Training the whole team
on AI documentation

Consent to record starts before the clinician enters the room. For booked patients, the front desk is usually the first place anyone mentions that the practice uses an AI scribe, so reception staff need a short, accurate script and a chance to practise saying it aloud. Two minutes at a team meeting is enough to rehearse it; leaving staff to improvise is how patients end up hearing something garbled or alarming.

A workable script covers four things in plain language: the clinician uses a secure tool that helps write up the notes, it listens to the consult so the clinician can pay attention to the patient, the clinician checks and signs everything before it goes in the record, and the patient can say no with no change to their care. Something like: "Dr Chen uses a tool that helps write up her notes. It listens during the appointment so she can focus on you, and she checks everything before it goes in your file. Are you happy with that? It's fine to say no."

Reception should also know the two-sentence answers to the common follow-ups: who can hear the recording, where the information is stored, and what happens if they decline. Anything beyond that gets handed to the practice manager or the clinician. A printed patient explainer at the desk and a line in the appointment reminder both support the conversation. All of this applies to any AI scribe, whichever vendor the practice chooses.

In most Australian general practices, nurses run substantial clinics of their own: immunisations, wound care, chronic disease reviews, health assessments and care plan work. If nurses will document those encounters with the scribe, they need the same hands-on training as the doctors. That means starting and stopping a session, confirming consent at the start of the encounter, and understanding that the draft only becomes part of the record once it has been reviewed and signed under the practice's usual authorship rules.

Nurses are also heavy readers of other people's notes. Recalls, results follow-up and treatment room handovers all depend on the doctor's documentation, so nurses should be told what an AI-drafted note looks like in the first weeks and encouraged to flag anything that reads oddly. In practice, downstream readers catch a lot of the early drafting quirks, and a nurse who mentions a strange phrase at the morning huddle is doing exactly what the rollout needs.

Someone has to own the rollout, and in most practices that is the practice manager. The clinicians decide whether the tool is clinically useful; the practice manager builds the scaffolding that lets them find out safely. That scaffolding is mostly paperwork and scheduling, and it is worth doing properly before the first recorded consult.

The practical checklist looks like this:

• A short written policy covering consent, where recordings and drafts live, and the requirement that a clinician reviews and signs every note before it enters the record
• An updated privacy collection notice and patient explainer that mention the tool
• A training session for each role: clinicians, nurses, reception
• One named go-to person for questions during the first month
• A clear path for incidents: what happens if a patient complains, or a signed note turns out to contain an error
• A calendar date, about a month in, to review how it is going and decide what changes

Week one belongs to a small pilot group: one or two clinicians who volunteered, using the scribe in routine consults. Expect the early drafts to need editing. Clinicians are learning what to say aloud during the consult, and narrating examination findings and the plan in the room produces markedly better drafts as well as better-informed patients. A two-minute check-in at the morning huddle keeps small irritations from festering.

Week two widens the circle. The remaining interested clinicians come on, nurse-led clinics start using it, and reception begins running the consent script for all booked patients so the conversation becomes routine rather than exceptional. Keep the daily check-ins going; most of the useful feedback arrives in this week, once the novelty has worn off and the tool is meeting real complexity.

Measure a few things lightly. How long clinicians spend on notes after their last patient, how many drafts needed heavy correction, and how patients responded to the consent question. Handwritten tallies are fine. The point is to give the one-month review some substance beyond impressions.

Most rollouts that stall trace back to a small set of avoidable errors, so it is worth naming them before they happen:

• Skipping the reception briefing, so patients hear about recording for the first time inside the consult room and the clinician spends the opening minutes doing damage control
• Signing drafts without reading them. This is the fastest route to an error sitting in a clinical record with a clinician's name on it
• Training only the doctors and leaving nurses and reception to work things out from corridor conversations
• Going quiet in the room. The scribe drafts from what is said, so a clinician who examines in silence gets a thin note
• Switching it on for every consult on day one instead of building up from routine presentations
• Nobody owning feedback, so small fixable irritations accumulate until the tool is quietly abandoned

One behaviour decides whether an AI scribe is safe in a practice: every draft gets read, corrected and signed by a clinician before it becomes part of the record. Train this explicitly from the first session. Read every line, check medications and doses, check laterality, and check that recorded negatives were actually elicited. The clinician who signs the note is its author, and AHPRA's record-keeping expectations rest on that authorship regardless of what tool produced the first draft.

Build the habit into the timetable rather than the clinician's goodwill. Reviewing and signing before calling the next patient works for most consults, and it keeps the encounter fresh in memory when corrections are needed. Batching twenty unsigned drafts to the end of the day recreates the after-hours documentation pile the practice was trying to eliminate, and review done at the end of a long day catches fewer errors.

When reception, nurses and clinicians all understand that nothing enters the record unreviewed, the consent conversation gets easier to have honestly, and downstream readers know the notes they rely on carry the same authority they always did.