Patients should not have to accept an AI scribe simply to obtain an adequate consultation. Whether refusal genuinely preserves their care depends on the alternative workflow, not just an opt-out sentence. This article examines England's scribing guidance and proposes a service-design standard; it does not assert a universal right to reject every healthcare technology.
Start with the specific tool and the applicable guidance
An AI scribe, an imaging algorithm and a patient-facing triage form do different jobs. A policy about dissent from consultation recording should not silently become a legal conclusion about all three.
As checked on 10 October 2026, NHS England's ambient-scribing information-governance guidance says that a patient's dissent means the tool cannot be used. It also expects services to prepare for creating summaries without it. That is a practical requirement, not merely a notice-writing exercise.
The question for a practice is therefore concrete: what happens in the next minute when someone declines? The answer should not depend on which receptionist receives the request, whether the usual clinician is present, or whether the patient knows the organisation's governance vocabulary.
A formal choice can still be difficult to exercise
Imagine a fictional service offering a convenient appointment through an AI-supported route. A patient who does not want recording is told to telephone another number, repeat the booking process and wait for an unspecified alternative. The policy may describe a choice, but the practical cost of exercising it is considerable.
Contrast a second fictional service. The patient can express their preference during booking or at the appointment. The clinician uses the established documentation method, and the preference is handled without a public discussion at reception. The encounter may look different, but the patient still receives appropriate assessment and follow-up.
Neither example is an observed comparison. They expose a distinction that organisations can test: whether refusal changes the technology used or obstructs access to the underlying service.
In its 11 September 2026 response on safe AI use, Healthwatch emphasised transparency, oversight and individual agency, including the ability to decline scribing for sensitive discussions. That is patient-advocacy evidence about what should matter, not proof that all current services meet it.
Define equivalence around care, not identical mechanics
An editorial standard of "equivalent care without this tool" should not promise identical appointment length, identical documentation processes or immediate availability in every setting. It should ask whether the person can still receive a clinically appropriate assessment, understandable information, necessary follow-up and a route to raise concerns.
This distinction matters because a different workflow can be reasonable. A clinician may type more when recording is declined. A service might arrange additional support for communication. The important issue is whether differences respond to clinical and accessibility needs or merely make the non-AI option unattractive.
A useful local specification would identify the alternative at each stage: booking, assessment, documentation, communication and follow-up. It would also identify the person authorised to resolve a problem when the usual alternative is unavailable.
The specification should be tested with people who are unfamiliar with the system. A route that works only for a digitally confident staff member is not a convincing patient alternative.
Accessibility is part of the route, not an optional extra
Some people will need information in a different format or additional communication support before they can make a meaningful choice. A web page explaining the opt-out is not sufficient for someone unable to access or understand that page.
The Accessible Information Standard, checked on 10 October 2026, addresses identifying, recording, flagging, sharing, meeting and reviewing relevant information and communication needs. Its scope concerns needs associated with disability, impairment or sensory loss; it should not be described as a single policy covering every possible language or access issue.
For AI deployment, a proposed practical test is whether the explanation and alternative remain usable with an interpreter, assistive communication or a carer's appropriate support. The patient should not have to master the AI interface to avoid using it.
There is also a distinction between a person choosing not to use a tool and a person being unable to use it. Recording both as "patient refused digital" can obscure different barriers and lead to an unsuitable response next time.
Avoid making dissent a judgement about the patient
A refusal does not by itself demonstrate poor understanding, hostility to innovation or unwillingness to engage with care. A patient may understand the proposed benefits and still prefer a different arrangement.
The GMC's patient-partnership standards, checked on 10 October 2026, require respectful communication and attention to individual needs. A proposed operational extension is to record the relevant preference neutrally, without attributing a motive that has not been established.
Staff should also avoid repeated persuasion after a clear decision. Explaining a misunderstanding can be helpful; making the patient repeatedly defend their choice is different. Training scenarios can ask staff to distinguish those situations without treating either automatic reassurance or immediate argument as the default response.
Measure what happens after refusal
A credible audit would follow the pathway rather than count the number of people who declined. Relevant questions include whether the appointment proceeded, whether an alternative was explained, whether the patient had to repeat their request and whether any follow-up was lost during the change of workflow.
Compare results cautiously. Patients choosing different routes may have different needs, and a longer encounter is not necessarily a worse encounter. The purpose is to identify avoidable barriers and inappropriate differences, not to manufacture a crude AI-versus-no-AI league table.
Staff experience belongs in the assessment too. A practice cannot sustainably offer an alternative that exists only because one clinician does substantial unrecognised work. The organisation should provide the time, training and documentation process necessary for the choice it advertises.
A better introduction makes the alternative ordinary
A proposed introduction could say: "This tool helps prepare a draft note. We can continue without it, and you can ask for it to be stopped during the appointment." The wording should be used only where the service can deliver what it says.
The strongest implementation is one in which neither accepting nor declining becomes the main event. Patients receive a clear explanation, the clinician adapts the workflow and the consultation returns to the problem that brought the person there.
That is a more useful success criterion than maximising the proportion of recorded appointments. A technology intended to support care should not make the patient's ability to question its use depend on their confidence, persistence or digital skill.
Frequently asked questions
Does refusing an AI scribe mean refusing medical care?
No. The documentation tool and the clinical assessment are different, and England's scribing guidance expects services to prepare an alternative way to produce the record.
Must the non-AI appointment be identical?
Not necessarily. The proposed standard is clinically appropriate, accessible care without avoidable disadvantage, rather than identical appointment mechanics.
Does this establish a right to refuse every healthcare AI system?
No. Different technologies, purposes and jurisdictions require their own assessment, so the scribing guidance should not be extended indiscriminately.
Turn a patient-access concern into a focused learning reflection →