The Green Zone

Three months ago, you would have spent an evening writing it from scratch. Today, you are going to have a first draft in three minutes. And you are going to do it completely safely.

In the last two lessons, we covered the data protection rules and why they exist. Now I want to show you what you can actually do. Four green zone tasks with worked examples. No patient data involved. No governance concerns. Just genuine time savings.

Here is how I would do it. I open ChatGPT and type the following:

“I am a GP partner in a UK NHS practice. Write a protocol for managing annual chronic kidney disease reviews. Base it on current NICE recommendations. Include which blood tests to order at each review, what to check clinically, when to refer to nephrology, and how to structure the recall system. Format it as a numbered checklist that a practice nurse could follow.”

Three minutes later, I have a two-page protocol. It covers eGFR monitoring, urine albumin-to-creatinine ratio (ACR) testing, blood pressure targets, medication review, and referral thresholds.

Is it perfect? No. It never is. When I check it against NICE, I find that one of the blood pressure targets needs adjusting and the referral threshold is slightly off. But those two corrections take me five minutes. Writing the entire protocol from scratch would have taken an hour.

NICE guideline NG203 on chronic kidney disease is 78 pages long. You do not have time to read 78 pages.

So you ask AI: “I am a UK GP. What are the key recommendations from the NICE guideline on chronic kidney disease that are most relevant to primary care? Focus on monitoring frequency, referral criteria, and blood pressure targets.”

The AI gives you a structured summary. Monitoring intervals based on CKD stage. The ACR thresholds that trigger referral. The blood pressure targets for patients with and without diabetes. The recommended medications at each stage.

But here is the important part. You are not pasting the guideline text into AI. You are asking AI about publicly available recommendations. The AI draws on its training data to answer your question. And you verify every key point against the actual guideline before you rely on it.

This is the difference between using AI as a shortcut to understanding — which is safe — and feeding copyrighted content into AI for processing — which is not.

You have a diabetes clinic next week. You want a one-page leaflet about managing sick days for patients on SGLT2 inhibitors, because you have noticed that several patients do not know the sick day rules.

You type: “Write a patient information leaflet about sick day rules for patients taking SGLT2 inhibitors like dapagliflozin or empagliflozin. Use plain English at a reading age of 11. Include when to stop the medication, what symptoms to watch for, when to contact their GP, and when to go to A&E. Use NHS terminology and UK spelling.”

The AI produces a clear, well-structured leaflet. It covers the key points: stop the medication if you are vomiting, have diarrhoea, or have a high temperature. Watch for signs of diabetic ketoacidosis. Contact your GP if you are unsure. Go to A&E if you feel very unwell or are breathing unusually fast.

You check the clinical content against the current guidance. You adjust one sentence where the AI has been slightly imprecise about when to restart the medication. You print it and hand it out next week.

Total time invested? About ten minutes, including verification. Time to write the same leaflet from scratch? At least thirty minutes, probably more.

And notice what was not in that prompt. No patient names. No NHS numbers. No identifiable information. A generic leaflet for a generic clinical situation.

A patient mentioned that their friend was prescribed a new weight loss injection and asked whether it might help them. You know about semaglutide and tirzepatide, but you are not sure about the latest NICE guidance on eligibility criteria for NHS prescribing.

You type: “I am a UK GP. What are the current NICE recommendations for prescribing GLP-1 receptor agonists for weight management in adults? Include the BMI thresholds, any comorbidity requirements, and whether these are available in primary care or specialist services only.”

The AI gives you a structured overview. BMI thresholds, the requirement for a specialist weight management service referral, the comorbidity criteria, and the current availability through the NHS.

You check the key points against NICE and the local formulary. You now have a clear answer for your patient, and the whole process took less than five minutes.

This is AI at its most useful. Not replacing your clinical knowledge, but saving you the time it takes to find and synthesise information you could have found yourself.

All four tasks share two features.

First, no patient-identifiable data was involved at any point. The inputs were your clinical questions and publicly available guidelines. The outputs were generic documents and summaries.

Second, you reviewed the output before using it. The AI produced a first draft. You verified it against reliable sources. You made corrections where needed. You remained the clinician.

In the next lesson, we need to talk about the tasks that are less clear-cut. The amber and red zones — where the stakes are higher and the boundaries matter more.