Key takeaways
- Government policy backs ambient voice technology to cut clinical admin, with rollout support planned within two years[1].
- MHRA guidance from July 2026: products solely for transcription, summarising, drafting letters or suggesting codes for review are not medical devices[2].
- Before go-live, a practice needs clinical safety work under DCB0160, a DPIA, staff training and a clear process for clinicians to check every output[3][4].
- 11% of GPs use AI tools they obtained themselves[5]; bringing scribes under practice governance is the first job.
What AI scribes are and why policy favours them
Ambient voice technology, usually called AI scribes, listens to a consultation, transcribes it and drafts notes, letters or summaries for the clinician to check. The appeal is time. The government's 10 Year Health Plan commits to rolling out ambient voice technology, digital triage and the single patient record to cut clinical admin within two years[1]. It estimates that saving just 90 seconds per GP appointment would generate over 2,000 full-time-equivalent GPs' worth of capacity[6].
Adoption is already under way, often informally. In a 2025 Nuffield Trust and RCGP survey of 2,108 GPs, 28% said they currently use AI tools in clinical practice[7]. Of all GPs, 13% use tools provided by their practice and 11% use tools they obtained independently[5]. That second group is the governance problem every practice manager should know about.
The 2026 position on medical device status
In July 2026 the MHRA published guidance, developed with NHS England, clarifying how medical device law applies to ambient voice technology[2]. The key line is about intended purpose.
| What the product is intended to do | Regulated as a medical device? | What it means for the practice |
|---|---|---|
| Transcription only | No | Still needs safe deployment, governance and output review |
| Summarising clinical conversations | No | Clinician checks every summary before it enters the record |
| Drafting letters | No | Letters are signed off by the clinician |
| Suggesting clinical codes for a clinician to review | No | Codes are confirmed by a person |
| Supporting diagnosis, treatment or prevention | Yes | Product must meet medical device requirements |
| Taking automated action, such as placing orders without clinician review | Yes | Product must meet medical device requirements |
The guidance also asks NHS boards and executive teams to assure themselves that deployment is supported by appropriate clinical oversight, local governance, staff training and procurement processes, and that any change in a product's functionality that may alter its regulatory status is identified and reviewed[2]. Clinicians remain responsible for reviewing and verifying AI-generated transcripts, summaries and other outputs[2]. For a practice, that last point is the heart of safe use.
What to have in place before go-live
- Clinical safety: DCB0129 sets requirements for technology manufacturers and DCB0160 covers how health and care organisations deploy and use health IT safely; both are mandatory[3]. Ask the supplier for its DCB0129 safety case and complete your own DCB0160 work with a Clinical Safety Officer.
- Data protection: where processing using new technologies is likely to be high risk, UK GDPR requires an impact assessment before processing starts, and large-scale processing of health data is a named example[4]. Complete a DPIA covering recording, transcription, storage, retention and any use of data by the supplier.
- Supplier assurance: a contract with data processing terms, clarity on hosting and subcontractors, security assurance, and confirmation of the product's intended purpose and regulatory status.
- Integration: how drafts reach the clinical record, and confirmation that nothing is filed without a clinician's approval.
- Training: how to start and stop recording, how to check drafts, what to do when the output is wrong, and when not to use the tool.
- Patient information: a privacy notice update, information in the waiting room and online, and a simple way for patients to say no.
- Change control: a named person who reviews supplier updates for new features that could change the regulatory status.
Why GPs are cautious, and how governance answers it
Among GPs not using AI, 89% were concerned about professional liability and medico-legal issues, and 88% about a lack of regulatory oversight[8]. Patient privacy and data security worried 82%[9]. These concerns are reasonable, and the adoption checklist above is largely a response to them: clear regulatory status, clinical safety documentation, a DPIA and a practice process that keeps the clinician in control.
Talk to your medical defence organisation before rollout. Ask how it views the use of ambient scribes and what it expects to see in your records of review and consent.
An adoption pathway for a practice
Find out what is already in use
Ask every clinician, in a no-blame way, whether they use any AI tool for notes or letters. Pause unapproved tools until assessed.
Check the ICB position
Many ICBs have preferred products, funding or assurance work already done. Start there.
Choose and assure the product
Confirm intended purpose, regulatory status, DCB0129 documentation, security and data processing terms.
Complete DCB0160 and the DPIA
With a Clinical Safety Officer, build a hazard log and safety case for your setting, and complete the DPIA.
Pilot with a few clinicians
Run for four to six weeks with clear exclusions, audit a sample of outputs and record time saved and errors found.
