Best AI Tools for Clinicians in the UK: MHRA Status and Data Location

UK clinicians can use AI scribes such as Jane, Tandem and Heidi for notes, and ChatGPT or Claude for work with no patient data. MHRA status depends on function: a scribe that only transcribes or summarises is not a medical device; one that suggests diagnoses is. Data location then decides what your DPIA must cover.

That rule comes from MHRA guidance first published on 29 July 2026, and it cuts both ways. A "registered Class I" badge tells you how a vendor chose to register, not that its product is safer than an unregistered summariser, while a tool that adds insights, coding or suggested diagnoses can fall inside the device rules whatever its disclaimer says. An NHS service also follows NHS England's ambient scribing guidance and completes its own safety and data assessments; a private practice has more freedom but should keep the same records.

Follow me on Instagram@sagnikteaches

The two questions that sort the shortlist

Most tool comparisons list features. Features matter less than two questions that either rule a product in or out.

Connect on LinkedInSagnik Bhattacharya

Is it a medical device, and which regime applies?

MHRA's guidance on ambient voice technology-enabled products works through examples. A scribe solely intended to produce a transcript or a summary of a clinical conversation for the clinician to review is not a medical device, and nor is one that drafts a discharge summary or letter from that conversation. A product becomes a device when it has a medical purpose: offering "generated insights" with suggested diagnoses or treatment options, being marketed as guiding diagnosis and treatment planning, or ordering follow-up tests without clinician confirmation. Where a device provides decisive information for a diagnosis, it may be class IIa, and class IIa and above need certification by an approved body (for UKCA) or a notified body (for CE). MHRA adds that its guidance documents describe its current thinking and are recommendations unless they cite a statutory requirement.

Subscribe on YouTube@sagnikteaches

NHS England's guidance on AI-enabled ambient scribing products has been updated to align with the MHRA guidance. It sits alongside a national, self-certified AVT Supplier Registry, which sets expectations for suppliers including a DTAC assessment (the Digital Technology Assessment Criteria) and MHRA registration where applicable. NHS England says it does not endorse suppliers on the registry and only runs preliminary completion checks, so the registry is a starting point for your due diligence, not a substitute for it.

What you can do without a lawyer is ask each vendor the same three things in writing: what is the product's intended purpose, is it registered with MHRA and at what class (or CE marked under the EU rules and at what class), and who is the legal manufacturer. Then look the product up on MHRA's public registration database, which NHS England's guidance links to for exactly this check. Registration is a statement about the product, not a guarantee that it is safe in your clinic. Class I is a manufacturer's own declaration, so it carries less assurance than a notified-body certificate, and neither replaces your own clinical safety assessment.

Where does the data sit, and who can see it?

"UK data" can mean four different things: where recordings are stored, where transcripts and drafts are stored, where the AI model runs, and where support staff can view data from. Ask about each separately. Also ask whether audio is kept at all, for how long, and whether content is used to train models. A vendor that answers the storage question but not the processing question has answered half of it.

Jane, Tandem, Heidi and general assistants side by side

Read that table as of November 2026 and as a starting list of questions, not a verdict. Vendors change pages often, and the gaps in it (Tandem's UK regulatory route, Heidi's paid price, Jane's NHS safety documents) are things you ask the vendor, not things you assume.

Jane AI Scribe: for private practices that already use Jane

Jane is a practice-management platform, and its AI Scribe works inside it. According to Jane's UK page the scribe can record a live session or take dictation after the appointment, transcribe it, and draft a chart note from a template such as a SOAP note, optionally using the last ten chart entries or the latest intake form as context. The clinician reviews and edits before signing. Jane's UK FAQ says it is classified as a Class I medical device and registered with MHRA, that recordings and drafts are stored on servers in London, and that the data is not used to train AI models.

Jane also gives advice on patient consent that is worth copying: tell patients before the appointment that you use an AI-assisted tool for charting, let them ask questions or opt out, and use the intake form to capture consent. It adds that professional bodies may have their own guidance, which is a prompt to check yours.

Who it suits, with an illustrative example

A three-practitioner physiotherapy clinic already uses Jane for bookings and charts. The scribe is a switch inside the system they use, at £15 per practitioner a month for unlimited notes (billed in GBP), with a free tier of five notes a month each to try it first. For a clinic of three, that is about £45 a month, roughly one extra treatment slot in the whole month, so the question is time saved, not cost. The clinic runs a two-week pilot with one clinician, compares note-writing time and edit rates, and decides.

