Ambient AI Scribes in Viet Nam: How Healthcare Organizations Should Govern Audio Capture, Documentation Accuracy, Patient Communication and Data Protection
IMPLEMENTATION INSIGHTS
Ambient AI Scribes in Viet Nam: How Healthcare Organizations Should Govern Audio Capture, Documentation Accuracy, Patient Communication and Data Protection
IMPLEMENTATION INSIGHT
Last reviewed: 3 October 2026
An ambient AI scribe can make a clinical encounter feel surprisingly simple. A clinician speaks with a patient, focuses more on the conversation than the keyboard, and receives a draft note at the end. What happens behind that experience is much less simple.
The system may capture speech from the patient, clinician and other people in the room; convert speech into text; process information through one or more cloud services; use generative AI to select, summarize and restructure information; and return documentation that may later become part of the medical record. Depending on the product and configuration, raw audio, transcripts, prompts, intermediate data or generated outputs may be retained for different periods, or may not be retained at all.
That makes ambient scribing more than a documentation-efficiency tool. It is simultaneously a clinical workflow, personal-data-processing activity, information-quality process and human-AI interaction.
International implementation is moving quickly. NHS England updated its ambient-scribing guidance to Version 3 on 29 July 2026, covering clinical safety, information governance, product regulation, implementation and continuing monitoring alongside the potential benefits of reducing documentation burden. The guidance is specific to England and should not be treated as a statement of Vietnamese law.
For healthcare organizations in Viet Nam, the relevant question is therefore not simply, “Can this technology write the note faster?” It is: “Can we allow an AI system to capture a clinical conversation, process sensitive information and help create a medical record while preserving patient trust, documentation accuracy, confidentiality and clear human accountability?”
Ambient AI scribing is not simply digital dictation
Traditional dictation usually begins with deliberate clinician input. The clinician decides what should be documented and dictates those words for transcription. Ambient AI scribes work differently: they capture speech or conversation with minimal user intervention and increasingly combine speech recognition with generative AI to produce task-specific outputs such as clinical notes, summaries, letters or other workflow documentation. NHS England uses a similarly broad concept of ambient-scribing products.
That distinction matters because the system may process much more than what a clinician would deliberately dictate. It may hear symptoms, medication discussions, family history, unrelated personal information, another person's voice, sensitive questions, background conversation and material that the clinician ultimately decides should not appear in the record.
The final note is also not simply a transcript. A generative system may select, omit, reorganize, compress and reinterpret information. In practical terms, it transforms what was said into what is written. Governance therefore needs to address at least three different layers: audio capture, intermediate processing and final clinical documentation. A product can perform acceptably at one layer and still create risk at another.
Audio capture is not the same as permanent audio recording
The word “recording” can be misleading because ambient products do not all handle audio in the same way. One product may create and retain an audio file. Another may stream audio for transcription and delete it soon afterward. Another may retain a transcript but not the original audio. A vendor may temporarily buffer audio in one location, process text in another and store only the generated note.
Healthcare organizations therefore need an architecture-level answer rather than a marketing statement such as “we do not store recordings.” Relevant questions include: Is raw audio created? Where is it processed? Is it retained and for how long? Is a transcript retained? Who can access it? Are subprocessors involved? Where are backups located? Can patient information be used for analytics, quality improvement, model training or product development? What actually happens when the vendor says data are deleted?
Even transient audio capture remains personal-data processing when identifiable personal information is captured or otherwise processed. Deleting the input later does not remove the need to govern the processing that occurred before deletion. For that reason, audio-capture architecture should be a core due-diligence issue, not a technical footnote.
The evidence is promising, but ambient scribes are not error-free technology
Ambient scribes are attractive because clinical documentation is burdensome, and early evidence suggests that they can reduce at least part of that burden.
A 2026 prospective multicentre study in Dutch general practice involved 12 GPs or GPs in training and 535 observed consultations. Ambient scribing reduced documentation time by an average of 42.7 seconds per consultation, while total consultation time did not significantly change. The study also identified potential benefits in perceived workload while reporting limitations involving summary accuracy, non-verbal information and some sensitive or linguistically complex encounters.
A pragmatic prospective pilot at UC Davis Health offers another useful perspective. Thirty-one physicians used an ambient listening AI scribe to create 7,545 notes, and physicians conducted detailed quality assessments on 356 of them. Among those reviewed notes, accidental omissions occurred in 18%, hallucinations in 11.5% and accidental inclusions in 9.3%; 19 of the 356 reviewed notes contained an error rated as posing serious or imminent risk if left uncorrected. Because detailed review covered only a subset of all notes, those percentages should not be treated as prevalence estimates for ambient AI generally. The more important conclusion is that careful clinician review remained necessary even when overall note quality was high.
