Does an AI receptionist meet CQC expectations in a GP practice?

The honest answer is that the CQC has not addressed this specific thing. Its guidance on artificial intelligence in GP services exists, it is detailed, and answering the practice telephone is absent from it. What the guidance does give a practice is a clear framework of systems and processes that any AI tool has to sit inside, and that framework is what an assessment will test. A practice can meet it. It cannot cite a CQC blessing for an AI receptionist, because none has been written.

GP mythbuster 109 on the use of artificial intelligence in GP services, published 14 July 2025, is the document, and this post stays inside what it actually says.

Not on answering the telephone. GP mythbuster 109, published 14 July 2025, is the CQC guidance on artificial intelligence in GP services, and the uses it describes are ambient voice scribes, triage through a structured online questionnaire, results processing, clinical documentation, diagnosis and treatment planning support, chatbots, and predictive modelling for missed appointments and staffing. A conversational system answering the practice phone appears nowhere in it. The expectations still reach such a system, because they are written about a practice's systems and processes and those cover any AI tool it adopts. This is general information about published CQC guidance and not legal or clinical advice.

Two details in the guidance matter more than they first appear. The CQC defines AI as technologies that simulate human intelligence to perform complex tasks by learning from data, and it separates that from rules-based automation, which follows fixed logic and produces the same outcome for a given input every time. A scripted telephone menu is the second thing. A system that interprets what a caller says is the first, and the guidance follows it.

The other detail is that the CQC already lists administrative AI in general practice without treating it as exotic. Under predictive modelling it names predicting missed appointments and helping to schedule them, and forecasting staffing levels. Administrative AI in a GP practice is, on the regulator's own account, ordinary.

The practice's systems and processes. GP mythbuster 109 names ten areas an assessment covers, running from how the tool was procured through to whether it works fairly across the practice population, and taking in risk assessment, named responsibilities, human oversight, learning from errors, data protection, consent and staff training on the way. The CQC states it will check that AI is being used as a support tool, and that it will look for the evidence in a practice's audits, significant incident log or other quality improvement activity. The practice that cannot produce that evidence is the one with the problem.

The ten areas are set out below in the order the guidance uses. This is the list to walk into an assessment with, and the practice that has an answer for each one is in a strong position whatever the tool happens to be.

That turns on whether it influences clinical decision making. The CQC states that an AI tool used in a practice which influences clinical decision-making is likely to qualify as a medical device under MHRA regulations, with classification driven by intended use and running from Class I at the lowest risk to Class III at the highest. A system that books appointments and takes messages sits outside that. A system that assesses symptoms or assigns urgency is doing the thing that pulls it in, and a practice should then ask the supplier for MHRA registration, UKCA or CE marking, and the classification documents.

The guidance is specific about the evidence to request. A practice should ask an AI supplier for MHRA registration, UKCA or CE marking, and device classification with the clinical safety and regulatory documentation, and the CQC notes that by law a manufacturer must know and disclose this where the product is regulated. Registered devices can be checked on the MHRA public access registration database. The classification depends on intended use, which is why the boundary a system is built to matters so much: booking an appointment and taking a message keeps a tool clear of clinical decision-making, and assessing symptoms does not.

The guidance adds a caution worth carrying: software should only be used within the constraints described in its instructions for use, and adopters who go beyond them take on significantly more risk. A tool sold for scheduling and quietly used for triage is exactly that scenario.

Where DCB0160 applies, yes, and the CQC states the scope. Its wording is that when adopting an eligible digital technology on behalf of the NHS, providers need to first meet the safety standards set by NHS Digital, so a tool that is neither eligible nor adopted on behalf of the NHS is not caught by the standard on the face of the guidance. Most general practice delivers NHS-commissioned services, eligibility is a question for the ICB, and the ten assessment areas apply either way. Where DCB0160 does apply, the clinical safety officer must be a senior clinician, hold current registration with aprofessional body such as the General Medical Council or the Nursing and Midwifery Council, and have training in digital clinical safety and clinical risk management to practitioner level. The CQC notes that a practice without that expertise in house may seek it from the commissioning organisation, the primary care network or a third party provider.

The clinical safety officer owns the process defined in the standard, which the guidance describes as evaluating the evidence that clinical risks are mitigated or accepted, keeping the risk management documentation in order, maintaining the clinical safety case report, hazard log and clinical risk management plan, and organising hazard workshops. The CQC also distinguishes the two NHS standards cleanly: DCB0129 applies to the developer of the technology and DCB0160 to the adopter, so a supplier's compliance never discharges the practice's own.

The eligibility wording is worth reading before concluding it lets a practice off. Most general practice delivers NHS-commissioned services, the question of what counts as eligible is one for the ICB rather than the supplier, and the ten assessment areas set out above apply to the practice's use of AI whatever the answer on DCB0160 turns out to be. Procurement, governance and human oversight are assessed either way.

Telling them is the requirement, and explicit consent is a separate question. The CQC states that the type of consent needed depends on the technology and its intended use, and that because AI technologies are new a practice does need to tell people it is using them. Its wording is that this is about being transparent and allowing people to object, rather than asking for explicit consent. The same guidance calls it essential that practices offer a non-digital route to access care, so a patient who declines the automated route still has to be able to reach the practice.

The worked example in the guidance concerns AI scribes, where the CQC says a practice does not need explicit consent to use one for tasks delivering individual care, and may rely on implied consent under the common law duty of confidentiality, while still needing to tell patients they are in use. Transparency carries the obligation. On human oversight the guidance is direct in requiring a practice to demonstrate that AI is being used as a support tool, with the CQC adding that this is never a replacement for human oversight.

A practice can say its AI tool was procured against DCB0160 and DTAC, that a named clinical safety officer owns the hazard log, that oversight is evidenced in audit, that patients are told, and that a non-digital route remains open. That set of claims is checkable and it maps onto what the CQC says it will look for. A claim that the CQC approves of AI receptionists is not available to anyone, and a supplier offering it is describing a document that does not exist.

Where that leaves the telephone is a matter of scope. A system that answers, identifies itself, takes a name and a number, books into the appointment book against rules the practice set, and moves anything clinical to a person, is doing administrative work the guidance already recognises in other forms. The rules on regulated AI intake draws the same boundary across four markets, and how an AI employee is scoped and governed on our side is on how we work.