AI notes for GP consultations: the tool that doesn't need a procurement process

The standard GP appointment in England is 10 minutes. Inside those 10 minutes, you are expected to take a history, examine where needed, reach a diagnosis or management decision, prescribe or refer, and document everything accurately in the patient record. The documentation is usually the last thing — rushed, done after the patient has left, from memory.

Research published in the British Medical Journal found that GPs spend approximately 40% of consultation time on administrative tasks, including documentation. In a 10-minute appointment, that is 4 minutes of typing. Across a 30-patient clinic session, it is 2 hours of documentation — most of it done in the gaps between appointments or at the end of a session when clinical recall is at its least reliable.

AI clinical documentation tools now exist that can change this. The question for GPs is which one is appropriate for clinical use — not just technically capable, but legally sound, practically accessible, and compatible with the real conditions of general practice in the UK.

The GP documentation problem in numbers

The data on administrative burden in general practice is stark:

• GPs in England spend an estimated 2–3 hours per day on documentation beyond clinical decision-making
• 42% of GPs report feeling burnt out, with administrative burden cited consistently as a primary driver
• Documentation that happens after the patient has left — as memory fades, as the next patient waits — is more likely to contain gaps, imprecision, or errors
• NICE's consultation guidance notes that clinician workload and documentation pressure are patient safety considerations, not merely efficiency concerns

These are not abstract statistics. They describe what happens during every GP clinic session, every working day, in every practice in England.

Why existing AI tools don't work in GP practice

Three cloud-based clinical AI tools have attracted attention in GP circles: Heidi Health, Nabla, and Freed. All three are capable products for the contexts they were designed for. All three have the same architectural problem for NHS GP use.

They send patient audio to cloud servers. Patient consultation audio is GDPR Article 9 special category health data. Processing it through a third-party cloud system requires a DPIA (Data Protection Impact Assessment), explicit lawful basis for the data transfer, patient consent disclosure, and — if the processor is a US company — attention to Standard Contractual Clauses for cross-border transfers under UK GDPR. NHS England's Caldicott Principles require that patient-identifiable data is protected at every stage. Cloud transmission is a stage.

This doesn't make cloud tools unusable in general practice — some GP partnerships have navigated the procurement pathway and DPIAs to use Heidi or similar tools legitimately. But the pathway is not simple, and it requires practice manager or ICB involvement that most individual GPs cannot drive alone.

They cost $119–150/month per clinician. For a GP partnership with six partners considering a clinical AI tool, the annual cost is $8,600–10,800 for a single product. This requires a practice-level procurement decision, budget approval, and an ROI case. It is not a tool a GP registrar can install on their phone this afternoon.

They require stable internet. GP surgeries generally have Wi-Fi, but broadband reliability varies — particularly in rural practices, older buildings, and during the peaks of a busy morning clinic. Tools that depend entirely on cloud connectivity have a failure mode in the same environments where GPs most need documentation support.

On-device processing: a different architecture for a different governance position

An AI clinical documentation tool that processes audio entirely on the clinician's device changes the governance picture significantly.

When patient consultation audio is processed on-device:

• No data leaves the device. There is no cloud transmission, no third-party data processor, no data transfer to assess.
• No DPIA is required for a third party. The device is the processing environment. The clinician is the data controller. No external party receives patient data.
• No Standard Contractual Clauses. There is no cross-border transfer because there is no transfer.
• The Caldicott position is clean. Patient-identifiable data is on the clinician's device — the same device where their email, calendar, and patient-related correspondence already live under existing acceptable use frameworks.

This is what the ICO describes as data protection by design and default: privacy built into how the product works, not managed through contractual controls after the fact.

For a GP who wants to use AI documentation today — not after a six-month practice-level procurement process — on-device processing is the legally defensible path that doesn't require external approval to walk.

The Kuulo workflow in a GP consultation

Here is what a consultation looks like with Kuulo:

Before the consultation. Open Kuulo on your iPhone. Select the GP Consultation or Clinical Consultation template. The template is pre-structured to capture presenting complaint, history of presenting complaint, relevant past medical history, examination findings, assessment, and management plan — the standard sections of a UK GP consultation note.

During the consultation. Tap record. Place the phone face-down on the desk or on the windowsill. The transcription runs entirely on your device's processor, using the Neural Engine in your iPhone's Apple Silicon chip. You conduct the consultation in your normal way. If you prefer, you explain to the patient that you're using a note-taking app on your own device, that nothing goes to any external system, and that it simply helps you produce accurate documentation.

After the consultation. The patient has left. You tap to generate the summary. Within 30–60 seconds, you have a structured clinical note: the patient's symptoms and history in their own words (subjective), your examination findings (objective), your working diagnosis or differential (assessment), and the management plan including prescriptions, referrals, and safety-net advice (plan). The note is on your screen.

Into the record. Review, correct any errors — the AI is good, not infallible — and copy the relevant sections into EMIS, SystmOne, or Vision. The full transcript remains on your device for reference.

Total documentation time per consultation: approximately 90 seconds, compared to the 4 minutes Nuffield Trust data suggests is the current average.

The patient consent conversation

Explaining recording to patients is simpler than it might seem, and easier when the tool is on-device.

A straightforward statement before the consultation begins: "I use a note-taking app on my phone to help me produce accurate consultation notes. It works entirely on my phone — nothing goes to any external system. Is that okay with you?"

Most patients respond positively to this framing. The absence of a cloud system, a company, or an account makes the explanation simple and honest. Patients who decline should have their decision respected immediately and without consequence — the consultation proceeds as normal without recording.

The GMC's guidance on patient confidentiality does not specifically address AI note-taking, but its principles are clear: patients should understand how information about them will be used. On-device processing simplifies that explanation considerably.

A note for GP registrars and locums

The procurement pathway for enterprise clinical AI tools (Heidi, Nabla) typically runs through a GP partnership or practice. GP registrars rotating between practices and locum GPs who work across multiple sites don't have a consistent practice-level relationship through which to procure a tool.

Kuulo is a personal device app. It follows the clinician across every practice, every rotation, every locum session. The SOAP note template generates the same structured output whether you're in a rural practice in Devon or a city centre surgery in Leeds. There is no IT provisioning requirement. The only requirement is the iPhone in your pocket.

For GP registrars specifically: the documentation you produce on placement is a clinical learning record. A structured, AI-generated consultation note reviewed and corrected after each appointment is better portfolio evidence than a retrospective summary written at the end of the day. The habit of accurate, timely documentation is one of the most transferable clinical skills. Kuulo makes it practically achievable under the time constraints of GP training.

The comparison with enterprise clinical scribes

The 4 minutes per consultation that AI documentation can recover is not a small number. Across a 30-patient clinic, it is 2 hours. Across a working week, it is a half-day. The GP who reclaims that time is not more efficient — they are less likely to be running 20 minutes behind by 11am, less likely to rush through the last four patients, less likely to document retrospectively from a memory that has been overwritten by the rest of the morning.

The technology to do this exists, runs on the device in your pocket, costs nothing to start, and processes patient data in the only way that is architecturally sound for NHS general practice. The only remaining question is whether you'll try it on your next clinic session.