AI scribes are not just admin tools. In the NHS, they are information-governance tools, patient-trust tools, and clinical-record tools. The governance questions go far beyond "does the note look accurate?" — extending to consent, objection, privacy, vulnerable patients, clinical appropriateness, record accuracy, and how ambient listening fundamentally changes the dynamics of the clinical consultation.
NHS England's IG guidance states that ambient scribes listen in the background to capture speech or conversations and produce task-specific outputs such as summaries or letters. Patients should be told at the beginning of the session if an ambient scribe is being used, and patients can ask for one not to be used. The guidance covers use for individual care and does not extend to research use or training of new AI products, which have different legal requirements.
What NHS England Guidance Says
The practical essentials: ambient scribes capture consultation speech and produce clinical documentation — notes, summaries, letters, coding suggestions, and task lists. Organisations may deploy them to reduce administrative burden and improve consultation quality. Patients should be informed proactively. Patients can object, and the consultation must proceed normally without the scribe if they do. Organisations must assess product safety, output accuracy, and information governance compliance.
NHS England's clinical guidance (April 2025, updated to Version 2 in April 2026) covers implementation, clinical safety, information governance, and integration — aligned with the AVT Supplier Registry. The IG guidance, reviewed by the ICO and National Data Guardian, was published with a template DPIA in March 2026. The guidance references DCB 0129 (manufacturer) and DCB 0160 (deploying organisation) clinical safety standards. East Lancashire Hospitals Trust is hosting a national programme to deploy AVT across the NHS in England.
Why Patient Objection Matters
Consent and objection are not only legal issues. They are trust issues — and trust determines whether ambient scribing is sustainable at scale.
Some patients will disclose less if they know an AI tool is listening. Consultations involving mental health, sexual health, safeguarding, domestic abuse, or stigmatised conditions may be particularly affected — the patient may self-censor, omit critical details, or avoid raising sensitive topics entirely. Other patients may welcome ambient scribing because the clinician appears less screen-focused and more present — the documentation happens in the background rather than competing with eye contact.
The key governance point: the conversation must remain clinically safe even if the patient refuses the scribe. Objection should be logistically straightforward (the clinician presses a button, the scribe is off) and must not create any pressure — implicit or explicit — to accept. If the workflow makes it awkward or time-consuming to proceed without the scribe, patients will feel pressured to agree.
Transparency is the legal requirement (NHS England is clear that patients must be informed). But transparency serves a deeper purpose: patients who discover retrospectively that AI was listening during their consultation will understandably feel their privacy was violated — even if the processing was technically lawful under Article 6(1)(e) and Article 9(2)(h) of UK GDPR. Proactive disclosure builds the trust that sustainable long-term AVT adoption requires.
The Mental-Health, Learning-Disability, and Neurodivergence Gap
This is where the governance picture becomes significantly more complex. The Royal College of Psychiatrists (RCPsych) and NHS England announced in May 2026 that they are developing guidance for Ambient Voice Technology in clinical practice with people with mental health needs, learning disabilities, or neurodivergent people. The announcement states that this group may be more vulnerable because of their conditions, capacity issues, and sensitivities around privacy and trust. Guidance is expected by September 2026.
The gap matters because some of the consultations where privacy, trust, and capacity are most sensitive are precisely the consultations where ambient scribes are being deployed — and where generic guidance may be insufficient.
A patient experiencing persecutory delusions may interpret ambient listening as surveillance, potentially exacerbating their symptoms. A patient with severe anxiety or OCD with surveillance-related themes may find the knowledge of recording deeply distressing. A patient with a learning disability may not understand what the scribe does or what their objection rights are. A patient whose capacity is fluctuating may be unable to give or withhold meaningful informed agreement at the start of the consultation.
The RCPsych/NHS England guidance — when it arrives — will need to address these specific scenarios with the clinical nuance that generic AVT guidance cannot provide.
When Might an AI Scribe Be Clinically Inappropriate?
Practical guidance for clinicians considering whether to use an ambient scribe in specific clinical scenarios.
Acute psychosis or paranoia — ambient listening may be interpreted as surveillance, potentially worsening symptoms or damaging the therapeutic relationship.
Safeguarding disclosure — a patient disclosing abuse, neglect, or exploitation needs maximum confidence that their disclosure is handled with sensitivity. The presence of recording may inhibit disclosure of information that is critical to their safety.
Domestic abuse — a patient accompanied by a controlling partner may be unable to disclose safely if they know the conversation is being captured — particularly if the partner might later demand to see records.
Suicidal ideation and self-harm — risk assessment conversations require the clinician's full attention and the patient's trust. The quality of the risk assessment may be compromised if the patient feels the conversation is being recorded.
Capacity assessment — assessing mental capacity requires nuanced clinical interaction. The assessment itself should not be complicated by the patient's confusion or distress about the recording.
Severe anxiety around surveillance — patients with anxiety disorders, PTSD, or trauma-related hypervigilance may find ambient listening genuinely distressing rather than merely inconvenient.
Child protection — consultations involving child safeguarding require particular sensitivity around recording, consent (who consents for the child?), and information sharing between agencies.
Consultations requiring interpreter nuance — ambient scribes may not accurately process interpreted consultations, and the presence of recording may affect the interpreter-patient dynamic.
Patients with cognitive impairment — patients who cannot understand what the scribe does cannot meaningfully be informed about it — creating a consent gap that generic transparency measures do not address.
Suggested Clinician Wording
A reusable script:
"I sometimes use an approved tool that listens during the consultation and helps draft the clinical note. I will check the note before it goes into your record. You do not have to agree to this, and your care will not be affected if you prefer me not to use it. Would you like me to use it today?"
For time-pressured GP appointments:
"I'm using an AI note-taker today — it helps me focus on you rather than typing. I'll check the note before saving it. You can opt out at any time — just let me know."
For sensitive consultations where the clinician decides not to use the scribe:
(No script needed — the clinician simply does not activate the scribe. The absence of the tool does not need to be explained.)
Practice Implementation Checklist
Privacy notice updated to include AVT use. Waiting-room information available (poster, leaflet, or digital display). Staff script prepared and practised by all clinical staff. Patient objection workflow defined and tested (how to proceed without the scribe smoothly). Process for correcting AI-generated notes before saving to the record. Role-based access configured for who can view AI-generated content. Supplier due diligence completed against AVT Supplier Registry requirements. Data storage location confirmed (UK GDPR-compliant hosting). Audio retention policy documented (typically deleted after verified summary produced). Model training policy confirmed (patient data not used for AI training without explicit legal basis). DPIA completed using NHS England template. Clinical safety case documented per DCB 0129/0160. Incident-reporting process established for AI-related documentation errors. Staff training delivered on clinical review standards, consent processes, and objection handling.
Where iatroX Fits
iatroX is positioned around the clinical knowledge layer — not the documentation layer. The governance principles are shared: clinical AI needs explicit governance, source transparency, professional accountability, and patient trust. Documentation tools capture what happened. Clinical knowledge tools help verify what should happen. Both need governance proportionate to their clinical impact.
Use iatroX for clinical knowledge verification alongside documentation tools →