A September 2026 review in BMJ Digital Health & AI reached a broader conclusion: ambient-scribe research has focused heavily on productivity and note generation while giving less attention to sociotechnical effects on workflow, professional roles, organizational processes and the patient–clinician relationship. The authors argue for treating ambient scribes as complex interventions rather than simple productivity tools.
The evidence therefore supports neither extreme. Ambient scribes should not be dismissed as inherently unsafe, but neither should fluent note generation be mistaken for a mature, solved clinical technology.
Documentation accuracy is a patient-safety control, not an editing preference
A fluent note can still be wrong.
Ambient AI may omit a clinically important symptom, misattribute information to the wrong speaker, misunderstand a medication name, convert uncertainty into certainty, place information under the wrong heading or generate a plausible statement that was never actually said. It may also fail to capture important information that was never verbalized, including examination findings, gestures, visual observations or other clinical context.
These errors may be particularly difficult to detect because generated text often looks polished. Healthcare organizations should therefore treat ambient-generated documentation as draft documentation requiring meaningful human review, not as automatically authoritative clinical information. NHS England likewise emphasizes professional review of ambient-scribe outputs before they are relied upon as part of individual care.
Meaningful review is more than confirming that a note reads well. The reviewer should be able to identify omissions, unsupported additions, wrong-speaker attribution, medication errors, changes in clinical meaning and failure to capture important context. The workflow also needs enough time for that review to be real. If an ambient scribe saves several minutes of typing but creates a two-second “approve” habit, the organization may have exchanged documentation burden for automation bias.
A clinician's signature does not make an inaccurate AI-generated note safe unless the review process makes meaningful error detection realistically possible.
Viet Nam's medical-record framework still places human accountability at the center
The current consolidated Law on Medical Examination and Treatment is Consolidated Document No. 26/VBHN-VPQH dated 26 February 2026. It protects the confidentiality of information in medical records and other private information provided during care, subject to legally permitted exceptions. The framework also requires people authorized to use medical-record information to maintain confidentiality and use it for the permitted purpose.
Circular No. 13/2025/TT-BYT has governed electronic medical-record implementation since 21 July 2025. It defines the electronic medical record as one created, updated, displayed, signed, stored, managed, used and accessed electronically, and it provides recognized forms of electronic signing or confirmation by healthcare workers, patients or their representatives.
Ambient AI does not displace these human accountability arrangements. An AI system may contribute substantially to a draft, but the healthcare organization still needs an identifiable professional process for determining what becomes part of the patient's medical record.
Current Vietnamese medical-record rules reviewed for this article do not impose a general requirement that every AI-assisted clinical note visibly display an “AI-generated” label. Healthcare organizations may nevertheless choose to preserve AI provenance internally so that they can later identify which system and version contributed to a note, when it was used and who reviewed the final documentation. That is an organizational governance control rather than a universal visible-label requirement.
Personal-data law makes ambient scribing particularly sensitive in Viet Nam
Ambient scribes process information that sits at the center of Viet Nam's personal-data-protection framework.
The Personal Data Protection Law No. 91/2025/QH15 has been effective since 1 January 2026. Article 26 provides that collecting and processing health information requires the data subject's consent, except in the cases specified in Article 19(1). Decree No. 356/2025/NĐ-CP separately lists health status among categories of sensitive personal data and requires organizations processing sensitive personal data to establish access restrictions, processing procedures and security measures.
Where consent is required, Article 9 provides that it must be voluntary and based on the data subject being informed about the data being processed, the processing purpose, the relevant controller or controller-processor and the data subject's rights and obligations. Consent must be expressed clearly and in a verifiable form; silence or failure to respond does not constitute consent.
Vietnamese healthcare organizations should therefore be cautious when importing consent practices from other jurisdictions. NHS England states that explicit consent is not generally required for use of an ambient scribe for individual care in England and that implied consent may be relied upon under the English common-law duty of confidentiality, provided people are informed and given an opportunity to dissent. That is a UK-specific legal approach.
It should not simply be copied into Viet Nam.
