edged sword: it enables fast 24/7 support that improves patient experiences at lower cost. But if it misfires in a critical moment it can deeply undermine the relationship. We wanted to know what people will accept from AI in this setting, so we asked over 2,000 people across the US, UK and Australia.
We asked people how they would feel about AI answering a simple question sent through their healthcare provider's patient portal, such as whether a prescription has shipped. How quickly the answer arrived made the biggest difference. 58% were comfortable with an instant AI answer. When the same answer took 15 minutes, that fell to 20%. After an hour, it was 9%. Being able to reach a person helped, but it didn't make up for a slow answer.
Most people are comfortable letting AI handle routine healthcare tasks. 82% are comfortable with it booking appointments and 70% with billing. Prescription refills divided opinion: 44% were comfortable and 42% were not, even though the AI wasn't deciding anything about the medication. We asked people to pick the one thing that would most increase their willingness to use AI. The top answer was evidence that it's accurate, with clear communication about its limits (26%). Being told they were talking to AI was the least popular of the five named options, picked by just 9%.
For healthtechs, the longer an AI takes to respond, the less likely people are to accept its answer. The rest of this report shows which tasks people are ready to hand to AI, and what makes them comfortable doing it.
"58% accepted an instant AI answer. Make them wait 15 minutes and it falls to 20%."
Steve Hind · Co-founder and CEO, LorikeetIntroductionHealthtech's front door is now AI
Healthtechs compete on speed and convenience. Telehealth, virtual care and online pharmacy companies promise care without the waiting room, and AI support is how many of them answer the questions that come with it: booking, billing, account changes and refills.
To see what people will accept from that AI, we surveyed over 2,000 people across the US, UK and Australia. We asked how acceptable AI help would be for five health support tasks, how response time and access to a person change that, and what would most increase their willingness to use AI in health support.
Most people are comfortable using AI for routine health tasks. How fast it responds matters most. 58% accepted an instant AI answer, 20% a 15-minute one and 9% one that took an hour. A way to reach a person lifted the 15-minute figure to 46%. Scheduling and billing are widely accepted. Prescription refills divide people. And the strongest reassurance is an AI that is accurate and clear about its limits.
58%
accept an instant AI answer, even with no way to reach a person. Make them wait 15 minutes and it drops to 20%
82%
accept AI help with appointment scheduling
26%
chose accuracy with clear communication about its limits as the thing that would most increase their willingness to use AI
What's new
This is Lorikeet's fourth consumer study in 2026, and the first about health support.
Key takeawaysSeven numbers that matter
58%
accept an instant AI answer, even with no way to reach a person. At a 15-minute wait it falls to 20%
More than 2x
as many people accepted a 15-minute AI reply when they could also reach a person (46%). Without that option, only 20% accepted a 15-minute AI reply
9%
accept an AI answer that takes an hour. With a person on call it's 22%
82%
accept AI help with appointment scheduling. For prescription refills it's 44%
26%
chose accuracy with clear communication about its limits as the thing that would most increase their willingness to use AI. Being told they were talking to AI was the least popular of the five named options, picked by just 9%.
51%
went straight to a doctor or care team when an AI didn't fully answer their health question
2.2x
as likely to report using AI health support, among people who used telehealth in the past year, compared with those who didn't
Finding 01The 15-minute rule
accept an instant AI answer, even with no way to reach a person
Illustrative
It's 9pm and a patient opens their portal to ask whether their prescription has shipped. If the answer comes back in seconds, most people don't mind that it came from AI. If it takes 15 minutes, fewer than half the people would find it acceptable, even with a person on call.
We asked people how acceptable it would be for an AI to answer a simple question in a patient portal, and tested different reply times. The speed to respond was critical. 58.2% accepted an instant AI reply with no way to reach a person. When the reply took 15 minutes, acceptance fell to 20.2%.
An hour was a hard no. Even with a person on call, 21.7% accepted the wait and 69.9% called it unacceptable.
Acceptance of an AI reply, by response time
A simple, nonclinical question in a patient portal. Half the people we asked could also reach a person.
Proof point
At Eucalyptus, now part of Hims & Hers, Lorikeet cut response time from 20 hours to 90 seconds and scaled patient support 4x without adding headcount.
Finding 02A reachable person still matters
as many people accept a 15-minute wait when they know they can reach a person
Illustrative
The same patient is still waiting for an answer. A visible "talk to a person" option changes how the wait feels, even when the reply takes just as long.
