Key takeaways
- UK adults are cautiously positive: about half expect AI to improve healthcare, yet healthcare is also where they most want AI carefully managed.
- Direct UK evidence on AI answering practice phones is thin, so the safest guide is what drives trust in data and AI generally: openness, control and a human route.
- Keep an AI phone assistant to admin and signposting; clinical triage raises safety and medical device questions.
- Say it is automated at the start, offer a person at any point and tell callers exactly what will happen next.
What the evidence does and does not tell us
Practice managers considering an AI phone assistant usually ask one question first: will patients accept it? There is surprisingly little UK survey data that asks patients specifically about AI answering a GP or dental phone line. What exists is broader evidence about public attitudes to AI and data in healthcare, plus clinicians' own views. Read together, they point clearly to the design choices that matter.
The government's public attitudes tracker found that 52% of UK adults expect AI to have a positive impact on healthcare[1]. At the same time, healthcare was the area people most often said needs careful management of AI to prevent negative outcomes, chosen by 29%[1]. Optimism and caution sit side by side.
Familiarity is rising
Patients are not meeting AI for the first time on your phone line. Four in ten UK adults use AI chatbots for personal purposes at least once a month, up from 34% in the previous wave of the tracker[1]. Many callers will already have a view about what automated assistants do well, such as answering simple questions at any hour, and what they do badly, such as handling anything unusual.
40%
of UK adults use AI chatbots for personal purposes at least monthly[1]
Trust in the NHS itself remains high: 85% of UK adults trust it to act in the public's best interest, the highest of any organisation tested[1]. But trust in the NHS to be open and transparent about how it uses data was lower, at 69%[1]. Patients extend goodwill to their practice; an AI assistant that feels evasive spends it.
Clinicians share the same worries
GPs' views mirror the public's. In a 2025 Nuffield Trust and RCGP survey, 28% of GPs said they used AI tools in clinical practice[2]. Among those who did not, 89% were concerned about professional liability and medico-legal issues, 88% about a lack of regulatory oversight and 82% about patient privacy and data security[2]. If the clinical team is uneasy, patients will hear it at the front desk.
There is also a regulatory line to respect. The MHRA says many software and AI products used in health and social care are regulated as medical devices[3]. An assistant that assesses symptoms or decides urgency is in very different territory from one that answers questions about opening hours or explains how to request a repeat prescription.
Design choices that build or break trust
| Design choice | Builds trust when | Breaks trust when |
|---|---|---|
| Disclosure | The assistant says it is automated in its first sentence | Callers find out halfway through, or never |
| Human route | Saying 'reception' or pressing a key reaches a person or a call-back | The caller is looped back to the menu |
| Scope | It handles admin questions and signposting only | It appears to judge symptoms or urgency |
| Urgent help | Emergency and NHS 111 advice comes early and clearly | Urgent advice is buried after routine options |
| Next steps | It confirms what will happen and when, in plain words | The call ends without a clear outcome |
| Data | Recording and data use are explained briefly, with a link to the privacy notice | Callers are not told the call is recorded or transcribed |
If you can only get one row right, make it the human route. Patients forgive an assistant that cannot answer a question; they do not forgive one that stops them reaching a person. Test the route from every point in the call flow, including after a misunderstanding, and make sure it still works when reception is busy, for example by offering a call-back rather than a long hold.
A greeting that sets expectations
The first ten seconds decide most of the caller's attitude. A clear opening does three jobs: says what the assistant is, says what it can do and gives a way out.
Example greeting: 'Thank you for calling Riverside Surgery. I am the surgery's automated assistant. I can help with opening times, how to request appointments and prescriptions, and passing a message to the team. If you need a person, just say reception. If this is a medical emergency, hang up and dial 999.'
Ava, for example, is set up to answer practice information questions, signpost urgent help and pass anything about a patient's care to the reception team. Whatever system you use, test the greeting with patients of different ages and with callers whose first language is not English, and time how long it takes before the caller reaches the thing they rang about.
Being open about data on the call
Concern about AI in health care is closely bound up with concern about data. In the same government tracker, 57% of adults identified data not being held securely, or being hacked or stolen, as a risk[1]. Only 35% agreed that people have control over who uses their data and how[1]. An AI phone assistant that records or transcribes calls touches both worries.
