Writing With AI

This is the lesson where general-purpose AI earns its keep. Outbound practice writing is high-volume, rarely contains patient-identifiable data when done properly, and benefits enormously from a fast first draft.

The green zone tasks from Module 2 — protocols, guideline summaries, patient information, clinical questions — were the start. This lesson extends the same approach across everything your practice publishes or sends.

Every safe use of AI for practice writing follows the same shape.

You provide the facts, the audience, and the constraints. The AI produces a structured first draft. You correct, cut, and approve. The judgement — what is true, what is appropriate, what your practice is willing to say — never leaves the human.

The four-element prompt from Module 2 does the heavy lifting: Role (who the AI should write as), Task (what to produce), Context (the facts and audience), Constraints (length, reading level, tone, what to avoid). The difference between a useless draft and a near-final one is almost always in the context and constraints you supply.

Patient information is where AI drafting shines, because the hard part of patient writing is not knowing the medicine — it is pitching the language.

Ask for a reading age of eleven to twelve. Ask for short sentences and everyday words. Ask for a “when to seek help” section. Then check every clinical fact against the guideline yourself, because you sign it, not the AI.

Worked example. A practice changing its repeat prescription ordering process needs to tell patients. The prompt: “You are writing for an NHS GP practice. Draft a notice for patients explaining that from the first of September, repeat prescriptions can no longer be ordered by telephone, and should be ordered via the NHS App, online, or in writing. Reading age eleven to twelve. Warm but clear. Include what patients should do if they cannot use online services. Under two hundred words.”

The draft that comes back will need your practice’s details and a human read-through for tone. It will also take four minutes instead of forty.

Complaint responses are some of the hardest writing a practice does. The emotional temperature is high, the stakes are real, and the writing has to be accurate, empathetic, and legally sensible all at once.

AI can help — with the structure, not the substance. It can turn your bullet-point account of events into a well-organised, professional response. It can suggest a tone that acknowledges distress without admitting what should not be admitted. It can catch defensiveness you did not notice in your own draft.

But the boundaries matter more here than anywhere else in this lesson.

Individual referral letters contain patient data, so general AI is out for the letter itself — that is embedded-tool or human territory. But the scaffolding around referrals is fair game.

AI is excellent at producing referral templates: a standard structure for a two-week-wait referral, a checklist of what your local dermatology service wants included, a proforma for ADHD referral information. Build the template with AI once, then fill it with patient data inside your clinical system, where it belongs.

The same logic covers clinical protocols and standard operating procedures, as you saw in Module 2 — and letters to third parties that contain no patient data at all: a letter to the ICB about a service issue, a note to a care home about a process change, correspondence with suppliers.

A practical warning from experience: AI drafts have a recognisable flavour — slightly too smooth, slightly too enthusiastic, fond of certain phrases. If everything your practice publishes reads that way, patients notice, and the writing stops sounding like you.

The fix is editing. Cut the filler. Shorten the sentences. Put your practice’s plain way of saying things back in. The AI gives you the clay; the shaping is yours. A good rule: if you would be embarrassed to read a sentence aloud to a patient at reception, it does not go out.