This is not a prompt list. It is a school: every lesson ends where a real tool begins
— the interaction checker, the drug lookup, pregnancy and breastfeeding — and every
exercise is marked against a rubric written by a pharmacist, not against whatever a
model happens to think that day.
First things first: no lesson here will ask you to upload a real patient's
report, blood results or medication list. Everything you practise on is synthetic and ships
with the site. Teaching otherwise would put you in legal trouble — and us for recommending
it — so that is exactly what lesson one is about.
The four levels
Always free
L1
Foundations and safety
- What generative AI is, and what it is not
- Patient data, GDPR and confidentiality
- Hallucinations and how to verify
- Bias, automation bias and prompt injection
- De-identify before you ask, in photo and voice too
Account
L2
Everyday practice
- Anatomy of a good prompt
- Everyday emails and messages
- Summarising a research paper
- Pulling data out of a PDF
- Reviewing what you already wrote
Account
L3
Clinical AI
- Reading a medical report with AI
- Structured medication review
- The interaction checker: dataset or AI
- Triage support, when to refer
- Exercises on synthetic cases
Premium
L4
Owners: business and automation
- The map of a pharmacy's 17 areas
- GitHub, Vercel, Supabase and Claude Code
- Analytics and Search Console
- A chatbot and an app with AI Studio
- Building AI agents for your business
Level 1 — Foundations and safety Always free
Thirteen lessons, in this order. The first ones are the what and the
concrete risks — text, image, voice, injection; the last ones, the
judgement to decide on your own.
-
1 · Patient data and privacy
What you may type, what you may not, and how to de-identify in ten seconds. Required before everything else.
-
2 · What happens to what you type: memory and history
A text that identifies nobody is not automatically harmless to send. Where it goes, how long it is kept, and who else can read it.
-
3 · What generative AI is, and what it is not
A model does not look the answer up: it writes it. Everything else follows, with a classifier for which tasks you can hand it.
-
4 · Hallucinations and how to check an answer
An invented reference does not look like a mistake: it looks like a citation. With a detector for sources that lead nowhere.
-
5 · Train your judgement: catch the error
You will not write a single prompt here: the AI has already answered and your job is to say what is wrong. Three cases, marked against a rubric.
-
6 · Bias: who the answer leaves out
Bias does not show up in the wrong answer: it shows up in the one that looks right for everyone and is only right for some. Four counter situations.
-
7 · Automation bias: when you trust AI more than you should
Not a bias of the model, it's yours: the more confident an answer sounds, the less it gets checked. And it always sounds equally confident.
-
8 · Images and voice: the multimodal risks
A photo identifies by what is around it, not just by what you ask. The risk the text lesson does not cover.
-
9 · Prompt injection: when the text you paste gives the orders
A model cannot tell your instruction apart from the text you paste it. Where this actually shows up in a pharmacy, and how to spot it.
-
10 · Professional responsibility: the sign-off is still yours
AI is not a source and is not cited as one. What you document, what you delegate, and what does not stop being yours because a machine wrote it.
-
11 · ChatGPT, Gemini or Claude: what actually changes
Three questions that really do separate them and five things that do not change by changing model. No comparison tables that expire in a month.
-
12 · When the patient brings an AI answer
It happens more and more. How to separate what it gets right from what it does not, without sounding like you're competing with what they read.
-
13 · When NOT to use AI: the risk traffic light
Green, amber and red, with the reasoning written down: what happens if the answer is wrong, and who pays for it. The close of Level 1.
Level 2 — Everyday practice Account
The seven jobs a pharmacy actually hands to an AI, with where each one
fails.
-
1 · Anatomy of a good prompt
Five pieces, and the one almost nobody writes is the one that changes the answer most. With a composer that assembles it and tells you what you left out.
-
2 · Writing everyday emails and messages
Not the difficult email you've been avoiding: the ordinary one, every day. The four details that turn it into a useful draft.
-
3 · Summarising a research paper
What you hand over is almost always the abstract, not the paper. What is not in there, and why the 50 % in the headline is usually a 1 %.
-
4 · Extracting data from a PDF
A text PDF and a photo of a piece of paper are not read the same way. In the second the model fills in the blur, and a filled-in digit looks like a read one.
-
5 · Using AI to review what you already wrote
Not to write for you: to tell you what reads badly. A different use, with completely different reliability.