Decide and scale
Review the pilot with the CSO and partners, update documents, train everyone and set a review date.
What goes into the record
Decide in advance what the permanent record will contain. Most practices file only the clinician-approved note or letter, not the raw transcript or audio. That keeps the record readable and limits how much sensitive data is stored. Whatever you choose, write it into the DPIA and the safety case, and make sure the product is configured to match.
- Approved note: always reviewed and edited by the clinician before saving.
- Transcript: usually not kept, or kept briefly for checking, then deleted.
- Audio: deleted as soon as the transcript is produced, unless your assessment justifies keeping it.
- Audit trail: a record that the AI tool was used for that consultation and who approved the output.
If a patient later asks for their records, they are entitled to what you hold. Retaining less, with clear rules, makes those requests simpler.
NHS dental practices and other settings
The same principles apply to NHS dental practices, community services and private clinics. Dental consultations often involve charting, treatment options and costs, so test whether the product handles tooth notation and treatment plans accurately before relying on it. Practices that run both NHS and private work should apply one governance process to both, so clinicians do not use different tools under different rules depending on who is paying.
Talking to patients
Patients should know when a consultation is being recorded and transcribed and be able to say no without affecting their care. A short script at the start works well:
"I use a secure tool that listens to our conversation and drafts my notes, so I can focus on you. I check everything before it goes into your record, and the recording is deleted afterwards. Are you happy for me to use it today? It is fine to say no."
- Adjust the deletion wording to match your actual retention settings.
- Offer to switch it off part-way through if the conversation becomes sensitive.
- Record the patient's decision in the way your DPIA and safety case require.
- Do not use the tool when a patient lacks capacity to agree and no suitable decision-maker is present, unless your local policy covers it.
After go-live: keep checking
Safety work does not stop at launch. Review a sample of AI-drafted notes and letters each month against the clinician's memory of the consultation, log any errors in the hazard log, and watch for patterns such as missed negatives, wrong medicines or details that were not said. Re-check the product's regulatory status whenever the supplier adds features, because a tool that starts suggesting diagnoses or placing orders may become a medical device[2].
28%
of GPs said they currently use AI tools in clinical practice in 2025[7]
Example: bringing an informal tool under control
Example: in a fictional practice, two GPs had started using a free AI note-taking app on their own phones. The practice manager paused its use, explained why, and worked with the ICB to assess an approved product instead. The practice's Clinical Safety Officer completed a short safety case, the DPIA identified retention settings that needed changing, and a six-week pilot showed time savings with a handful of correctable errors. Both GPs moved to the approved tool, and the practice added AI tools to its staff induction.
Frequently asked questions
Is an AI scribe a medical device?
It depends on intended purpose. MHRA guidance from July 2026 says products solely for transcription, summarising, drafting letters or suggesting codes for review are not, while those supporting diagnosis or treatment or taking automated action are[2].
Do we need a DCB0160 safety case for an AI scribe?
Yes. DCB0160 covers how health and care organisations deploy and use health IT systems safely and is mandatory[3]. Work with a Clinical Safety Officer.
Do patients have to consent to AI scribes?
Patients should be told and given a simple way to decline. Record their decision as your DPIA and safety case require, and keep the explanation honest about retention.
Can GPs use free AI apps on their own phones?
They should not use unapproved tools with patient data. Bring any tool under practice governance with a DPIA, safety assessment and data processing terms first.
References
GOV.UK (Department of Health and Social Care) (2025). Fit for the future: 10 Year Health Plan for England (opens in a new tab) ↑
GOV.UK (Medicines and Healthcare products Regulatory Agency) (2026). MHRA clarifies regulatory status of ambient voice technologies used in the NHS (opens in a new tab) ↑
NHS England (2026). National review of clinical risk management standards DCB0129 and DCB0160: consultation questions (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (2026). UK GDPR Article 35: Data protection impact assessment (opens in a new tab) ↑
Nuffield Trust (2025). How are GPs using AI? Insights from the front line (opens in a new tab) ↑
GOV.UK (Department of Health and Social Care) (2025). Fit for the future: 10 Year Health Plan for England (opens in a new tab) ↑
Nuffield Trust (2025). How are GPs using AI? Insights from the front line (opens in a new tab) ↑
Nuffield Trust (2025). How are GPs using AI? Insights from the front line (opens in a new tab) ↑
Nuffield Trust (2025). How are GPs using AI? Insights from the front line (opens in a new tab) ↑