Imagine: a patient has a chronic disease review booked for tomorrow morning. At 5pm today, an AI agent calls them. It confirms attendance. It asks whether medications have changed. It checks for new symptoms. It asks them to confirm their home blood pressure reading. It thanks them and says their GP will have a summary ready.
By the next morning, a structured brief is waiting in the clinical record: confirmed attendance, updated medications, new symptom summary, home BP reading, and the AI-generated patient summary drawn from the EHR. The clinician reads it in 90 seconds.
The individual components all exist: AI voice calling (QuantumLoop EMMA, InTouchNow), patient data capture (online consultation tools), and pre-charting (Navina, DeepScribe). They just have not been connected in UK general practice.
What the Pre-Visit Call Would Include
The call would be conversational but structured: confirm attendance, gather symptom or medication updates, collect patient-reported data (home BP, weight, blood glucose), identify new concerns the patient wants to raise, flag any urgency requiring rescheduling or escalation, and explain what to expect at the visit.
The output would be structured data — not a free-text transcript — that integrates with the clinician's pre-visit preparation.
The Consent Architecture
An AI agent calling a patient's personal phone, asking health questions, and recording responses requires explicit, informed consent under UK GDPR. This means a Data Protection Impact Assessment, clear patient information about data use, an easy opt-out mechanism, and transparency that the caller is AI.
Patients must know they are speaking to an AI. They must be able to opt out easily. And patients who decline must receive the same quality of care.
Clinical Safety
If the patient discloses something urgent during the pre-visit call — chest pain, suicidal ideation, acute confusion — the agent needs immediate escalation: transfer to a human clinician, automatic urgent flagging in the clinical system, or direction to 999/111. This logic must be clinically validated and regularly audited.
EMIS/SystmOne Integration
The pre-visit data packet needs to land in the clinical system in a reviewable format. Current EMIS and SystmOne APIs support third-party data input, but the integration requires collaboration with clinical system providers.
The Guideline Layer
When the pre-visit call identifies a concern — home BP exceeding 150/90, or new exertional chest pain — the system should flag it with guideline context. "This patient's home BP exceeds the NICE NG136 threshold for treatment review."
iatroX provides this layer. A pre-visit system using iatroX's NICE/CKS/BNF-grounded knowledge for clinical flagging would ensure every alert is guideline-current and citation-linked.
What Needs to Be True
Reliable voice AI handling diverse accents and languages. Robust consent and opt-out mechanisms. Tested clinical safety escalation. Native EMIS/SystmOne integration. DTAC compliance. Patient acceptance. And a guideline-grounded knowledge layer for clinical flagging.
None is impossible. Some are already proven in adjacent categories. The gap is in connecting them.
Conclusion
Pre-visit AI calling is the logical next step after AI reception and ambient scribing. It addresses a genuine workflow gap in UK primary care. The first vendor that builds this with proper governance, clinical system integration, and guideline grounding via platforms like iatroX will create a category.