AI triage in general practice is widening the gap between access and care
A BMJ rapid response from a general practitioner takes aim at the rapid spread of AI-led triage, ambient note-taking, and total triage systems in primary care. The central complaint is not about technology alone. It is about what happens when software begins to stand between doctor and patient, deciding who is seen, how quickly, and on what terms.
The argument lands squarely in the consultation room. General practice depends on continuity, context, and an ability to interpret symptoms against a patient’s history, family situation, frailty, and lived experience. Algorithmic questionnaires can sort symptoms into categories, but they cannot weigh tone of voice, prior distrust, subtle deterioration, or the small clinical cues that often separate reassurance from escalation. In that sense, the concern is not merely operational. It is diagnostic, relational, and safety-related.
Total triage may move demand around the system, but it does not make demand disappear.
Algorithmic filtering cannot replace clinical judgement
The response describes total triage as a process in which patients are guided toward a probable outcome through algorithmic sorting. That model may look efficient on a dashboard, yet it can flatten the complexity of real primary care. A patient with chest discomfort, anxiety, diabetes, frailty, and an unstable home situation does not present as a single problem to be routed through a form. The same symptom can carry very different significance depending on who is describing it, what else is happening in their life, and how they have behaved clinically in the past.
Primary care medicine has always worked through accumulation of context. A long-term clinician learns what is ordinary for a patient, what is new, and what is quietly concerning. An automated questionnaire cannot carry that memory. It cannot sense the hesitation that sometimes reveals risk, or the mismatch between reported severity and underlying decline. Even when triage systems claim to improve throughput, they may do so by shifting work elsewhere rather than reducing it, which can leave patients with more contacts, more friction, and less certainty that anyone has truly understood the problem.
The response also points to a practical cost that is easy to miss in technology evangelism: when systems are designed to divert patients away from clinicians, they may create the appearance of access while making the experience more fragmented. A patient who cannot book ahead because the software forbids it is not the same as a patient who has genuinely obtained timely care. The difference is visible at the bedside, not on the access report.
Safety, frustration, and the loss of clinical nuance
One of the strongest parts of the critique comes from the perspective of frail and older patients. For them, a triage system that demands form completion, repeated clarification, or interaction with a voice interface can be more than irritating. It can be a barrier to care. The rapid response describes these systems as frustrating, unsatisfactory, and at times unsafe, particularly when they replace a human conversation that might have revealed deterioration, confusion, or caregiver concern.
That concern is clinically credible because frailty rarely behaves like a single-organ disease. It travels through mobility, cognition, continence, falls risk, hydration, medication tolerance, and social support. A rigid digital funnel is poorly suited to that reality. A clinician can hear when a patient’s story is incomplete or when a relative is trying to speak around a deteriorating condition. Software tends to take the question literally and the answer at face value, which is exactly the wrong posture for complex primary care.
The response is equally skeptical of the way some practices present AI not as an aid but as a cure for workload. There is a clear distinction between a tool that supports decision-making and a system that becomes a buffer between clinicians and patients. In the latter case, the professional role shrinks toward remote processing, while the core work of listening, examining, and interpreting is pushed back or broken apart.
Ambient note-taking may save time, but it also changes the consultation
The same concerns extend to ambient AI used for clinical documentation. Proponents often describe feeling more present with patients because they no longer have to type. The response is openly unsparing about that claim. Presence in a consultation is not created by delegating note-taking to software; it depends on attention, restraint, and clinical engagement. If the technology produces long, bland, or inaccurate notes, the supposed gain can quickly become another administrative burden.
In practical terms, the consultation is not a recording session. It is an exchange shaped by uncertainty, interruption, and multiple problems raised in the same encounter. Software built to capture linear information can struggle when the patient raises one issue, then another, then another, each requiring a different clinical frame. A note generated by automation still has to be checked, edited, and made safe. Any time saved during the visit may reappear afterward as quality-control work.
There is also a quieter professional issue here: the quality of the written record affects how medicine is practiced later. A human note can preserve nuance, soft signals, and the logic behind a plan. A sterile autogenerated note may flatten those details, leaving future clinicians with less context and less confidence. The consultation then becomes less of a relationship and more of a transaction. For general practice, where the record often carries the memory of the doctor-patient relationship, that is not a trivial change.
Primary care risks losing its human center
At the heart of the rapid response is a defense of general practice as a personal service grounded in local knowledge. Primary care does not exist solely to route patients efficiently. It exists to look after a population across time, through illness, recovery, uncertainty, and decline. When technology is deployed primarily to sort, filter, and shorten encounters, the service can drift away from that mandate.
The critique is not a blanket rejection of technology. It is a warning against accepting digital systems as the answer before asking whether they solve the right problem. In this framing, the real shortage is not software. It is clinicians, time, and the space to apply judgment properly. If an AI system increases barriers while claiming efficiency, it risks trading access for convenience and continuity for speed.
The final message is blunt and grounded in daily practice: patients still need a real human clinician who can listen, examine, interpret, and adapt care with empathy and nuance. Technology may support that work, but it should not displace it. The unanswered issue is whether primary care will continue to adopt systems that make the service look more streamlined while making the consultation less humane, less contextual, and less safe.