Key takeaways
- A basic AI receptionist answers and takes a message; an AI employee is built to finish a defined task within rules you approve.
- The real test is what happens after the greeting: the wording used, where it stops, what it records and how a caller reaches a person.
- Neither should triage or give clinical advice, and any write-back to a clinical record needs a DPIA and clinical safety work first.
- Patients are moving online fast, so a tool that only covers the phone line solves part of the problem.
- Judge suppliers on five live test calls from your own practice, not on a scripted demo.
Two labels for very different products
Suppliers use "AI receptionist", "virtual receptionist", "voice agent" and "AI employee" almost interchangeably, so the label tells a practice manager very little. A more useful distinction is about outcomes. A basic AI receptionist picks up the call, says something sensible and usually ends with a message for the team. An AI employee is designed to complete a defined piece of work: give the correct answer in your wording, route the caller to the right service, record what happened and hand over cleanly when a person is needed.
Both can sound polished on a demo. The difference only shows when you follow a real call through to the end and ask who still has work to do afterwards.
Side by side
| Question | Basic AI receptionist | AI employee |
|---|---|---|
| Where do answers come from? | A generic script or general knowledge, which may guess | Wording your practice approved, with a clear 'I don't know' when something is missing |
| What happens to a routine request? | A message is taken for someone to deal with later | The request is completed within agreed rules or routed to the named team |
| Urgent symptoms? | May treat them as any other message | Agreed 999 and NHS 111 wording, no follow-up questions |
| Reaching a person? | Can loop callers round the menu | A route to reception agreed in the call plan |
| Records? | A short email or note per call | Call record and transcript under your retention rules |
| Improvement? | Same answers until someone rewrites the prompt | Early calls reviewed and approved wording added |
What finishing a task really involves
In a practice, very few calls are only questions. Most are requests that need an action and a record. When you assess a product, walk through these stages for a typical call and note where the tool stops and your team has to pick up.
Understand the request
Recognise what the caller wants in their own words, including accents, background noise and callers who change their mind halfway.
Check the rules
Look up the practice's approved answer or process rather than improvising one.
Act or route
Give the answer, explain the right route, or pass the caller to the correct team with the reason attached.
Confirm with the caller
Tell the caller what will happen next and when, so they do not ring back to check.
Record
Leave a record the team can find, with no more personal data than the task needs.
Hand over
Make the person picking it up able to act without calling the patient back to start again.
A basic receptionist often handles the first stage well and then jumps straight to taking a message. An AI employee should cover every stage for the tasks it is set up for, and say clearly which tasks are outside its scope.
Five calls that show the difference
- "Are you open on the bank holiday?" Both should answer, but only one built on your approved wording will give this year's hours rather than a guess.
- "How do I get an appointment?" The receptionist may start asking about symptoms. The employee explains your routes, such as the online consultation form, the NHS App and staffed phone times, without triaging.
- "I'm ringing about my blood test." Neither should discuss results. The employee passes the caller to reception through the agreed route instead of collecting health details in a voicemail.
- "My chest is tight and my arm hurts." The only acceptable response is your agreed 999 wording, straight away.
- "Can I just speak to someone?" The employee offers the agreed transfer or the direct line and its hours. Looping the caller back to the menu is a failure.
Run exactly these calls, in your own voice and in a colleague's, during every demo. They expose more than any slide deck.
Why the distinction matters now
Phone access is improving but is still a pressure point. In the 2026 GP Patient Survey, 56.9% of patients said it was easy to contact their practice by phone, up from 49.7% in 2024, and 30.8% said their last contact attempt was online, nearly double the 2024 share[1]. Patients increasingly move between phone, website and app for the same request, so a tool that answers calls but cannot follow your wider process only shifts the work around.
Source: Ipsos (GP Patient Survey provider for NHS England), 2026[1]
Show the data
| Period | Value |
|---|---|
| 2024 | 16.8% |
| 2025 | 22.4% |
| 2026 | 30.8% |
NHS England has set the goal of ending engaged tones and promoted queue messages and call-back options as part of digital telephony[2]. The government's 10 Year Health Plan estimates that saving just 90 seconds per GP appointment would free more than 2,000 full-time GPs' worth of capacity[3]. Admin time works the same way: the value of an AI tool lies in the follow-up work it removes, not in the fact that it answered.