When the AI reply took 15 minutes, people were more than twice as likely to accept it if they could also choose to reach a person: 45.5% with that option, 20.2% without.
The human option raised acceptance at every wait. It added 21.0 points to an instant reply, 25.4 points at 15 minutes and 12.4 points at one hour. The lift peaks at 15 minutes (25.4 points) and halves by one hour (12.4). At one hour, the best result was 21.7%.
Points added by offering a way to reach a person
Finding 03Admin is accepted. Refills divide people.
accept AI help with appointment scheduling
Illustrative
A GLP-1 member messages their clinic twice in one week. First to move a check-in, then to ask for their next refill. To the clinic, both are admin. To the member, the second one feels medical.
82.0% of people accept AI help with appointment scheduling. For prescription refills the figure is 44.2%, with 42.1% calling it unacceptable.
That is a 37.7-point gap between scheduling and refills. Billing sits at 69.5%, non-diagnostic symptom support at 67.4% and help with test results at 66.5%. People were divided on refills even though the question made clear the AI would not decide whether a medication was appropriate.
Share who say AI help is acceptable, by task
Our interpretation (the survey did not ask why): a refill is mostly admin, but people treat it as clinical. The line people draw seems to be whether a task feels medical.
Finding 04Honesty about limits wins people over
chose evidence of accuracy, with clear communication about its limits
Illustrative
A member asks whether a side effect is normal. The reply that builds trust says what it can tell them, what it can't, and how to reach their clinician now.
Asked to pick the one thing that would most increase their willingness to use AI in health support, the top answer was evidence of accuracy with clear communication about its limits (26.3%).
Being told they were talking to AI was the least popular of the five named options, picked by 9.2%.
What would most increase your willingness to use AI in health support? (choose one)
Other answers: 8.9%.
Finding 05When AI doesn't resolve it, the work lands on the care team
went first to a healthcare professional when AI didn't fully answer their question
Illustrative
When the AI can't answer, the question doesn't go away. It turns into a call to the clinic or a message to the care team.
51.1% of people whose AI health-support interaction left a question unresolved went first to a healthcare professional, by phone or patient portal.
15.2% tried another AI tool, website or digital service. 11.4% had a telehealth or in-person visit.
First step after an AI interaction that left a question unresolved
This question covered any AI health support, including tools people use on their own, so it does not isolate AI a provider deployed.
Finding 06Telehealth users are already on board
as likely to report using AI health support, among people who used telehealth in the past year
People who used telehealth in the past year were 2.2 times as likely to report using AI health support: 62.7%, against 28.6% of people who had not used telehealth.
This is the one result specific to digital health users.
A correlation between two self-reported measures.
In practiceWhat this looks like at a healthtech
The data describes what people will accept. Here is what happens when a healthtech deploys AI support.
What it meansWhat healthtech leaders should do now
What good AI health support looks like
It answers in seconds. It handles the admin work people already accept from AI: scheduling, billing and account questions. It says what it can't do and where to go next. And a person on the care or support team is always within reach, with an easy way to get there.
01
Measure AI response time in seconds
A 15-minute wait cut acceptance of an AI reply from 58.2% to 20.2%. After an hour it was 9.3%. A way to reach a person lifted the 15-minute figure to 45.5%.
02
Keep a person reachable, and show people how to get there
The option to reach a person lifted acceptance at every wait, by up to 25.4 points. Make the handoff visible from the first reply.
03
Start with scheduling, billing and account questions
82.0% accept AI for scheduling and 69.5% for billing. This is the admin volume that fills a healthtech support queue.
04
Build refills with clear limits and a fast handoff
Refills divide people (44.2% accept, 42.1% don't), even when the AI makes no clinical call. Say exactly what the AI does, and route anything clinical to a person.
05
Have the AI say what it can't do
Accuracy with clear communication about limits was the top reassurance (26.3%). Being told they were talking to AI was the least popular of the five named options (9.2%). When the AI can't resolve a question, 51.1% go straight to a clinician, so tell people where to go next.
Ask Claude about the report
Add the Claude skill and ask your own questions, or describe your AI support setup and get a readiness scorecard.
Get the Claude skill →Lorikeet for healthcare
Concierge-level care experiences for every patient and member.
Lorikeet is the AI customer support platform built for healthcare. It books appointments, verifies benefits and refills prescriptions inside your rules, tells patients what it can't do, and hands off to your care team with the full context. HIPAA compliant, with BAAs and audit trails.