What to check: Jane's UK FAQ does not mention NHS clinical-safety documents (DCB0129 or a DTAC statement), so if you deliver NHS-funded care, ask Jane for them. Also ask which functions its Class I registration covers, since MHRA's July 2026 examples treat a plain transcript or summary as outside the device rules. Also check how the scribe handles a mid-session interruption or a second person speaking, and confirm that a patient who opts out gets the manual route without friction. What small clinics pay for AI scribes puts the price in context.

Tandem Health: a notified-body route, with data in Europe

Tandem describes an AI scribe that captures consultations and prepares structured notes for clinician review, with a coding assistant and clinical decision support alongside it. Its own site says those products are CE marked under the EU medical device regulation as Class IIa devices, externally assessed by a notified body; that all patient data is processed and stored in data centres within Europe; that conversation audio is deleted after transcription and never stored; and that data is never used to train AI. It lists ISO 27001 and ISO 13485 certification, UK Cyber Essentials Plus and compliance with NHS data security standards including the Data Security and Protection Toolkit (DSPT), and claims more than 10,000 care organisations across Europe. Those figures are Tandem's, not independently checked. Its pricing page lists a free plan and a Pro plan priced in euros and billed yearly, so check the price for your account.

Two things for a UK buyer. First, a CE-marked device and a UKCA-marked device are different routes. GOV.UK's guidance on regulating medical devices says general medical devices that comply with the EU MDR can be placed on the Great Britain market until 30 June 2030, so a CE mark is accepted for now but has an end date; ask Tandem about its UK registration and its plan for after the transition. Tandem also announced in April 2025 that it had teamed up with Accurx to bring its scribe to NHS staff, so some practices will meet it inside software they already use. Second, "Europe" is a region, not a country: ask whether UK patient data stays in the UK or is processed elsewhere in Europe, and confirm that your data-processing agreement and DPIA say the same.

Who it suits, with an illustrative example

A group GP practice wants the higher-assurance route a notified-body certificate implies, and it already uses a messaging platform that offers Tandem's scribe inside it. The practice manager writes four questions to the vendor: is audio ever stored, where is processing done, what is the UK regulatory route, and can you give us your clinical safety documentation for our DCB0160? She files the replies with the DPIA. If any answer is vague, the pilot does not start.

Heidi: NHS-scale adoption and the UK compliance pack

Heidi's UK page claims use by 70,000 clinicians, 15 NHS trusts and more than 1,200 GP practices, and says it powers what it calls the largest AI scribe procurement in NHS history. Its UK compliance page says data is hosted locally in the UK, that it is compliant with DTAC and that it follows the Data Security and Protection Toolkit standards. A post dated 1 September 2026 says Heidi has completed its DCB0129 clinical safety assessments, which can help customers prepare their own DCB0160 documents, and a post of 4 September 2026 adds that its DTAC assessment was a full internal one and that its DSPT submission is current. A free plan is advertised; I did not find a paid price on its UK pages.

On MHRA status, the same 4 September post says Heidi is currently registered as a Class I medical device under MHRA guidance for its summarisation function, and that it is working out what Class II certification would look like as it adds features. Read that against MHRA's July 2026 examples: a scribe that only summarises may not be a device at all, while added functions can move it up a class. Ask Heidi which functions the registration covers, which features you will switch on, and check the entry on MHRA's register yourself.

Who it suits, with an illustrative example

A hospital outpatient team wants a scribe that its trust's digital safety lead can approve. The lead reads the vendor's DCB0129 summary, writes the trust's DCB0160 safety case with the clinical safety officer, runs the DPIA using the template in NHS England's information governance guidance on ambient scribing, and signs off a six-week pilot with two clinicians. The vendor's paperwork saves effort, but the trust's own sign-off cannot be delegated.

General assistants for the non-patient half of the job

A large share of a clinician's admin involves no patient data: policy drafts, audit templates, rota wording, patient information leaflets, teaching material, reflective-practice prompts, correspondence templates with placeholders. For these, a general assistant on a business plan is a sensible tool. ChatGPT Plus and Claude Pro are $20 a month on individual plans (Claude Pro is $17 a month billed annually), the business plans start higher and add admin controls, and eligible Microsoft 365 business licences include Copilot Chat (Basic) at no extra cost. OpenAI says it does not use ChatGPT Business or Enterprise data for training by default; consumer plans offer an opt-out, so check the setting on any individual account.

