Key takeaways
- AI call handling is best suited today to non-clinical work: practice information, explaining routes, signposting urgent help and passing requests to staff.
- It should not triage, give clinical advice or decide who gets an appointment without formal clinical safety assurance.
- Before any pilot, complete a DPIA and a clinical safety assessment, agree a data processing agreement and tell patients.
- GPs' main worries are liability, regulation and privacy; a pilot plan should answer each one in writing.
Where AI fits in the practice front door
AI is already part of general practice. In a 2025 Nuffield Trust and RCGP survey of 2,108 GPs, 28% said they currently use AI tools in clinical practice, most often for documentation and note-taking[1]. Phone answering is a newer use. The government's 10 Year Health Plan says it will use digital telephony to ensure all GP practice phones are answered quickly and build AI-powered online advice into the NHS App to end the 8am scramble[2].
For a practice manager, the practical question is narrower: which calls could an AI system handle safely now, and what has to be in place first? The answer depends far more on the task than on the technology.
Start with the problem rather than the product. Pull a week of call data and a sample of calls, and sort them by reason. If a large share are information questions, such as opening hours, how to order medication or whether results are back, there is a clear case for automation. If most calls need a clinical decision or a booking, AI call handling alone will not change much, and the effort is better spent on capacity, triage and online routes.
What AI call handling can and cannot do today
| Task | Suitable now? | Safeguards |
|---|---|---|
| Opening hours, address, parking, closures | Yes | Practice-approved wording; updated when hours change |
| Explaining routes: online form, NHS App, phone times | Yes | Explains the route; does not decide which route a patient must use |
| Signposting 999 and NHS 111 | Yes, with agreed wording | Wording agreed by clinicians; triggers on red-flag phrases; never delays |
| Taking a message for reception (non-clinical) | Yes | Clear hand-off; staff action within a stated time |
| Overflow at peak times and after hours | Yes, for the tasks above | Callers can still reach a person when the practice is open |
| Booking appointments into the clinical system | Only with full assurance | Clinical safety case, DPIA, integration approval |
| Collecting symptoms or triaging | No, not without regulated, assured tools | May be a medical device; needs formal clinical safety work |
| Giving clinical advice | No | Clinicians only |
The line between information and triage matters legally as well as clinically. The MHRA says many software and AI products used in health and social care are regulated in the UK as medical devices[3]. A system that decides how urgent a caller's symptoms are, or which clinician they should see, is in different territory from one that reads out opening hours.
What GPs and patients worry about
Among GPs not using AI, the biggest concerns were professional liability and medico-legal issues (89%), lack of regulatory oversight (88%) and patient privacy and data security (82%)[4]. Those are reasonable concerns for a phone system that patients will speak to about their health.
Patients are cautiously open. In the government's public attitudes tracker, 52% of UK adults expected AI to have a positive impact on healthcare, but healthcare was also the area where people most often said AI needs careful management to prevent negative outcomes[5]. Patients accept AI more readily when they are told what it is doing, can reach a person and know their data is protected.
Governance before a pilot
Define the scope in writing
List exactly which call types the AI will handle and which it will pass to staff. Start with information and routing only.
Complete a DPIA
Calls include health information. Record what data the system hears, stores and transcribes, where it is processed, for how long and who can access it.
Carry out clinical safety assessment
Under the NHS clinical risk management standard for deploying health IT, DCB0160, the practice or its ICB should assess hazards such as missed urgent cues. Ask the supplier for its DCB0129 safety case.
Agree a data processing agreement
Cover retention, deletion, sub-processors, location of processing, breach notification and what happens at contract end.
Tell patients
Say in the greeting that an automated assistant may answer, how to reach a person, and that calls are recorded. Update the privacy notice.
Plan monitoring
Review a sample of calls and transcripts each week, especially any involving urgent signposting, and keep a log of issues and fixes.
Our guides to safety rules for AI handling patient calls and connecting AI calls to clinical systems go into more detail on the safety case and integration.
Questions to ask any supplier
- Exactly what does the system do on a call, and what does it hand to staff?
