Referral letters and AI patient privacy: what a GP referral gives away in ChatGPT or Claude.

The referral letter is the document a GP writes most often, and the one most often pasted into ChatGPT or Claude to be tightened, reordered or made easier for a specialist to read. It carries the patient’s identity and Medicare number, the presenting problem, the history, the medications, the results and often a line about how the patient is coping at home. This page marks every item in a routine referral that the Privacy Act, the NSW health records law and Good medical practice bear on, and shows how the same letter is drafted on Ermos from the patient record inside the practice.

The GP referral letter

Written after a consultation when a GP asks a specialist or allied health practitioner to assess or treat a patient. It is sent to the specialist and, in this example, copied to the patient.

An invented referral letter. The Medicare number and the provider number are placeholders. Hover or tap any highlighted item for the analysis; the full list is below.

What leaves the practice

1. 1
   Name, date of birth, address and Medicare number

   Four identifiers that tie everything below to one person. The Medicare number is an identifier assigned by an agency, and APP 9.2 limits when an organisation may use or disclose such an identifier. All of it is personal information, and the practice holds it as a health service provider. Privacy Act guide.

2. 2
   The presenting problem and the ECG finding

   A symptom history and an investigation result. Health information is sensitive information under the Privacy Act, with stricter rules on use and disclosure than ordinary personal information. Patient data guide.

3. 3
   Medical and family history

   Diagnoses, smoking history and a parent’s heart attack. The family detail is health information about someone else that also bears on the patient’s own risk. Patient data guide.

4. 4
   Medications and an allergy

   A list that reveals each condition being treated, with doses. Part of the record Good medical practice 10.5 expects a doctor to keep accurate and secure. Patient data guide.

5. 5
   Bereavement, living alone and a named daughter

   Social context a GP records because it affects care, naming a third party and describing a fear. Private detail that Good medical practice 4.4 asks doctors and their staff to hold in confidence. Patient data guide.

6. 6
   The request and a travel date

   What the GP wants done and by when, tied to the patient’s plans. It is the GP’s clinical judgement written for one patient, and the trip says when she will be away from home.

7. 7
   Pathology results

   Laboratory values against a named patient. Health information, and the kind of result a practice must also keep from leaving by accident. Patient data guide.

8. 8
   The referring GP and the practice

   Ties the letter to a named doctor and a provider number. A doctor’s name is on the public register, but the provider number is a billing identifier, and the letter links both to this patient’s record.

9. 9
   Information from My Health Record

   Something the GP learned by accessing the patient’s My Health Record. Section 59(2) of the My Health Records Act 2012 prohibits using or disclosing health information obtained through the My Health Record system unless Division 2 of Part 4 authorises it, where the person knows or is reckless that it does not, and section 61 authorises participants to use and disclose it to provide healthcare to the patient. Patient data guide.

10. 10
    The standard closing request

    Courtesy wording the practice uses on every referral. Nothing in it is about any patient.

What happens in Frontier Models (ChatGPT, Claude)

• The whole document is transmitted to the provider’s servers. On personal plans the user cannot choose where it is stored or processed, and for the major providers that has generally meant outside Australia, under that country’s legal process. Some business plans offer storage in Australia, but the provider still receives the content.
• On consumer tiers, prompts can be retained and used to train future models.
• The provider’s staff and systems can reach the content under the provider’s own policies.
• Once sent, the practice no longer controls it. Retention and deletion happen on the provider’s terms.

The same document on Ermos

Part of AI data leakage in healthcare practices. See how this works in practice: Referral letters on Ermos.

Common questions