A Vietnamese healthcare organization should determine whether its proposed ambient-scribe processing requires consent under Article 26 or whether a specific case under Article 19(1) applies. Article 19(1) includes, among other cases, processing necessary to perform an agreement between the data subject and the relevant organization. The existence of a general healthcare relationship, however, should not automatically be assumed to cover every form of AI audio capture, third-party cloud processing or secondary use; applicability needs to be assessed against the actual processing activity and purpose.
The appropriate conclusion is therefore not that every ambient-scribe deployment always requires a separate signed consent form, and not that consent is unnecessary. It is that the organization should establish and document why the proposed audio capture and health-data processing are legally permitted before deployment.
Patient communication remains important even when consent is not the operative requirement
Legal permission to process information and good clinical communication are related but different questions.
A patient entering a consultation may reasonably assume that the conversation is being heard by the people in the room. They may not expect an AI service to process the conversation, generate intermediate text, communicate with external cloud infrastructure or involve a commercial technology provider.
Healthcare organizations should therefore develop a simple and consistent way of explaining ambient scribing before use. Depending on the actual system and applicable legal position, patients should be told in plain language what the tool does, whether audio is retained, what happens to the resulting documentation, who remains responsible for the final record and how concerns can be raised.
The explanation does not need to become a technology lecture. A practical communication model answers questions such as: What is capturing the conversation? Why are we using it? What happens to the audio or transcript? Who checks the note? What can I do if I have concerns about its use?
International evidence suggests that context matters. The Dutch study identified sensitive conversations as one area in which ambient scribing may affect patient or clinician behavior. Clinician discretion therefore remains important. A tool that is appropriate for a routine follow-up may not be appropriate for every consultation, and the user interface should make it easy to pause or stop capture immediately when necessary.
The primary patient may also not be the only person whose data are processed. Caregivers, parents, interpreters, spouses, children, other family members or staff may speak during the encounter, while the patient may disclose personal information about third parties. A patient's consent or other applicable processing basis should not automatically be assumed to resolve the position for every other person's data. Multi-person consultations, children, people with impaired decision-making capacity and incidental third-party information therefore need explicit organizational procedures.
Audio, transcripts, drafts and final records should not automatically share one retention period
A common design mistake is to treat everything produced by an ambient scribe as part of the medical record.
The final reviewed clinical note may need to be retained under medical-record requirements. Raw audio may serve only as temporary input. An intermediate transcript may exist solely to create a draft. System logs may serve cybersecurity, quality-assurance, incident-management or audit purposes. These artifacts have different functions and may justify different retention arrangements.
The organization should therefore distinguish:
raw audio → transcript → generated draft → final clinical record → provenance/audit information.
For each layer, it should be able to answer: Why are we retaining this information, for how long, who may access it and what event triggers deletion or archival?
“Keep everything in case we need it later” is not a sound data-governance policy. Nor is immediate deletion automatically correct if another legal retention requirement or a legitimate patient-safety need applies. Retention should follow a defined purpose and applicable law.
Impact assessments and cross-border processing should be addressed before go-live
Many ambient AI systems use cloud infrastructure. Cloud use does not automatically make a product unsuitable for healthcare, but it makes the data architecture central to governance.
Article 20 of the Personal Data Protection Law treats several arrangements as cross-border transfers, including transferring personal data stored in Viet Nam to a system outside Viet Nam, transferring data from an organization in Viet Nam to an overseas organization or individual, and using a platform outside Viet Nam to process personal data collected in Viet Nam.
Healthcare organizations therefore need a data-flow map showing where raw audio, transcripts, prompts, intermediate information and outputs travel; which organizations process them; which countries are involved; which subprocessors participate; what information returns to the healthcare organization; and where residual copies or backups may remain.
For organizations acting as personal-data controllers or controller-processors, Article 21 requires a personal-data-processing impact-assessment dossier to be prepared, retained and submitted to the specialized personal-data-protection authority within 60 days from the first day of processing, subject to the statutory exemptions. The Law also requires that these dossiers be updated when relevant information changes.
Where a deployment involves a cross-border transfer under Article 20, the person or organization carrying out the transfer must also prepare a cross-border-transfer impact-assessment dossier and submit one original dossier within 60 days from the first cross-border transfer, again subject to the statutory exemptions.
These are not merely procurement-formality issues. Ambient-scribe architectures can change after deployment through new subprocessors, model providers, storage arrangements or cloud regions. Impact-assessment documentation therefore needs to be maintained and updated in accordance with the applicable legal requirements.
Vendor contracts should address permitted processing purposes, security, confidentiality, retention and deletion, subprocessors, incident notification, data export, service termination, model training, secondary use and access to information needed for investigation.
