AI receptionist for UK clinics: GDPR, patient data and what to ask a vendor
By Yevhen Lukyanov, founder of Fantastic AI Studio · Updated 10 November 2026
TL;DR — An AI receptionist can answer booking, opening-hours and price questions for a UK clinic, but anything touching a patient's health is special category data. Before go-live, settle your lawful basis, complete a DPIA, tell patients it is an AI, and keep clinical judgement with staff.
A receptionist assistant looks like an administrative tool, but in a clinic the conversation often turns clinical in one message: "I've had a rash for a week, can I see someone?" From that point the assistant is handling information about a person's health. That does not rule an assistant out. It changes what you must settle first, and it is why a clinic needs a tighter brief than a hotel or an estate agent. This is general information, not legal advice; the regulators named below are the authority.
Why patient conversations are special category data
UK GDPR Article 9 lists data concerning health as a special category and prohibits processing it unless a specific condition applies. Two conditions matter most for clinics: processing necessary for the provision of health care or treatment (Article 9(2)(h), which is subject to the professional-secrecy safeguards in Article 9(3)), and explicit consent (Article 9(2)(a)). Which one fits a given use is a decision for your clinic and your data protection adviser, and it should be written down before launch rather than after.
The practical point is scope. An assistant that only gives opening hours, prices and directions should not need to collect health details at all. If it does collect a reason for a visit, a symptom or a medication, you are processing special category data through a third-party system, and the questions further down apply.
Do you need a DPIA?
Article 35 of the UK GDPR requires a data protection impact assessment where processing is likely to be high risk, and names large-scale processing of special category data as one case. The NHS England Digital Regulations Service says a DPIA is required by law before large-scale processing of special category data using innovative technology, strongly advises one before deploying technology in a health or social care setting, and says to update it when the processing changes. Neither source defines "large scale", and a small clinic may be below it. Even so, CQC lists a DPIA among the principles it expects providers to follow when they use AI. A short, honest DPIA is cheaper than explaining its absence later.
What CQC expects when a registered provider uses AI
If your clinic provides a regulated activity in England, it is registered with CQC. CQC's statement on AI, last updated on 21 May 2026, says it does not approve specific technologies and does not create a separate AI rulebook. Instead it applies existing regulations, including consent (Regulation 11), safe care and treatment (Regulation 12) and good governance (Regulation 17). It sets out principles for providers: AI should support, not replace, human decision-making; its outputs should be continuously monitored; people should be told about the role of AI in their care and offered non-digital routes where needed; staff should be trained; and there should be clear accountability and a way to report and investigate problems. CQC also says the presence or absence of AI does not predict a specific rating. CQC regulates England only; Scotland, Wales and Northern Ireland have different regulators.
For a receptionist assistant, those principles translate directly into design choices: a visible "you are talking to an AI" message, a phone number and a human hand-off in every conversation, a named person who reads transcripts, and a record of what the assistant is and is not allowed to do.
Where an assistant stops and a person starts
| Task | What the assistant does | What stays with staff |
|---|---|---|
| Opening hours, location, parking, prices | Answers from the details you have approved | Any price or exception not in those details |
| Booking and rescheduling | Offers the slots you release, captures the request, confirms by the channel you choose | Matching the appointment type to the clinician, and any clinical screening questions |
| Reminders and confirmations | Sends approved wording at agreed times | Handling replies that mention symptoms or distress |
| Insurer and self-pay questions | Gives the policy you have published | Pre-authorisation and case-by-case decisions |
| Symptom or medication questions | Does not answer. Says a clinician or the practice team will respond, and passes it on | All of it |
| Urgent or distressed messages | Shows your emergency wording (999 or NHS 111) and alerts the on-duty person | Every judgement about urgency |
| Complaints and sensitive records requests | Acknowledges and routes with the conversation attached | All of it |
One boundary deserves its own sentence. If a tool is meant to assess symptoms or support diagnosis, it may count as a medical device regulated by the MHRA, which publishes guidance on when software applications are considered medical devices, with separate material on symptom checkers. A receptionist assistant should be built and described so it plainly does not do that. Ask any vendor to state the tool's intended purpose in writing.
Questions to ask a vendor before patients meet it
- What patient data does it collect, where is it stored, and for how long? Can you set the retention period and delete a conversation on request?
- Will the vendor sign a data processing agreement, and which sub-processors (for example the language-model provider) see the conversation? Is any of it used to train their models?
- Is the assistant restricted to administrative tasks, and can it be configured never to ask about symptoms?
- What does it say when someone describes an urgent problem, and who is alerted?
- Can you read every transcript, correct a wrong answer within a day and switch the assistant off immediately?
- Which calendar or practice-management system does it connect to? Ask for the system's name and confirm yours before you sign.
- Who supplies the DPIA inputs: a description of the data flows, the sub-processor list and the security measures?
What it costs
UK AI-only receptionist software typically runs £149–£399 a month, according to Hand On Web's 2026 comparison, but generic software is not built around your appointment types, your clinicians or your governance. A supervised assistant of the kind we build costs £3,500–7,500 to launch plus £249–749 a month, after a £1,500 audit that is credited to the build if you go ahead. Our cost guide compares the four main routes.
Compare AI-only software, hybrid services, a part-time receptionist and a supervised assistant.
AI receptionist cost in the UK, 2026Where we fit, and what we can't claim yet
Fantastic AI Studio is a new studio. We do not yet have a published UK clinic case study, and we are not a data protection adviser or a clinical safety assessor: your clinic remains responsible for its DPIA, its lawful basis and its clinical governance. What we offer is a scoped, supervised assistant limited to administrative work, with transcripts you can read and a human hand-off in every conversation. Every engagement starts with the audit, and if a different route suits your clinic better, the audit will say so.
See how a supervised assistant handles bookings, reminders and enquiries for a UK private clinic.
AI receptionist for clinicsSources
- legislation.gov.uk — UK GDPR Article 9: processing of special categories of personal data
- legislation.gov.uk — UK GDPR Article 35: data protection impact assessment
- NHS England Digital Regulations Service — Step 2: do a data protection impact assessment (read 10 November 2026)
- CQC — Artificial intelligence in health and social care: CQC's role, expectations and plans (updated 21 May 2026)
- MHRA — Medical devices: software applications (apps) (updated 1 July 2023)
- Hand On Web — AI receptionist cost UK: 2026 comparison
- ICO — Guidance on AI and data protection