Walk into a growing number of Australian consulting rooms this year and you may notice something new sitting between you and your doctor: a phone or laptop, quietly listening. The device is running an “AI scribe”, software that records the conversation, transcribes it and drafts the clinical note, freeing the doctor from typing while you talk. The technology is being pitched as an antidote to burnout and paperwork. But a fresh article in the Medical Journal of Australia puts the question that has largely been skipped in the rush to adopt: how do patients actually feel about it?
The context: a quiet revolution in the consulting room
AI scribes, also called ambient documentation tools, have moved from novelty to near-mainstream in Australian general practice with remarkable speed. Products from home-grown startups such as Melbourne’s Heidi Health and Lyrebird Health, alongside international offerings, promise to hand clinicians back the hours they lose to record-keeping. For a workforce battling burnout and shrinking bulk-billing margins, the appeal is obvious. Surveys of doctors have consistently found that administrative load is one of the biggest drivers of dissatisfaction, and anything that lets a GP look at their patient instead of a screen has a ready market.
Yet the technology sits on sensitive ground. A medical consultation is one of the most private conversations most people will ever have. Recording it, even to produce a better note, introduces a third party into the room: the software vendor, its servers and, in some cases, the large language models that power the transcription and summarisation. The MJA article’s core provocation is that clinicians and vendors have raced ahead on efficiency while doing comparatively little to ask whether patients are comfortable being recorded at all.
The news: patient trust is the missing piece
The central argument canvassed by the MJA is that patient acceptance cannot be assumed. Adoption to date has been driven largely by clinicians and practice managers weighing time savings and note quality. Patients, by contrast, are often only lightly informed, sometimes with a verbal heads-up, sometimes with a sign in the waiting room, sometimes not at all. The piece frames this as a consent and communication gap that could undermine trust if it is not addressed deliberately.
The reasons patients might hesitate are not hard to guess. There are worries about where the recording goes, how long it is kept, whether it is used to train commercial AI models, and who could access it in the event of a data breach. There is also a subtler concern: whether people will censor themselves, holding back an embarrassing symptom or a mental-health disclosure, if they know a machine is capturing every word. Against that, many patients may welcome a doctor who is more present and less distracted, and who produces a more accurate record of what was actually said.
Two viewpoints: efficiency versus autonomy
Clinicians and the vendors backing them tend to emphasise the upside. In their telling, AI scribes reduce errors of omission, capture nuance that a rushed doctor might forget, and let practitioners spend their limited consultation minutes on the human parts of medicine. Advocates argue that, handled well, with a clear explanation and a genuine chance to opt out, most patients accept the tools once they understand the benefit to their own care.
The counter-view, which the MJA gives real weight to, is that convenience for the clinician is not the same as informed consent from the patient. Privacy and consumer-health advocates have warned that “opt-out by default” or buried disclosures fall short of the standard a recorded medical conversation demands. The Royal Australian College of General Practitioners has already published guidance urging members to obtain informed consent, understand where data is stored, and check whether a product meets Australian privacy obligations. Medical defence organisations have echoed the caution, noting that the doctor, not the software company, remains legally responsible for the accuracy of the note the AI drafts.
The Australian stakes
This is a distinctly Australian problem as much as a global one. Health information is among the most tightly protected categories under the Privacy Act, and the Office of the Australian Information Commissioner has made clear it expects organisations deploying AI to handle personal data lawfully and transparently. Many AI scribe vendors process audio offshore or rely on foreign-owned cloud infrastructure, raising questions about data sovereignty that matter more for medical records than almost any other kind of information.
There is also a regulatory grey zone. The Therapeutic Goods Administration regulates software that functions as a medical device, but a tool that merely transcribes and summarises a conversation may fall outside that net even as it shapes the clinical record. That leaves a patchwork in which individual practices, rather than a national framework, effectively decide how consent is sought and how data is safeguarded. For a health system already grappling with My Health Record adoption and public wariness after high-profile data breaches at Medibank and elsewhere, trust is a finite resource. Mishandling AI scribes could burn it quickly.
The equity dimension matters too. Rural and outer-suburban practices, chronically short-staffed, may lean hardest on AI scribes to keep appointment lists moving, meaning the patients with the least choice of GP could be the most exposed to the technology, whether or not they are comfortable with it. Culturally and linguistically diverse patients, and those with lower digital literacy, may find “informed consent” hardest to give meaningfully when the explanation is a single line on a waiting-room poster.
What’s next
The MJA piece reads less as a verdict than as a prompt: before AI scribes become the invisible default in Australian medicine, the profession needs to design consent, transparency and data-handling with patients in the room rather than as an afterthought. Practically, that could mean standardised plain-language explanations, a genuine and easy opt-out, clear rules on data retention and model training, and a preference for tools that keep Australian health data onshore.
Expect the colleges and regulators to keep tightening their guidance as adoption climbs, and expect patient-experience research, exactly the kind the MJA is calling for, to catch up to the technology it is trying to evaluate. The efficiency case for AI scribes is largely settled. The trust case is not, and that is the one that will decide whether the microphone on the desk becomes a routine part of good care or a quiet source of unease.
Sources: Medical Journal of Australia
