A patient asking whether a consultation is being recorded needs an answer about the consultation, not an infrastructure slogan. European-owned processing may be relevant to the service's governance, but it does not explain what the microphone captures, what the software produces or what happens when the patient would prefer not to use it.
A change to sovereign infrastructure can therefore change part of the explanation without replacing its core. Patients still need to understand the task, the clinician's role and their practical choices.
This article focuses on ambient scribing for individual care in England, with sources reviewed on 19 October 2026. The suggested wording is original and should be used only where it accurately describes the service operating in that consultation.
Begin with the questions a patient is actually asking
A patient may want to know whether the software is listening continuously, whether another person can hear the recording, whether the information helps train a model or whether the computer is making the clinical decision.
These questions are related but not interchangeable. Answering that the servers are in Europe does not establish the retention policy. Saying that the clinician reviews the note does not explain who processes the audio.
A useful opening explanation has a small number of concrete facts. Further detail should be available when the patient asks for it, without requiring them to understand cloud architecture before deciding whether they are comfortable.
The general sovereignty guide can help staff understand the technical vocabulary before translating it into ordinary language.
Transparency and explicit consent are not the same issue
NHS England's guidance for health and care professionals says explicit consent is not required for the individual-care activity it describes. It nevertheless requires informing patients at the start and respecting dissent.
That is a scoped position, not permission for every research, product-improvement or recording purpose. It also does not turn a conversational question about comfort into the organisation's UK GDPR lawful basis.
For the clinician, the operational requirement is straightforward: know the approved arrangement, explain it honestly and have a workable alternative. A policy that allows objection but leaves staff unable to continue without the tool is incomplete in practice.
An opening explanation for a documentation-only encounter
For a fictional service where an approved tool processes the conversation to prepare a draft note, a clinician could say:
"I use a note-taking tool that processes our conversation and prepares a draft. I check and correct the note before adding it to your record. Tell me if you would rather I took the notes without it."
This wording describes the immediate task and the opportunity to object. It does not invent a storage location, deletion period or guarantee that no supplier can process the information.
If other functions are being used, explain those separately. A script describing only note preparation should not conceal a patient-specific recommendation or an automated communication function operating in the same interface.
The purpose is an accurate explanation of today's encounter, not a memorised statement about every feature sold under the brand.
What a verified infrastructure statement can add
Suppose a clinic has confirmed that its relevant processing is operated through European-owned infrastructure. It could explain:
"Our organisation has chosen a service arrangement using European-owned infrastructure for this processing. The privacy information explains the companies involved and how the information is handled. That is separate from my responsibility to check the clinical note."
This statement is appropriate only when the arrangement has actually been confirmed. An announcement that a supplier offers a new option does not establish that the clinic has purchased or enabled it.
For Tandem and Accurx users, that distinction is particularly important. A partner-delivered implementation needs its own confirmation rather than inheriting a new description from another deployment.
Avoid promising a deletion process that is not yours
Different services can handle raw audio differently. Tandem's published security page states that audio is processed in real time and deleted immediately after transcription. That is a supplier statement about audio, not a universal description of every transcript, generated note or supporting record.
A clinician should not substitute a generic script saying that recording is kept until the note is approved when that is not how the selected service operates. Nor should immediate audio deletion be expanded into "nothing is stored".
Before a rollout, prepare an internal answer covering each relevant information type: audio, transcript, generated draft, approved record and operational information. The patient-facing explanation should remain consistent with that account.
When the answer is unknown, acknowledge it and obtain the correct information. An invented reassurance is harder to repair than an honest boundary.
Respond to an objection without turning it into a debate
A patient may decline because of confidentiality, unfamiliarity, a previous experience or a reason they do not wish to discuss.
For a service with a manual alternative, suitable wording could be:
"That is fine. I will continue without the AI note-taking tool and make the clinical record myself. We can carry on with what brought you here."
The alternative still involves ordinary documentation. Stopping the scribe does not necessarily mean no clinical record will be created, and the patient should not be left with that impression.
Where an objection comes after capture has started, separate stopping further processing from handling information already processed. Staff should follow the relevant local procedure rather than promise that every copy has instantly disappeared.
A patient should not need to prove that the technology is unsafe to express a preference about its use.
Keep sensitive conversations and other people visible
A consultation may involve an interpreter, a relative or information about someone who is absent. A concise automated note can still misattribute whose condition, preference or concern was discussed.
The facilitator of the consultation should make the use of the tool clear to those participating and consider whether the setting is appropriate. When third-party information appears in the draft, review whether it is necessary and accurately attributed.
For example, an invented consultation may include a carer explaining their own difficulty providing support. The relevant record might need to capture the resulting care issue without reproducing every private detail about the carer.
This is a documentation judgement, not a reason to remove clinically important safeguarding information automatically. Difficult cases need the organisation's established clinical and information-governance processes.
Align the spoken explanation with the written notice
The privacy notice can explain the organisations involved, purposes, information categories, retention and applicable rights in more detail than the opening conversation.
Its role is not to correct an inaccurate verbal explanation afterwards. If staff say that all information remains in one country while the notice describes another processing route, patients are receiving contradictory accounts.
A useful implementation review compares the consulting-room script, website notice, reception material and supplier configuration. Treat a material processing change as a reason to revisit that set, not only the contract.
The ICO's transparency guidance for AI provides the broader data-protection context. The organisation should determine the actual notices and legal arrangements with appropriate expertise.
Rehearse the awkward questions before the first clinic
A short fictional role-play can test whether staff can explain the service without overstating it.
Ask one person to play a patient who says, "Does European mean nobody else can access it?" Ask another to object halfway through. In a third version, an interpreter joins after the opening explanation.
The useful endpoint is not a perfectly polished sales pitch. It is whether staff can distinguish known facts from unresolved questions, stop the tool appropriately and continue the clinical encounter.
Record what the exercise reveals: perhaps nobody knows the transcript-retention period, or the written notice describes a different product tier. Resolve those points before asking clinicians to reassure patients confidently.
Trust depends on what happens after the explanation
A good introduction cannot compensate for a note that nobody checks or an objection that staff disregard. Conversely, a clear response to a patient's concern can make an unfamiliar technology easier to understand without promising that every patient will welcome it.
Sovereign infrastructure may be one relevant benefit. The patient should still be able to understand what was used, challenge an inaccurate record and receive care when the tool is not used.
That is a stronger professional account than "it is European, so there is nothing to worry about".
Frequently asked questions
Does a sovereign deployment remove the need to tell patients about an AI scribe?
No. Infrastructure arrangements do not replace transparency about the actual use of the tool in the consultation.
Can clinicians use one standard statement about audio deletion for every product?
No. The statement must match the selected service and information type. Immediate deletion of audio does not establish identical deletion of transcripts, drafts or approved records.
What should happen when a patient declines the scribe?
In the individual-care arrangement described by NHS England, dissent must be respected. Follow the local process and continue through an appropriate alternative documentation workflow.
Read the wider iatroX guide to AI-scribe transparency and patient choice