-
6 · Patient-facing material
Being understood is the easy half. The hard half is what you may claim — and the model will not stop you there.
-
7 · Preparing team training
The raw material is not a topic: it is what actually goes wrong at your counter. Twenty minutes, one decision and the exact sentence you say.
Level 3 — Clinical AI Account
What you do with a patient in front of you, in the order you do it — and an
exercise at the end to check it stuck.
-
1 · Reading a medical report with AI: the 8-step method
Extract, structure, flag the uncertain, separate fact from interpretation, cross-check, ask, summarise and review. Every step with where it fails.
-
2 · Medication review
Six steps, and AI helps with two, gets in the way of one and is no use in three. The hard step is not the analysis: it is the list.
-
3 · AI alongside the interaction checker: what it adds and what it does not
The site's checker is not just AI: it has a curated dataset, and AI only steps in where that dataset does not reach. Three origins, read differently.
-
4 · Preparing the patient conversation
AI prepares the conversation, it does not have it. The five questions that do not work, and the one that tells you in twenty seconds how they really take it.
-
5 · Checking against the SmPC
The model does not read the SmPC: it writes what one usually says. Where each answer lives, and why a citation from memory reads just as well.
-
6 · Triage support: when AI helps you decide to refer
It can help you not forget a warning sign. It cannot decide for you whether to refer. Exactly where the line sits.
-
7 · Exercises with synthetic clinical cases
The AI has already answered and your job is to say what is wrong. Two cases, marked against a rubric: missing the serious one does not pass.
Level 4 — Owner: business, website and automation Premium
The map of what can be automated, and then the tools you build it with:
your own site, the measuring, an assistant and an app.
-
4 · The stack: GitHub, Vercel, Supabase and Claude Code
Premium
The four pieces you need for your own site, what each does and the order you set them up in. With the four costliest mistakes.
-
5 · Coding with Claude Code or Codex
Premium
An agent is not a chat: it opens your project and writes on its own. The four-rule method and the five traps you cannot see by reading the code.
-
6 · Google Analytics and Search Console
Premium
One tells you what happens inside your site, the other how people get to it. How to install them without ending up at zero, and why position 6 and 11 are different problems.
-
7 · Building a chatbot for your pharmacy
Premium
Building it is the easy part. The hard part is what it talks about, what it does not, and who answers for it. With the prompt that makes it safe.
-
8 · An app with Google AI Studio and the free lite models
Premium
What "free lite model" actually means, and the cascade that turns the quota into something you can rely on.
-
AI for pharmacy owners: the map of 17 areas
Premium
The seventeen things an owner can take to AI, what data each one needs, and which are already built into this site.
-
Agents · 1 · What an AI agent actually is
Premium
A model in a loop, with tools and a stopping rule. The loop, the cost arithmetic, when it is the wrong tool, and the six named failures.
-
Agents · 2 · Your first agent, step by step
Premium
From an empty folder to watching it chain two tools on its own. The full code, no libraries, and four ways to break it on purpose.
-
Agents · 3 · The tools: the contract
Premium
How to write one the model uses well, and the four conditions for one that writes: idempotency, confirmation, hard caps and logging.
-
Agents · 4 · The agent in production
Premium
Where it lives, what gets logged, the four weekly numbers, prompt injection, and who answers when it gets things wrong in a pharmacy.
-
Building an agent that does your ordering
Premium
What the arithmetic works out and what the AI adds, with a simulator over a sample catalogue. The number one request from any pharmacy owner.
-
Building your pharmacy website with AI
Premium
AI drafts; what you may claim is decided by the law. With a duties auditor and a checker for unauthorised claims.
Practical cases 23 cases
-
Starting from zero
5 cases
If you have never used AI for anything, start here. Five things you already do in the pharmacy, done with help.
-
Counter and team
5 cases
What gets written, pinned up and said. Signs, protocols, training and the difficult conversations.
-
The pharmacy's numbers
5 cases
Your own data, asked in plain language. Sales, dead stock, purchase terms, the rota and payroll.
-
Being found
4 cases
Google, social, your first web page and the messages you already send. What somebody sees before they walk in.
-
Actually automating
4 cases
From a sheet that checks itself to your first agent. And the list of what you will never automate.
See them all →