Where AI Writing Helps a Clinic, and Where It Hurts
AI writing in a medical practice is one of those tools that looks harmless because the output is only words. In a clinic, words are the record β which is why the line between useful and dangerous needs drawing before anything is switched on.
Here is where we would draw it.
Where it helps
- Referral letters that follow a standard structure and pull from the patient record
- Recall and reminder messages in plain language
- Practice newsletters and health information that is reviewed before it goes out
- Internal summaries of long documents a clinician has already read
In each case the output is administrative, the clinician reviews it, and a mistake costs a phone call rather than a clinical outcome.
Where it hurts
Clinical notes written by a system that was not in the room. Advice written in the clinician’s voice. Anything produced for a patient to act on without a human reading it first.
The problem is not that the writing is bad. It is that plausible prose reads as authority, and a patient cannot tell the difference between a letter a doctor wrote and one a model assembled.
The consent question
If a patient’s information is being processed to generate a letter, the practice needs to know where that processing happens. That is a Privacy Act question, not a brand preference.
Our position is consistent across every clinical tool we deploy: inside the practice’s own environment, with the audit trail intact. That is why the AI Clinical Scribe is built that way rather than as a public web tool.
The reviewing clinician is the control
Every generated document enters the record as a draft. A named clinician signs it. If the practice cannot guarantee that step, it should not use the tool for anything patient-facing.
That single rule removes most of the risk and costs about twenty seconds per letter.
What good looks like
Letters that used to take twelve minutes take three. Recall messages go out on time instead of late. Clinicians spend their evening on the parts of the work that need a clinician.
Nothing patient-facing reaches a patient unread. That is the whole proposition, and it is worth being strict about. The referral side is covered by the Triage Parsing Agent.
What we build
We build automation and digital employees for Australian medical and allied health practices, and we will tell you which of our twenty specialised AI agents should stay away from patient-facing writing. Talk to us about your letters.
Start with one process
Tell us where the hours actually go and we will map it against what an agent can and cannot take. One process, measured before it is trusted. The first conversation is free and usually short.
Pingback: Why Re-Keying Data Costs More Than It Looks
3 November 2026