A GP Says AI Consultation Notes Are Making Her Job Harder, Not Easier

Key points

• An NHS out-of-hours GP says she finds meaningful errors "far more often" in AI-generated consultation notes than in those typed by colleagues.
• AI scribes, software that listens to a medical appointment and writes a summary automatically, have been flagged by an NHS watchdog for getting drug names and diagnoses wrong.
• The doctor says AI-written triage notes, assessments done before a patient sees a doctor, are consistently longer, repetitive and sometimes self-contradicting.
• Her own questioning of patients regularly produces a history that differs significantly from what the AI recorded.

Picture this. You are a doctor about to call a patient in. You read the notes. Your heart sinks.

That is the experience Dr Mary Gibbs describes every time she spots that an AI scribe has been used on a case. Dr Gibbs works as an out-of-hours GP in England, covering shifts when regular surgeries are closed. Before she speaks to a patient, she reads through a triage assessment, the initial phone check done by another clinician. When a human typed those notes, the summary is clean and useful. When AI wrote them, she says, the consultation that follows is "invariably long, with duplications, and sometimes contradictions."

So what exactly is going wrong?

The notes produced by AI scribes are unreliable enough that Dr Gibbs cannot trust them. She says the history she takes directly from patients turns out to be "significantly different" from what the AI recorded far more often than when a colleague wrote the notes.

AI scribes work by listening to a spoken consultation or phone call and then producing a written summary automatically. The promise is speed: less typing for busy clinicians, more time for patients. The reality, Dr Gibbs argues, is the opposite. She has to spend extra time cross-checking, re-asking questions and untangling contradictions that a human-written note would never have introduced.

This concern did not appear in isolation. The Guardian reported in late August that an NHS watchdog had already warned that AI scribes are getting drug names and diagnoses wrong, a finding that puts the problem squarely in the patient-safety column.

Should patients be worried?

Not panicked, but aware. A wrong drug name or a missed diagnosis detail in a summary is not automatically a patient-harming event. Good doctors check. Dr Gibbs checks. But checking takes time, and extra time in a stretched NHS system has a cost.

The practical point for anyone using out-of-hours or phone-first GP services: if something important was said during your initial call, say it again when you speak to the doctor. Do not assume the summary captured it correctly. Repetition in this case is a safety net, not an annoyance.

What happens next?

The wider rollout of AI scribes across the NHS is still moving forward, driven by genuine pressure to reduce admin load on clinicians. Voices like Dr Gibbs's are a necessary check on that momentum.

The question worth asking is not only whether AI saves time on paper, but whether it saves time in the room, with a real patient who may leave with the wrong information attached to their record. On that test, the early evidence is not encouraging.