- How does it detect emergencies, and what does the caller hear when it does?
- Does it ever collect symptoms or give advice? If so, what is its regulatory status?
- Where are recordings and transcripts stored and processed, and for how long?
- Can we see and change every piece of wording it uses with patients?
- How does a caller reach a person, and how quickly?
- What clinical safety documentation can you provide?
- What happens when it does not understand a caller, such as a strong accent, background noise or another language?
- Which other NHS practices use it for the same tasks, and can we speak to them?
Bringing the reception team with you
Reception staff will reasonably ask whether AI is coming for their jobs. Answer that directly before the pilot, not after rumours start. In most practices the honest answer is that AI takes repetitive information calls at peak times, so that people can spend longer on the calls that need judgement, kindness and local knowledge.
- Involve the reception lead in choosing the scope and writing the wording the system uses.
- Show staff transcripts from the pilot, including the mistakes and how they were fixed.
- Make clear who owns the call-back list and messages the system creates.
- Ask the team what they would most like taken off their plate, and test that first.
Staff are also your best safety monitors. They hear from patients when something went wrong on an automated call. Give them a simple way to log those reports.
Example: a cautious pilot
Example: a fictional practice, Ashgrove Surgery, piloted AI call handling for overflow calls only, between 8am and 9.30am and after 6.30pm. The scope was practice information, explaining the online form and NHS App, urgent signposting with clinician-approved wording, and taking non-clinical messages for reception. Anything patient-specific went to the call-back list.
The practice completed a DPIA and a clinical safety hazard log before go-live, updated its greeting and privacy notice, and had the reception lead review twenty transcripts a week. In the first month, the reviews found two calls where the assistant's urgent wording was correct but slower than the practice wanted; the clinical lead shortened the trigger list. The practice decided to continue for information and overflow only and to revisit booking once integration and assurance were available.
Patients who struggle with automated calls
Automated voices can be harder for some patients: people with hearing loss, speech differences, dementia, limited English or high anxiety. Build a fast route to a person into the design, such as saying 'reception' or pressing 0 at any point, and check that it works in practice. Patients with a recorded communication need may be better routed straight to staff. Watch abandonment and complaints from these groups closely during any pilot, and see our guide to phone access for vulnerable patients.
Where AI phone answering is heading
Products in this space vary widely. Ava, for example, answers practice information, explains routes, signposts 999 and NHS 111 with practice-agreed wording and passes patient matters to reception without taking health details, while appointment requests into clinical systems remain planned pending safety assurance. That split between what is live and what needs more assurance is a useful test for any product you consider.
Expect the boundary to move as integration, regulation and NHS App features develop. Each step should be backed by evidence from your own monitoring. If you are in Wales, check how any system handles the bilingual greeting rules, and for dental settings see AI receptionists in dental practices.
Measuring the pilot
- Share of calls the system completed without a hand-off, by call type.
- Hand-offs actioned by staff within the stated time.
- Urgent signposting events, each reviewed by a clinician.
- Callers who asked for a person, and how long it took to reach one.
- Complaints, compliments and staff-reported issues.
- Effect on queue length and abandonment at peak times.
Set a review point, such as six weeks, and decide in advance what results would lead you to stop, continue or widen the scope.
Frequently asked questions
Can an AI system triage GP patients over the phone?
Not without formal clinical safety assurance and, often, regulation as a medical device. Most practices limit AI call handling to information, routing and urgent signposting.
Do patients have to be told they are speaking to AI?
Yes, they should be. Say so in the greeting, explain how to reach a person and update the privacy notice.
What paperwork is needed before going live?
At least a DPIA, a clinical safety assessment under DCB0160 with the supplier's DCB0129 documentation, and a data processing agreement.
References
Nuffield Trust (2025). How are GPs using AI? Insights from the front line (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) ↑
GOV.UK (Medicines and Healthcare products Regulatory Agency) (2025). Software and artificial intelligence (AI) as a medical device (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 (Department for Science, Innovation and Technology) (2024). Public attitudes to data and AI: Tracker survey (Wave 4) report (opens in a new tab) ↑