One question deserves a direct answer before contract signature:
Can the supplier use our patients' conversations, transcripts or clinical notes to train, fine-tune or improve its models or services?
That should never be left to inference from a generic website privacy notice.
Local validation matters because Vietnamese clinical conversations are not generic English speech
A product that performs well in an English-speaking ambulatory clinic is not automatically ready for routine use in Viet Nam.
Vietnamese clinical encounters can include regional accents, English medical terminology, drug names, abbreviations, informal expressions, multiple speakers and switching between Vietnamese and English. Performance can also be affected by masks, telephones, background noise, crowded rooms and specialty-specific language.
Ambient documentation therefore has a layered failure pathway:
speech capture → transcription → interpretation → summarization → clinical note.
An error introduced at the first stage may become a fluent but incorrect statement at the last.
The Dutch 2026 study identified limitations involving summary quality, non-verbal information and complex conversational contexts. Nature A 2026 npj Digital Medicine perspective similarly emphasizes that moving ambient scribes from relatively simple outpatient environments into more diverse care settings creates additional clinical, technical and ethical challenges and argues for context-specific evaluation before wider deployment.
For Viet Nam, local validation should therefore include the actual languages, accents, specialties, patient populations, room conditions and workflows in which the organization intends to deploy the product.
The most useful question is not simply, “What transcription accuracy does the vendor report?” It is: “What clinically meaningful errors occur in our consultations, and can our clinicians reliably detect and correct them before the note becomes part of the medical record?”
Ambient scribing should remain within its approved function unless expanded capabilities are separately governed
Ambient products are evolving rapidly. A product that initially creates draft notes may later add coding recommendations, referral generation, diagnostic suggestions, order preparation, patient-message drafting or autonomous workflow actions.
That can materially change the risk profile of the technology.
Under Decree No. 142/2026/NĐ-CP, providers must classify AI systems before they are put into use. A deployer must coordinate with the provider to review and reclassify the system when deployment, modification, integration or a change in function or purpose creates new or higher risk. Providers must also review the classification when significant changes affect the original risk classification.
For systems already classified as high-risk, Article 13 goes further: significant changes affecting the original conformity assessment, including changes to principal functions, intended use, AI model, technical configuration, key input data or integration environment, may require reassessment of conformity.
These two processes should not be confused. A material change may trigger risk review or reclassification for an AI system generally; conformity reassessment is a specific requirement for high-risk AI systems when the conditions in Article 13 are met.
Ambient scribes are also not automatically high-risk under Vietnamese law simply because they are used in healthcare. The current high-risk list under Decision No. 33/2026/QĐ-TTg does not create a standalone healthcare category for ambient clinical documentation tools. Internal clinical risk may nevertheless be meaningful even when statutory high-risk classification does not apply.
That distinction is important: not legally high-risk does not mean clinically low-risk.
Medical-device status should similarly be assessed from the product's actual intended purpose and functionality rather than its marketing name. Viet Nam's current consolidated medical-device framework expressly includes software within the definition of a medical device when the statutory intended-use criteria are met, including specified purposes relating to diagnosis, prevention, monitoring or treatment. A documentation-only product should therefore not automatically be assumed either to be, or not to be, a medical device; the actual intended purpose needs to be assessed against the regulatory definition.
If a supplier adds new clinical functionality, the healthcare organization should reassess that function before enabling it rather than assuming that approval of the original scribe automatically covers the expanded capability.
Provenance should link the final note back to the AI-assisted workflow
Ambient documentation is one of the clearest real-world uses for AI provenance.
If a safety issue is discovered months later, the organization may need to establish whether a particular note was manually written or AI-assisted, which system and version generated the draft, which encounter was involved, who reviewed the draft and whether material edits were made.
That information does not need to clutter the visible clinical record. It can be maintained through structured metadata, system logs or linked governance records. Where clinically relevant, the organization should be able to reconstruct:
AI system → relevant version or configuration → clinical encounter → generated draft → human reviewer → final approved record.
This is especially useful if a vendor later reports a defect affecting one model version or the organization needs to perform retrospective review.
Provenance also helps preserve role clarity. The AI may contribute to generating the draft; the healthcare professional reviews and confirms the final clinical documentation. Those are related but different functions.
Monitoring should look beyond adoption rates and time saved
An ambient scribe can be popular with clinicians and still perform poorly in particular situations.
