In the last lesson, I gave you the framework. Role, task, context, constraints. Now I want to show you what that looks like in practice — with three prompts that I use regularly. Each one is a real task that saves me real time.
And at the end, I am going to show you how to build a personal prompt library so you never have to write the same prompt twice.
Example 1: A practice protocol
The task: You need a protocol for your practice nurses to follow when conducting annual COPD reviews.
The prompt: “I am a GP partner in a UK NHS practice. Write a protocol for annual COPD reviews to be used by practice nurses. Base it on current NICE recommendations. Include which assessments to perform, which questionnaires to use, what spirometry results to check, when to step up treatment, when to refer to respiratory, and what to document. Format it as a numbered checklist with clear action points.”
What you get: A structured two-page protocol covering inhaler technique assessment, MRC dyspnoea scale, CAT score, exacerbation frequency, smoking status review, spirometry interpretation, medication step-up criteria, referral thresholds, and recall scheduling.
What you do next: You read every line. You check the spirometry thresholds against NICE. You adjust the referral criteria to match your local respiratory service. You change the recall interval to match your practice system. You remove one recommendation that the AI included about a medication pathway that does not apply in primary care.
Time to produce from scratch: about an hour. Time with AI assistance: about fifteen minutes, including verification. That is a genuine time saving on a real task.
Example 2: A patient information leaflet
The task: Several of your patients on methotrexate for rheumatoid arthritis do not seem to understand the monitoring requirements. You want a simple leaflet to give them.
The prompt: “Write a patient information leaflet about blood test monitoring for patients taking methotrexate for rheumatoid arthritis. Use plain English at a reading age of 11. Explain why regular blood tests are needed, what the tests check for, how often they need to be done, what might happen if a result is abnormal, and what symptoms to watch for between tests. Use NHS terminology and UK spelling. Keep it under 400 words.”
What you get: A clear, well-structured leaflet that explains liver function tests, full blood count monitoring, and the importance of regular checks. It covers what to do if they feel unwell between tests and when to contact the surgery.
What you do next: You check the monitoring frequency against your local shared care protocol, because this varies between areas. You adjust one section where the AI has been slightly vague about what symptoms warrant urgent contact. You add your practice phone number.
Notice something important: at no point did you enter any patient’s name, NHS number, or personal details. This is a generic leaflet for any patient on methotrexate. That is what keeps it in the green zone.
Example 3: A clinical question
The task: A colleague asks you about the latest guidance on prescribing direct oral anticoagulants in patients with atrial fibrillation and chronic kidney disease. You want a quick, structured overview before your afternoon clinic.
The prompt: “I am a UK GP. What are the current NICE recommendations for prescribing direct oral anticoagulants in patients with atrial fibrillation who also have chronic kidney disease? Include the relevant eGFR thresholds for dose adjustment and contraindications. Specify which DOACs are preferred and any differences between them in renal impairment.”
What you get: A structured summary covering apixaban, rivaroxaban, edoxaban, and dabigatran. Dose adjustment thresholds. The preferred agent in significant renal impairment. Contraindications at very low eGFR.
What you do next: You check the dose adjustments against the BNF. You verify the eGFR thresholds against the NICE guideline. And here is the critical point — when I did this exercise myself, the AI gave me an incorrect threshold for one of the medications. It was close, but close is not good enough for prescribing guidance.
That is why verification is not optional. It is the most important step. The AI got you 80% of the way there in thirty seconds. But the last 20% — the verification — is where your clinical expertise matters most.
Iterating on responses
Here is something that many people do not realise. You do not have to get it right in one prompt. AI is a conversation, not a single question.
If the first output is too long, say: “Make it shorter, maximum 300 words.” If it uses jargon, say: “Simplify the language — this is for a patient with no medical background.” If it misses a topic, say: “Add a section about when to call NHS 111.” If the tone is wrong, say: “Make this warmer and less clinical.”
Each instruction refines the output. You are editing through conversation rather than editing a document. Some people find this faster. Some prefer to edit the document directly. Either approach works.
Building your prompt library
Here is a practical suggestion that will save you significant time over the coming months.
Start a note on your phone, or a document on your desktop, or a folder in your email. Every time you write a prompt that produces good results, save it. Label it with the task type: protocol, patient leaflet, clinical question, meeting preparation, staff communication.
Over time, you will notice that 80% of your AI use falls into a handful of categories. Having a reliable prompt template for each one means you never start from scratch. You just adjust the specific details and go.
I have templates for practice protocols, patient information leaflets, clinical question summaries, and meeting agendas. When I need a new protocol, I do not write a new prompt. I take my protocol template and change the clinical topic. The role, the format, the constraints — all stay the same.
That is how you move from occasionally using AI when you remember, to having it as a reliable part of your workflow.
In the next lesson, we need to talk about the hardest skill of all — recognising when AI has got something wrong. Because even with the best prompt in the world, these tools will sometimes produce confident, well-written, entirely incorrect answers.
Key Takeaway
Save prompts that work and build a personal library organised by task type. You do not need to write a new prompt every time — reuse your templates and adjust the clinical topic. And always verify: the AI gets you 80% of the way, but the last 20% is where your expertise matters.