Keep the line firm: no identifiable patient information in these tools unless your organisation has completed a DPIA and approved that tool for that use. Is it safe to put customer data into ChatGPT? explains why, and the answer is the same for patients with more at stake.

Three jobs these assistants do well, with examples

• A leaflet at the right reading level. Prompt: "Rewrite this leaflet on post-procedure care for a reading age of 11, keep every instruction, list anything you changed." Check every medical instruction against the original; AI can soften "do not" into "try not to".
• An audit template. Prompt: "Draft a one-page audit form for hand-hygiene checks in a small clinic, with columns for date, area, observer and action." No patient data, high reuse.
• A teaching outline. Prompt: "Outline a twenty-minute session for junior staff on safe handover, with three discussion questions." Review against your own protocols.

Dictation and transcription: the quiet alternative

If the regulatory or data questions slow an ambient scribe, speech-to-text dictation of your own summary, after the consultation, with no patient audio, is often easier to approve. It also avoids recording the patient, which changes the consent conversation. AI dictation software for doctors: dictation or ambient notes? compares the two approaches in detail.

A quick sum, illustrative: a clinician who sees 18 patients a day and saves four minutes of note time per patient gets back 72 minutes a day. Over 200 working days that is 240 hours. At even a modest value for clinical time, a £15-a-month scribe is cheap; the expensive items are the pilot, the paperwork and the risk of an unchecked error. Do not buy on the hours saved alone; measure how many edits clinicians really make.

A scoring sheet you can fill in for any scribe

Comparison tables date quickly; a scoring sheet does not. Score each candidate from 0 to 3 on each line, multiply by the weight and add up. Weights below suit a small private clinic; an NHS service should raise the weight on safety paperwork.

An illustrative scoring of three imaginary tools

To show the arithmetic only, three made-up candidates (not the products above) for a four-clinician clinic. Tool X scores 3 on regulatory, 2 on data location, 3 on retention, 1 on safety documents, 1 on fit, 2 on edit rate and 3 on cost: (3x3)+(2x3)+(3x2)+(1x2)+(1x2)+(2x2)+(3x1) = 9+6+6+2+2+4+3 = 32 out of a possible 48. Tool Y scores 2, 3, 2, 3, 3, 3, 2: 6+9+4+6+6+6+2 = 39. Tool Y wins despite a weaker regulatory answer because the sheet rewards documents and fit, so check that the weights match your real risks before you trust the total. If a tool scores 0 on either of the two triple-weight lines, drop it whatever its total.

A two-week pilot, with numbers you can collect

A pilot that ends with "everyone liked it" has told you nothing. Measure four things on one or two clinicians.

1. Minutes per note before and after. Time ten notes by hand in week one and ten with the tool in week two.
2. Edit rate. For twenty notes, count how many needed a change to a fact, medication, dose, side or plan, as opposed to wording.
3. Errors found in sampling. A second clinician reviews five notes against what they remember of the consultation, and logs omissions and inventions.
4. Patient opt-outs. How many declined, and whether the manual route worked.

Illustrative result: ten manual notes averaged 6.5 minutes; ten scribed notes averaged 3.0 minutes including review, saving 3.5 minutes. Of twenty notes, four needed factual corrections (a side confused, a dose misheard, a negative finding dropped, a follow-up date missing). That is a 20% factual-correction rate, which is the number that matters. It means every note needs a careful read, not a glance, and the time saving must be counted after that read.

Front desk and phones: a different regulatory question

Scribes get the attention, but front-desk AI (voice agents, chatbots, triage prompts) raises different issues. A bot that books appointments is a business tool; a bot that advises on symptoms starts to look like a clinical product and raises the device-status question again. Keep such tools to booking, directions and opening hours unless you have done the regulatory and safety work, and always give a route to a human. The clinic chatbot data and consent checklist is the place to begin.

The buying routine: eight steps from shortlist to sign-off

1. Define the job. Notes in consultations, letters after clinic, phone triage or all three. A tool good at one is not necessarily good at the others.
2. Ask the three regulatory questions in writing (intended purpose, MHRA or CE status and class, legal manufacturer) and save the replies.
3. Ask the four data questions (storage, processing, support access, audio retention and training use) and save the replies.
4. Collect the safety documents: the vendor's clinical safety case or DCB0129 material and its data-processing agreement, including sub-processors.
5. Complete your own DPIA and, for NHS-funded work, your own clinical safety assessment with a named clinical safety officer (DCB0160).
6. Pilot small. One or two clinicians for two to six weeks, with a patient-facing consent line and a manual route for anyone who declines.
7. Measure and sample. Time per note, edits, errors found in a sample of notes reviewed by a senior colleague.
8. Decide and record. Adopt, adjust or stop, with the date and the reason. Review at six months, since regulation and products both move.