Post-deployment monitoring should therefore extend beyond utilization rates and documentation time. Relevant signals may include correction rates, clinically important omissions, hallucinations, medication errors, speaker-attribution errors, note sections frequently edited, performance across languages and specialties, patient concerns, privacy incidents, use in sensitive consultations, downtime, vendor changes and signs of clinician overreliance.
NHS England's current guidance likewise treats ambient-scribe implementation as an ongoing clinical-safety and governance activity rather than something completed at procurement.
Monitoring should also look for workflow drift. If clinicians gradually stop making any contemporaneous notes because they assume the ambient tool will capture everything, a later failure may have greater consequences than it did during the pilot. If editing time progressively approaches zero, that might mean the product has improved — or it might mean review has become superficial.
Metrics should therefore trigger investigation rather than become simplistic targets. The safety question is not “How many clinicians are using it?” It is “Is the human–AI documentation process continuing to produce a reliable clinical record?”
A practical Ambient AI Scribe Governance Gate
Before approving or materially changing an ambient AI scribe, healthcare organizations can ask:
Intended use: Is the product only drafting documentation, or does it also analyze, recommend, code, order, refer or take other actions?
Audio capture: What exactly is captured, and does the system create or retain raw audio?
Participants: Whose voices and personal information may be processed: patient, clinician, caregiver, interpreter, family member or other people?
Consent or exception: Does Article 26 require consent for this processing, or does a specific case under Article 19(1) apply?
Patient communication: What will patients be told about the technology, its purpose, data processing and human review?
Sensitive encounters: When should clinicians pause or avoid the scribe because ambient capture is inappropriate for the conversation?
Data flow: Where do audio, transcripts, prompts, intermediate data and generated outputs travel and where are they stored?
Vendor roles: Which organizations and subprocessors process the information, for what purposes and under what contractual controls?
Secondary use: Can patient or organizational data be used for model training, product improvement, analytics or any purpose beyond producing documentation for the encounter?
Impact assessment: Have applicable personal-data-processing and cross-border-transfer impact-assessment obligations been addressed within the statutory timelines?
Cross-border processing: Does the architecture use overseas storage, foreign recipients or platforms outside Viet Nam to process personal data collected in Viet Nam?
Retention: How long are raw audio, transcripts, drafts, final notes and audit/provenance information retained, and why?
Accuracy: What clinically meaningful errors occur in the intended language, specialty and local workflow?
Human review: Who reviews the draft, what must they check, and does the workflow provide enough time for meaningful review?
Clinical-record integration: Can generated text enter the finalized record only after the required professional review and confirmation?
Provenance: Can the organization reconstruct which system and version contributed to a particular note?
Regulatory status: What AI risk classification and medical-device status apply to the functionality actually being deployed?
Change control: What happens when the vendor changes the model, function, data flow, subprocessors or infrastructure?
Monitoring: Which documentation-quality, safety, privacy, workflow and human-factors signals will be reviewed after go-live?
Incident response: Can the organization identify affected encounters, preserve evidence and stop the product rapidly if a safety or data-protection concern emerges?
Fallback: Can clinicians continue documenting safely if the ambient scribe is unavailable, suspended or withdrawn?
The Ambient AI Scribe Governance Gate is a practical organizational framework proposed for healthcare governance. It is not a statutory checklist or regulatory procedure established by Vietnamese law.
What this means for healthcare organizations in Viet Nam
The timing is particularly relevant because electronic medical records and healthcare AI are advancing together. Circular No. 13/2025/TT-BYT required hospitals to implement electronic medical records no later than 30 September 2025, while other healthcare facilities within its specified scope have until 31 December 2026.
Ambient AI may become attractive in this environment because it promises to reduce documentation burden and allow clinicians to spend more of the encounter interacting directly with patients. International evidence shows genuine promise, but it also explains why this should be treated as an organizational implementation rather than a personal productivity app. Current studies continue to identify omissions, hallucinations, workflow effects, sensitive-conversation concerns and the need for professional review.
Viet Nam now also has a substantially clearer legal environment than existed during many early ambient-scribe pilots. The Personal Data Protection Law and Decree No. 356 have been effective since January 2026, while the AI Law and Decree No. 142 have been effective since March and May 2026 respectively. Health information receives specific protection, certain cross-border processing arrangements trigger separate impact-assessment requirements, and AI deployment now sits within a formal risk-classification framework.
Ambient scribing therefore should not enter a hospital simply because an individual clinician subscribes to an app. It needs an organizational pathway for evaluation, approval, data governance, implementation, monitoring and incident response.
