AI medical scribes in the UK: what they do and how to use one safely

What AI scribes and dictation tools do

The term covers three slightly different jobs, and it helps to know which one you need.

• Ambient scribes listen to the consultation itself, with the patient's agreement, and draft a structured note from the conversation. They are what NHS England calls ambient voice technology (AVT).
• AI dictation turns what the clinician says after the appointment into text, increasingly with formatting and medical vocabulary built in. This is the newer version of the digital dictation many consultants have used for years.
• Letter and document drafting takes notes or dictation and produces the finished output: a GP letter, a referral, a discharge summary or a patient-friendly summary.

Many products now do all three. None of them removes the clinician's responsibility. They can mishear a drug name, drop a negative ("no chest pain" becoming "chest pain"), or add something that was never said. NHS England's guidance is clear that users must review and approve outputs before anything else happens with them.

AI tools for therapists

Counsellors, psychologists and psychotherapists are an obvious audience for AI notes, and an especially careful one. Recording a therapy session raises harder questions about consent and confidentiality than recording a physio assessment, and some clients will say no. Many therapists who use AI prefer to dictate a short summary after the session rather than record the session itself. Whichever you choose, the same governance steps below apply.

The main AI medical scribe options in the UK

From each vendor's own website, checked October 2026. Regulatory claims are as the vendor states them; verify them before you rely on them.

Being on this list isn't a recommendation. NHS England keeps a self-certified register of ambient scribe suppliers, but says plainly that it doesn't endorse any of them, and the same caution applies here. Several practice management systems also sell their own scribes, including WriteUpp and Jane, and Clinicminds includes one for aesthetic clinics. Our guides to practice management software and aesthetic clinic software cover those.

What NHS England's guidance on ambient voice technology says

NHS England's guidance on AI-enabled ambient scribing products was first published in April 2025 and is now on version 3, last updated on 29 July 2026. In summary, a deploying organisation should:

• Appoint a Clinical Safety Officer, a named clinician responsible for identifying and managing the risks.
• Complete DCB0160 documentation, the clinical risk standard for organisations using health IT: a safety case, a hazard log and a plan for monitoring the tool in use.
• Check the supplier has completed DCB0129, the matching standard for manufacturers.
• Complete a Data Protection Impact Assessment. NHS England provides a template in its information governance guidance.
• Check medical device status. Whether a scribe counts as a medical device depends on what the manufacturer says it is for. Summarising, suggesting codes or anything that edges towards clinical decisions needs checking against MHRA guidance.
• Keep a human in the loop. Outputs must be reviewed and approved before they are used.

Two further points are easy to miss. A pilot or proof of concept doesn't exempt anyone from these steps, and the guidance suggests limiting pilots to about four months. And liability for claims arising from the tool can still sit with the organisation using it, so contracts need to say clearly who is responsible for what.

Does this apply to private clinics?

The guidance is written for health and care settings in England and is aimed mainly at NHS organisations. DCB0129 and DCB0160 are legally required of NHS bodies and of providers delivering NHS services; a purely private clinic is not bound in the same way. They are still the recognised UK benchmark for clinical safety in health software, and following them is the simplest way to show you took safety seriously if something goes wrong. The DCB0160 standard is free to read.

UK GDPR, DPIAs and telling patients

A consultation recording and the note made from it are health data, which UK GDPR treats as special category data. That brings three practical duties.

Do a DPIA before you start

The ICO lists large-scale processing of special category data as something that always needs a data protection impact assessment, and treats innovative technology such as AI as a trigger when combined with other risk factors. It also says a DPIA is good practice for any major new project involving personal data (ICO: when do we need a DPIA?). For a small clinic, the DPIA doesn't need to be long, but it should answer:

• What is recorded, and is audio kept or deleted after transcription?
• Where is the data processed and stored, and by which subprocessors?
• How long are transcripts and drafts kept?
• Does the supplier use your data to train its models?
• Who can access the outputs, and how are errors caught before they reach the record?

Have a proper contract

The supplier will be processing patient data on your behalf, so you need a data processing agreement with it. Ask for it before you sign up, not after.

Tell patients

Update your privacy notice, tell patients at the start of the appointment that you use a scribe, and make it easy for them to say no. A patient who declines should get exactly the same care.

Can I use ChatGPT in clinical practice?

For tasks with no patient information, often yes, with care. For patient information, a personal ChatGPT account is the wrong tool.

Plenty of clinicians use general AI assistants for work that contains no patient information: drafting a leaflet, rewording a policy, summarising published guidance. That is reasonable, provided you check the output, because these tools can state wrong things confidently.

Putting identifiable patient information into a personal account is different. OpenAI says it may use content from its Free, Plus and Pro plans to train its models unless you switch that off (OpenAI Help Centre). More fundamentally, a personal account is an agreement between OpenAI and you as an individual, not a processing contract between your clinic and a supplier, which is what UK GDPR expects when someone else handles patient data for you. A general-purpose chatbot also hasn't been through the clinical safety and medical device steps that NHS England expects of a scribe.

Is ChatGPT GDPR compliant for healthcare?

No tool is GDPR compliant on its own; compliance depends on how you use it. OpenAI's business products are a different proposition from personal accounts: OpenAI says it doesn't train on business data by default, and since October 2025 it has offered UK data residency at rest for eligible API customers and new ChatGPT Enterprise and Edu workspaces. Even then, you would still need a lawful basis and a special category condition, a DPIA, a data processing agreement, transparency with patients and, for clinical documentation, a clinical safety assessment.

How to choose an AI scribe for your clinic

1. Decide the job. Ambient notes for every consultation, dictation after the appointment, or finished letters? The best tool for one is not always the best for the others.
2. Check it works with your systems. If it can't file into your practice management system, someone ends up copying and pasting, which wipes out much of the time saved.
3. Ask for the paperwork. DCB0129 safety case, MHRA registration status, DTAC, the data processing agreement, hosting location, retention periods and the training policy. A good supplier sends these without fuss.
4. Trial it properly. Use it with consenting patients for a few weeks, and time the notes and letters before and after, so you know what it is really saving.
5. Name a person responsible. Even in a small private clinic, someone should own the hazard log, review errors and decide when to stop.

If you would like help weighing up a tool, or working out whether AI is the right answer to your admin problem at all, our AI consulting work page explains how we approach it.