Yes, AI recording could make some patients less willing to disclose sensitive information. However, reported discomfort is not the same as a measured reduction in disclosure or a demonstrated adverse clinical outcome. The useful question is how a service preserves a genuinely private conversation when recording becomes routine.
The evidence concerns comfort, not a measured rate of withheld information
Healthwatch's research published on 16 July 2026 included representative polling of 4,039 adults. Reported comfort with AI scribes was 48% for routine health checks, compared with 23% for domestic abuse, 28% for mental health and 29% for sexual health. Its separately collected personal stories were self-selected, not a population estimate of errors or harm. See the Healthwatch findings.
These findings support taking the question seriously. They do not tell us how many people actually omitted information during a recorded appointment, whether any omission changed treatment, or whether the same patients would have disclosed more without a scribe. Those would require different research methods.
A sensible response is therefore neither to declare recording harmless nor to assume that every recorded consultation damages trust. It is to design the encounter so that people can change their minds, raise concerns and continue their care without having to justify their discomfort.
The consultation can change after recording begins
Consider a fictional patient who books a routine review but is also considering discussing controlling behaviour at home. The clinician explains that a digital tool will help prepare the note. The patient nods, intending to discuss the original problem only. Halfway through, the conversation opens a possible opportunity to disclose the second concern.
At this point, an apparently satisfactory introduction is not necessarily enough. The patient may wonder whether the recording will be retained, who can read the resulting summary, and whether an accompanying person will learn about the disclosure. The clinician cannot infer the answers to those concerns from the earlier nod.
This constructed example is not a reported AI incident. It illustrates why permission, understanding and comfort should not be treated as a single permanent state established at the start of an appointment.
A proposed conversational response is simple: "We can pause the recording and carry on. You do not need to explain why you would prefer that." The words only have value when the clinician can actually stop the relevant capture and processing.
Distinguish stopping the tool from stopping the clinical record
Pausing an AI scribe does not ordinarily mean that the clinician stops making an appropriate medical record. Patients should not be left believing that discussing something without the tool means the information will necessarily remain undocumented or unavailable to everyone involved in their care.
The explanation should distinguish the conversation, any audio, the generated draft and the approved clinical record. These are different objects with potentially different access and retention arrangements. Clinicians should describe their own organisation's arrangements rather than offer a generic assurance about what all AI systems do.
For England's individual-care scribing context, NHS England's information-governance guidance, checked on 10 October 2026, requires transparency and respect for dissent. Its statement that explicit consent is not required for the processing described does not remove the need for an understandable conversation.
An appropriate explanation might be: "I will still document the information needed for your care. We can discuss any concerns about what is recorded and who may see it." This is a proposed script, not a promise that every disclosure can be kept outside the record.
Make pausing practical, including when someone else is present
A pause button hidden in a settings menu is different from a pause that a clinician can confidently use during a difficult conversation. Before deployment, a team should test what happens to audio already buffered, whether transcription continues elsewhere, and how resumption is shown. These are procurement questions, not assumptions about a named supplier.
The encounter also needs a way to speak privately when an accompanying person is involved. A prominent invitation to explain why recording should stop may itself be uncomfortable. A neutral offer of private discussion can avoid requiring the patient to reveal the topic before the private conversation exists.
The GMC's communication standards, checked on 10 October 2026, emphasise listening, understanding patients' concerns and adapting communication to their needs. The implementation question is how the technology supports those obligations rather than competing with them.
A service could rehearse these transitions with fictional consultations: a patient who changes their preference, someone who misunderstands what recording means, and a person who wants to discuss a concern privately. Successful rehearsal would require a workable alternative, not merely the correct explanatory phrase.
Look for silent barriers, not just recorded objections
An opt-out count cannot establish that everyone else was comfortable. Someone may not understand the option, may feel awkward interrupting, or may decide not to introduce the sensitive subject. These are plausible explanations to investigate, not conclusions that can be read directly from a low refusal rate.
A proposed evaluation should ask patients privately whether they understood the tool, knew how to stop it and felt able to discuss everything they wanted. It should offer accessible ways to respond and avoid requiring people to describe the sensitive information they withheld.
Researchers should then distinguish awareness, comfort, disclosure and clinical consequences. An improvement in the first measure should not automatically be reported as improvement in all four. Any investigation involving sensitive disclosures also needs appropriate ethical and safeguarding arrangements.
Preserve the relationship, not the recording target
There may be consultations in which an approved scribe supports better attention and communication. There may also be moments when turning it off is the more useful clinical choice. The objective should be an adequate, honest consultation with a reliable record, not uninterrupted capture.
An organisation could therefore treat a well-handled pause as successful use of its policy rather than failed adoption. Staff should know that continuing without the tool is an available professional response, not an exception that requires defending afterwards.
For education, a team can use fictional role-play to practise explaining recording, recognising uncertainty about it and continuing a sensitive discussion. No real patient conversation needs to be uploaded to a learning platform. The skill being rehearsed is maintaining room for the patient's account, including information that a recording system cannot capture because it was never spoken.
Frequently asked questions
Does research prove AI scribes cause patients to withhold information?
No. The Healthwatch research discussed here measures attitudes and includes personal accounts; it does not establish a population-wide causal effect on disclosure or clinical outcomes.
Can a patient ask for an AI scribe to be stopped partway through?
In the England individual-care context described by NHS England, dissent must be respected. Teams should have a practical way to stop the tool and continue the consultation.
Does switching off the scribe mean nothing will be documented?
No. An appropriate clinical record may still be necessary, and the clinician should explain the distinction between AI recording and ordinary documentation.
Develop a learning activity around communication and patient trust →