Practical openness is simple. Mention recording or transcription in the greeting in one short sentence, point to the privacy notice on your website and make sure the notice explains what the assistant does with call content, who the supplier is, where data is stored and for how long. Keep messages taken by the assistant inside your normal systems, with the same access controls as other patient contacts, rather than in a separate inbox nobody owns.
Staff should be able to answer the follow-up question every receptionist will eventually hear: 'Was that a real person, and who heard my message?' A one-line answer agreed in advance avoids hesitation that patients read as evasion.
Groups to design for deliberately
- Older callers who prefer to speak to someone they know: make the human route obvious and fast.
- People with hearing loss or speech differences: offer text-based alternatives and avoid speech-only menus.
- Callers with limited English: keep sentences short and allow a request for an interpreter.
- Anxious or distressed callers: recognise words such as 'worried' or 'pain' and hand over promptly.
- Carers and proxies: explain how to call on someone else's behalf without collecting more data than needed.
Testing acceptance before go-live
Tell your patient group first
Explain what the assistant will and will not do, play the greeting and ask for reactions before anything goes live.
Start with out-of-hours or overflow
Patients judge a new system more kindly when it replaces voicemail or a long queue rather than a person they used to reach.
Measure the human hand-off
Track how often callers ask for a person and how long it takes to reach one. A rising rate signals a design problem.
Read complaints and comments weekly
Look for words like 'robot', 'couldn't get through' and 'didn't understand', and fix the call flow behind each.
Publish what you learned
A short note to patients after three months, saying what changed, signals that their views shaped the service.
Example: overflow first, then daytime
Example (fictional): a three-site practice introduced an automated assistant only for calls that would otherwise reach voicemail after 6.30pm and for the morning overflow queue. The assistant answered information questions and took messages for reception. In the first month, about one caller in six asked for a person; the practice reworded its greeting to explain the call-back option more clearly, and the rate fell.
The patient group asked for one change: a clear statement that the assistant never gives medical advice. The practice added it to the greeting and the website. Only after three months of stable feedback did it extend the assistant to daytime information calls.
The bottom line for practices
Government plans already point towards more digital routes into general practice, including digital telephony and AI-powered online advice in the NHS App[4]. Patients will meet more automation, not less. The practices that keep patients' trust will be those that are open about it, keep it within safe limits and make a person easy to reach. Map where it fits using patient journey mapping, and check it with a mystery patient audit once it is live.
Above all, decide in advance what the assistant must never do, write it down and share it with the team and the patient group: no clinical advice, no decisions about urgency, no promises about appointments it cannot book. Clear limits make it easier for staff to support the change, and easier for patients to trust what the assistant does say.
Frequently asked questions
Do patients mind talking to an AI on the phone?
UK evidence specific to practice phone lines is limited. Broader surveys show cautious optimism: 52% of adults expect AI to benefit healthcare, while healthcare is where most people want AI carefully managed[1]. Openness and a fast route to a person matter most.
Should an AI phone assistant say it is not a person?
Yes. Saying so in the first sentence avoids callers feeling misled and sets realistic expectations about what it can do.
Can an AI assistant triage symptoms?
That moves it into clinical territory. The MHRA says many AI products used in health care are regulated as medical devices[3]. Keep phone assistants to admin and signposting unless a product is approved for clinical use.
How do we know if patients accept it?
Track requests for a person, time to reach one, complaints mentioning the assistant and feedback from your patient participation group.
References
GOV.UK (Department for Science, Innovation and Technology) (2024). Public attitudes to data and AI: Tracker survey (Wave 4) report (opens in a new tab) ↑
Nuffield Trust (2025). How are GPs using AI? Insights from the front line (opens in a new tab) ↑
GOV.UK (Medicines and Healthcare products Regulatory Agency) (2025). Software and artificial intelligence (AI) as a medical device (opens in a new tab) ↑
GOV.UK (Department of Health and Social Care) (2025). Fit for the future: 10 Year Health Plan for England (opens in a new tab) ↑