For a fuller list of vendor questions, AI scribe security: audio, retention and vendor questions goes deeper than space allows here.

A realistic mistake: the trial that became the system

A clinic started a free scribe trial for one doctor. Within a month three more clinicians were using it, nobody had asked about processing location, and the DPIA had not been written. A patient asked where their recording went and nobody could answer. The fix cost a week: pause, ask the vendor the data questions, write the DPIA retrospectively and get consent wording in place. The lesson is that "free" and "just a trial" have no meaning for patient data; the paperwork starts before the first real consultation.

Consent and transparency: what patients should hear

Whatever tool you pick, patients should know before the consultation that an AI tool is being used, what it does, whether audio is recorded and kept, and that they can say no without consequences for their care. Jane's guidance suggests telling patients beforehand and using intake forms to record consent; the same pattern works for any tool. A short script at the start of the appointment helps: "I use a tool that turns my spoken notes into a draft, and I check it myself. Is that all right with you?" Rolling out an AI scribe without losing patient trust covers wording for posters, letters and reception.

How to keep this list honest after publication

Regulatory status and vendor features can change within weeks. Put a six-monthly reminder in your calendar to re-ask each vendor the regulatory and data questions, re-read NHS England's current guidance and MHRA's page on ambient voice technology, and check the register entries you relied on. If a vendor's answer changes, treat that as a trigger for a short review, not a footnote. Watch feature releases too: under MHRA's examples, a new "insights" button can change a product's device status even when nothing else about it has changed.

Questions UK clinicians ask before choosing an AI tool

Does an AI scribe have to be a registered medical device?

Not always. MHRA guidance first published on 29 July 2026 says a scribe solely intended to transcribe or summarise a clinical conversation for clinician review is not a medical device. Features such as suggested diagnoses, generated insights or autonomous orders give it a medical purpose, and then device rules apply. Ask each vendor for its intended purpose in writing.

Does 'data stored in London' mean the data never leaves the UK?

Not automatically. Check storage, processing and support access separately: where recordings sit, where the AI model runs, and where staff can view data. Jane states recordings and drafts are stored on servers in London; confirm processing and sub-processors in the vendor's data-processing documents.

Do I need a DPIA and DCB0160 for a private practice?

DCB0160 is the clinical safety standard for organisations deploying health IT, and NHS England's scribe guidance tells deployers to complete it with a DPIA and to check the supplier's DCB0129. Ask a clinical safety adviser whether it binds your private service. Do a DPIA either way: new technology processing health data is exactly what one is for.

Can I use ChatGPT or Claude for clinical notes?

Not with identifiable patient information unless your organisation has assessed and approved the tool for that use, with a DPIA and a business plan whose terms fit. For letters or policies with no patient data, general assistants are fine.

Further reads

• What Is an AI Scribe and How Does It Work in a Consultation? — How a scribe works in a real consultation.
• How a Small Clinic Can Use AI at the Front Desk Safely — Use AI at the front desk of a small clinic safely.
• Is an AI Scribe Worth It for a Small Private Clinic? — Whether a scribe pays for itself in a small private clinic.
• AI Scribe for Occupational Therapists: A Buyer's Checklist — A buyer's checklist written for another clinical profession.
• AI Scribe for Dentists: Will It Write Back to Your Software? — Will the scribe write back into your clinical system?
• AI Medical Transcription Services vs Human Transcribers for Small Clinics — When a human typist still beats AI transcription.
• AI Scribe for GPs: Costs, Free Tiers and the NHS Pilot Limit — Free and paid scribe options for GP practices.
• AI Platform for Doctors: One Suite or Separate Scribe and Phone Tools? — One suite or separate scribe and phone tools.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources, all read on 4 November 2026: MHRA guidance on ambient voice technology-enabled products (first published 29 July 2026) and GOV.UK guidance on regulating medical devices; NHS England guidance on AI-enabled ambient scribing products; Jane's AI Scribe for the UK FAQ and AI Scribe page; Tandem Health home, data and security, and pricing pages; Heidi's UK page, UK compliance page and posts of 1 and 4 September 2026; ChatGPT, Claude and Microsoft Learn plan pages. Vendor claims are the vendors' own; verify each before buying.