The technology can be genuinely useful. Its value, however, depends on what surrounds the model: how patients are informed, how data move, whether sensitive conversations can be protected, how drafts are reviewed, whether Vietnamese-language performance is adequate, how vendors and product changes are governed and whether the organization can reconstruct what happened later.
The most useful leadership question is therefore not, “Can AI write our clinical notes?”
It is:
“Can we use ambient AI to reduce documentation burden without weakening confidentiality, record accuracy, patient trust or human accountability?”
That is the standard responsible implementation should aim for.
Professional education notice
This article is intended for professional education and knowledge exchange. It does not constitute individualized legal, regulatory, clinical, cybersecurity, personal-data-protection or medical-device advice.
“Ambient AI scribe,” “ambient scribing,” “audio capture” and the Ambient AI Scribe Governance Gate are used here as practical technology and governance concepts. Product architectures differ substantially: some products retain audio, others process it transiently; some only generate documentation, while others perform additional clinical or workflow functions. Legal and governance assessment should therefore be based on the actual system, purpose, configuration and deployment rather than the product label.
Under Article 26 of Viet Nam's Personal Data Protection Law, collecting and processing health information generally requires the data subject's consent unless a specific case under Article 19(1) applies. This article does not conclude that every ambient-scribe deployment necessarily requires a separate written consent form, nor that consent may generally be dispensed with. Healthcare organizations should establish and document the legal position applicable to the specific workflow before implementation.
The NHS England guidance discussed here reflects the legal and information-governance framework of England. Its approach to implied consent for individual care should not be interpreted as the position under Vietnamese law.
Ambient AI scribes are not currently listed as a standalone high-risk healthcare category under Decision No. 33/2026/QĐ-TTg. This does not establish that all ambient scribes are legally low-risk or clinically low-risk. Classification depends on actual functionality, purpose and deployment context, and subsequent changes may alter that assessment.
References
[1] National Assembly of Viet Nam. Law No. 91/2025/QH15 on Personal Data Protection. Adopted 26 June 2025; effective 1 January 2026.
[2] Government of Viet Nam. Decree No. 356/2025/NĐ-CP Detailing Certain Articles and Measures for Implementation of the Law on Personal Data Protection. Issued 31 December 2025; effective 1 January 2026.
[3] Office of the National Assembly of Viet Nam. Consolidated Document No. 26/VBHN-VPQH on the Law on Medical Examination and Treatment. Consolidated 26 February 2026.
[4] Ministry of Health of Viet Nam. Circular No. 13/2025/TT-BYT Providing Guidance on Implementation of Electronic Medical Records. Effective 21 July 2025.
[5] National Assembly of Viet Nam. Law No. 134/2025/QH15 on Artificial Intelligence. Effective 1 March 2026.
[6] Government of Viet Nam. Decree No. 142/2026/NĐ-CP Detailing a Number of Articles and Measures for Implementation of the Law on Artificial Intelligence. Effective 1 May 2026.
[7] Prime Minister of Viet Nam. Decision No. 33/2026/QĐ-TTg Promulgating the List of High-Risk Artificial Intelligence Systems. Effective 15 August 2026.
[8] Ministry of Health of Viet Nam. Consolidated Document No. 08/VBHN-BYT on Medical Device Management. Consolidated 6 March 2026.
[9] NHS England. Guidance on the Use of AI-Enabled Ambient Scribing Products in Health and Care Settings. Version 3, updated 29 July 2026.
[10] NHS England. Using AI-Enabled Ambient Scribing Products in Health and Care Settings: Information Governance Guidance. 2026.
[11] van Linschoten RCA, van Loon CM, Joanknecht L, et al. Ambient scribe in general practice: a multi-perspective before-after longitudinal mixed-methods study. npj Digital Medicine. 2026;9:299.
[12] Taylor SL, Jost M, MacDonald S, et al. Quality of Clinical Notes Created by Ambient Listening Generative AI: Pragmatic Prospective Pilot Study. JMIR Medical Informatics. 2026;14:e86474.
[13] Ohde JW, Thompson A, Liu Z, et al. Barriers and opportunities of scaling ambient AI scribes for clinical documentation across diverse healthcare settings. npj Digital Medicine. 2026;9:369.
[14] Seuren LM, Williams R, Cresswell K. Beyond productivity: a NASSS-informed review of implementation risks and research priorities for ambient AI scribes in healthcare. BMJ Digital Health & AI. 2026;2:e000091.