· Registered Nurse
I gave it a fair go. For two weeks I let an AI assistant draft the appointment reminders and post-procedure check-ins I send patients by text. Faster, yes. Also wrong in a way that took me a few days to actually spot. So when someone at our clinic asks me — should nurses use AI to write patient messages — I’ve got a short answer. I draft patient SMS reminders by hand, not with AI. The rest of this is why, and the six-template trick that keeps that quick on a packed shift.
The 90 seconds it takes me to write one reminder by hand
Here is a real one from last Tuesday’s list. A patient booked for a longer skin-check appointment, anxious type, asks a lot of questions. I wrote: “Hi Sara, your skin check with the practice nurse is Thurs 10am. Pop in 5 mins early. Reply STOP to opt out.” Ninety seconds, sent.
The next patient on the list got different words. Post-procedure check-in, needs reassurance, not instructions. The opening line changed. Three patients, three opening lines, all in my own voice. That voice is the whole point.
When I trialled the AI draft, the wording drifted. Same clinical meaning every time, just paraphrased differently. Patients who’d had texts from me for years suddenly got a message that didn’t sound like me — and a few of them replied, confused about a step the AI had quietly reworded. That’s the thing: for a message someone reads on their phone in five seconds, sounding consistent beats being drafted fast.
What patient research actually says about AI-written messages
I don’t want this to be just my gut. The clearest study I’ve found is on patient preferences for AI-drafted responses to electronic messages. It reports that satisfaction drops once patients are told the message was AI-authored — a measurable dip versus believing a human wrote it (Ethics in Patient Preferences for AI-Drafted Responses, PMC11897835).
The same study makes the honest counterpoint. Over three-quarters of patients stayed satisfied regardless of who wrote the message, and the authors argue the dip shouldn’t override the duty to be transparent. Both things are true at once. The trust cost is small, but it’s real, and it doesn’t land evenly — it lands hardest on the patient who already feels uneasy about their care. That’s the patient I write for.
The disclosure problem I can’t get past
Say I keep using AI. Then I’m stuck choosing between two bad options. I can disclose it — and the research says satisfaction softens, which for an already-worried patient is the worst possible moment to plant a doubt. Or I don’t disclose it, and I’m sending words I didn’t write under my own name and registration. The second one sits worse with me, honestly.
And there’s an accountability question sitting underneath all of it. If an AI-reworded reminder tells a post-op patient the wrong day to stop a medication, I’m the registered nurse whose name is on it. AHPRA doesn’t have a column for “the model phrased it.” So the line stays mine, because the responsibility is mine — and that isn’t a workflow preference, it’s where my registration lives.
Consent, privacy and the wording AI gets wrong
Most of the SMS-and-AI writing online is framed around US HIPAA. I work under the Australian Privacy Act 1988, and the framing is different. The OAIC’s guidance on communications with patients is clear that consent — often implied for routine reminders, explicit when content strays toward marketing — has to be obtained and documented, and that a patient handing over a mobile number is not a blanket yes to everything (OAIC: Communications with patients).
Here’s where generic AI phrasing bit me. Indemnity guidance in Australia is blunt: an SMS should carry minimal clinical detail, a name and a callback, nothing a stranger reading over a shoulder shouldn’t see (Avant: Recommendations when communicating by telephone or SMS). The AI, trying to be helpful, padded my reminders out with the full procedure name and prep instructions. Tidy prose — and far more clinical detail in the body of a text than I’d ever put there by hand. The kind of brevity that protects a patient’s privacy just isn’t the model’s instinct. It’s mine.
Where I do let AI help (and where I won’t)
I’m not anti-tool. I let AI near the scaffolding all day. Spell-check, tightening a clunky internal email, suggesting a template structure before I rewrite it in my own words — fine. The pattern I learned the hard way mirrors a lesson from my dev side, the first time an AI coding tool lied to me about an API: the model is great at draft shape and unreliable on the specific words that carry consequence.
The actual patient-facing line stays mine. Non-negotiable. A reminder is a tiny artefact of the nurse-patient relationship, and that relationship is the thing I’m protecting, not the ninety seconds. I’ve written before about how a tool will confidently produce plausible, wrong specifics — my content agent once invented a fake nurse persona, and a patient’s text is the last place I want plausible-but-wrong.
How I keep hand-drafting fast on a busy shift
The fear is that hand-writing is slow. It isn’t, because I don’t start from blank every time. I keep a six-template phrase bank: appointment reminder, longer-appointment reminder, post-procedure check-in, results-ready callback, missed-appointment nudge, and pre-appointment prep. I drop a name in, tweak one line for the person, send.
When I switched back to that hand library after the AI fortnight, the reply-confusion rate dropped immediately — the messages sounded like me again, so people stopped second-guessing them. And some reminders don’t belong in a text at all. A frail patient, a sensitive result, a complex prep: that’s a phone call. Knowing when to not send an SMS is part of the skill, and no draft tool prompts you to pick up the phone. If you’re building your own bank, sanity-check the wording against your practice policy and your supervisor first.
My honest opinion
Speed is the wrong thing to optimise here. The pitch for AI-written patient messages is minutes saved, and the cost is a slow erosion of the one thing that makes a reminder land: it sounds like the person who’s been caring for you. So I’ll spend the ninety seconds. Internal admin, sure — point the AI at it all day. The words a patient reads in five seconds on a Thursday morning are a different thing, and for those the AI-versus-human question, for me, isn’t close. Do patients trust an AI-written appointment reminder a bit less? The data says yes. And trust is the whole job, so I’m not spending it to save a minute and a half.
TL;DR / Key Takeaways
- I draft patient SMS reminders by hand because voice consistency beats draft speed; when I trialled AI, patients replied confused about reworded steps.
- Research shows patient satisfaction drops when AI authorship is disclosed, leaving a no-win disclosure choice (PMC11897835).
- Accountability stays with the registered nurse, not the model — my name and AHPRA registration are on every message.
- Generic AI phrasing over-shared clinical detail; AU privacy and indemnity guidance wants minimal detail in an SMS.
- A six-template hand library keeps it fast (~90 seconds) and dropped my reply-confusion rate immediately.
Sources
- Ethics in Patient Preferences for Artificial Intelligence–Drafted Responses to Electronic Messages — PMC11897835
- Office of the Australian Information Commissioner (OAIC) — Communications with patients
- Avant — Recommendations when communicating with patients by telephone or SMS
Written by Stone, AHPRA-registered Registered Nurse, Sydney primary care. Reflects my own practice and is not medical or clinical-governance advice — check your own service’s policy and your supervisor before changing how you message patients. Last reviewed: 6 June 2026.
Fact-checked 2026-06-06. Last reviewed 2026-06-06.