Limits that apply to both
Neither kind of tool should triage, diagnose or give clinical advice. The MHRA says many software and AI products used in health and care are regulated as medical devices[4], and a phone tool that starts weighing symptoms risks drifting into that territory without the evidence to support it. Writing into a clinical system such as EMIS Web or SystmOne is also a different level of commitment: it needs a DPIA, clinical safety assurance and a data processing agreement before it is switched on.
Clinicians' concerns are real. Among GPs not using AI, 89% were worried about professional liability and 82% about patient privacy and data security[5]. A good supplier will expect those questions and answer them in writing.
As an example of how this looks in practice, Ava is built to be an AI Employee for practice phones: its live work is answering practice information, explaining routes, giving agreed urgent-help wording and passing patient matters to reception, while appointment and prescription requests are labelled Planned.
Example: a private clinic compares two demos
Example (fictional): a physiotherapy and sports medicine clinic in Leeds tests two products with the same six calls, recorded by the practice manager and a receptionist with a strong local accent.
Product A answered every call quickly and sounded natural, but four of the six calls ended with 'I've passed your message to the team', including the opening-hours question it could have answered. Product B was slightly slower, read the clinic's approved cancellation policy word for word, transferred the 'speak to someone' call within the agreed time and left a short summary for each call. In this example the clinic chose B, because A would have created as many callbacks as it removed.
What changes for the reception team
The choice between the two kinds of tool changes the receptionists' day in different ways. With a basic AI receptionist, the team usually inherits a queue of messages: each one has to be read, understood and often returned by phone, because the message rarely holds enough detail to act on. The calls have moved from the morning rush to mid-morning, but the work has not gone away.
With a tool built to finish defined tasks, the routine questions never reach the queue at all, and the requests that do arrive are already sorted by team with the reason attached. That frees experienced staff for the conversations that need judgement: the anxious parent, the patient who cannot use the online form, the caller who is upset about a referral.
- Agree who reviews AI call records each day and how quickly.
- Decide which requests go to which inbox or task list, and name an owner for each.
- Give receptionists a simple way to flag a wrong or outdated answer, and fix it within a week.
- Share early call summaries in team meetings so staff can see what patients are actually asking.
Involving the team from the start is not a courtesy. They know which answers change every season, which callers need a person straight away and which local services patients confuse, and that knowledge is exactly what makes the approved wording accurate.
Decision criteria for your shortlist
- Scope in writing: which tasks the tool completes, which it routes and which it refuses.
- Source of answers: your approved wording, with a visible process for updating it.
- Escalation: a tested route to a person and agreed urgent-help wording.
- Records: what is stored, for how long, and who in the practice can see it.
- Data protection: DPIA support, a processor contract and a clear statement on whether call data trains the product.
- Honest status: features labelled live or planned, with no promise that depends on an integration you do not have.
Patients will use these tools whatever you call them; 40% of UK adults already use AI chatbots for personal purposes at least monthly[6]. The question for a practice is not whether the voice sounds human, but whether the call is finished properly when it ends.
Frequently asked questions
Is an AI employee just a more advanced AI receptionist?
The difference is scope, not polish. A receptionist answers and takes a message; an employee completes defined tasks within your rules and hands over cleanly, which reduces follow-up work for the team.
Can either type triage patients?
Neither should. Phone tools should explain your routes and give agreed urgent-help wording, leaving clinical decisions to clinicians and NHS services.
Does an AI employee need access to our clinical system?
Not for answering practice information or routing calls. Booking or writing into EMIS Web or SystmOne needs a DPIA, clinical safety assurance and a data processing agreement first.
How should we test the difference?
Use five or six real call types from your own practice, including an urgent symptom and a request to speak to a person, and follow each call to the end.
References
Ipsos (GP Patient Survey provider for NHS England) (2026). 2026 GP Patient Survey results released (opens in a new tab) ↑
NHS England and DHSC (2023). Delivery plan for recovering access to primary care (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) ↑