What Is an AI Medical Scribe?

An ambient AI scribe runs in the background during a consult, transcribes the conversation, and returns a structured draft note in whatever format the clinician works in. The clinician reviews, edits, and signs off, and the finalised note becomes part of the record in the same way it would if it had been typed directly. Audio is not retained by Lyrebird after the consult ends, and the clinician remains the author of the record.

That is the product in one paragraph. The relevant detail, and where the differences between vendors live, is in how each stage of that loop is implemented and what the product is permitted to do under Australian regulation.

What happens during a consult

The capture stage records audio through a browser or desktop app while the consult proceeds as normal. With Lyrebird, audio is processed on Australian servers and deleted at the end of the consult, so no recording is retained by the vendor or accessible to the practice. See the compliance page for the full data handling position.

The draft stage is where the model does its work, reading the transcript, filtering out non-clinical content, and organising the remainder into a clinical format, whether that is SOAP, an issues-based note, a specialist letter, a discharge summary, or a template the clinician has built from their own notes.

The write-back stage differs significantly between products. For Bp Premier, Lyrebird writes the note directly into the record with a single click and places structured observations such as blood pressure, temperature, and weight into the correct Bp Premier fields without manual entry. For other EMRs, most scribes currently rely on copy and paste, and integration roadmaps vary. See the Best Practice integration and all supported integrations for more detail.

End-to-end, the loop typically completes in under a minute.

What an AI scribe is not

A scribe drafts documentation from what was said in the consult, and that is the defined scope of its function. It does not reason about the presentation, it does not generate a differential that was not discussed, and it does not propose a medication or investigation that was not raised by the clinician.

This distinction was informal until August 2025, when the TGA issued guidance on digital scribes that codified it. A scribe that transcribes and structures clinical conversation is not a medical device under the Therapeutic Goods Act 1989. A scribe that analyses or interprets the conversation and generates a diagnosis, differential diagnosis, or treatment recommendation not explicitly stated by the clinician is a medical device, and must be included on the ARTG to be supplied in Australia. The practical consequence is that documentation scribes and clinical decision support tools are now distinct product categories, with different regulatory obligations for the vendor and different disclosure considerations for the practice.

Lyrebird is a clinical documentation tool and does not generate clinical suggestions the clinician did not make.

Where the differences between vendors matter

A few points separate products in ways that are visible in day-to-day use and in regulatory posture.

Data residency. Processing and storing patient data in Australia avoids the offshore transfer disclosure that APP 8 otherwise requires. Lyrebird processes and stores all data on Australian servers. Several internationally-built scribes route part of the pipeline through the US or Europe, which does not preclude use in Australia but does shift the disclosure obligation back to the clinician.

EMR integration depth. The difference between a native write-back and copy-and-paste accumulates over the working year. Lyrebird has the only AI scribe integration with Bp Premier that Best Practice Software describes as "exclusive," developed in close partnership with Best Practice. The integration includes single sign-on, structured write-back of clinical observations as atomised values into the Bp Premier observations section, and persistent consent recording. See the Best Practice integration page for specifics.

Training data policy. Some scribes, particularly those offered free of charge, use patient conversations to train their models. Lyrebird does not use customer data for model training. Confirming this point in a vendor's contract is reasonable practice.

Template adaptability. The scribe's output becomes closer to the clinician's own style when it has examples of existing notes to learn from. The Gold Coast Hospital and Health Service evaluation identified template work as one of the factors that separated clinicians reaching significant time savings from those whose gains plateaued early. See the implementation lesson on optimising for reviewability for detail.

Audit trail. Per-consult, timestamped consent logging provides a record that can be produced if an MDO query or audit arises. Lyrebird logs consent status at the consult level, with export available.

What the evidence shows

Effect size varies with baseline. Clinicians whose documentation routinely extends beyond clinic hours have reported larger time savings than those who complete notes contemporaneously, and note-quality improvements have been more pronounced where baseline documentation was sparse. These variations are consistent across the published evaluations, and meaningful interpretation of headline figures requires context. See the GCHHS lesson on interpreting impact relative to baseline documentation quality for more on this.

The most substantial peer-reviewed evaluation in Australian practice is the Gold Coast Hospital and Health Service deployment, covering more than 100 clinicians across 21 specialties. The published findings included an 80% reduction in documentation time, an 88% improvement in note quality on independent review, and 84% of staff reporting a positive impact on efficiency.

Alder Hey Children's NHS Foundation Trust reported a 78% reduction in after-hours documentation and a 15% increase in consultations for specialists using Lyrebird. Dr Nuwan Athauda, an individual GP, reported recovering approximately an hour per day, which translated into two to three additional patients seen. Both sit within the range observed across Lyrebird deployments, though individual results depend on specialty, baseline, and the degree to which templates have been adapted.

Where the evidence shows it helps most

The GCHHS evaluation found that benefit was not uniform across specialties or consult types. Time savings were largest for clinicians whose baseline documentation extended beyond clinic hours, and note-quality improvements were most pronounced where baseline documentation was sparse. Procedural consults and very brief appointments showed smaller effects, which is consistent with scribe output being proportional to the amount of clinical conversation captured.

The implication for a practice evaluating a rollout is that the case for adoption varies by clinician and consult type, and the realistic expectation is a meaningful improvement in aggregate rather than a uniform one across every appointment. The GCHHS lesson on value being contextual covers the specifics.

Next steps

To evaluate Lyrebird directly, book a demo. For Bp Premier users, Lyrebird Free is available for all Bp Premier customers without a separate contract.