Preparing team training with AI | AI School for Pharmacists

"Prepare a training session on skincare for my team." It comes out instantly, it comes out long, it comes out well written, and it is useless: it is the average session on the average topic, and your team does not fail on the average topic.

The raw material for useful training is not a topic: it is what actually goes wrong at your counter. The model does not know that and cannot know it. It is the one thing you have to supply — and you already have it, without noticing.

Where that material already is, without looking for it

What you had to look up this month. Every time you said "let me check that" there was a gap. If you had it, the team has it.
Returns and exchanges. A product that comes back is usually a recommendation that did not fit.
The questions that repeat. Three people asking the same thing in a week is a topic; one is an anecdote.
What gets referred too often and what does not get referred enough. Both are training, and nobody ever asks for the second.

The format: twenty minutes, one decision

Team training in a pharmacy happens before opening or between customers. There are no forty slides; there are twenty minutes and people standing up.

Piece	How long	What it is
The case	2 min	A real counter situation, with no names
What you need to know	8 min	Only what changes the decision. No pathophysiology
The decision	5 min	What you recommend, what you refer, and where the line is
The sentence	3 min	How you say it, word for word, at the counter
Questions	2 min	And whatever nobody knows gets written down and looked up

"The sentence" is the part that is almost always missing and the only one anybody uses the next day. Knowing what to recommend and not knowing how to say it in ten seconds with somebody waiting is stopping halfway. Ask for it explicitly.

A prompt that works

Prepare a 20-minute session for a community pharmacy team.
It is one pharmacist and two technicians.

The real problem: this week we recommended a nasal decongestant
three times to people already on antihypertensives.

Structure: case (2 min) · what you need to know (8) · the decision
and where the line is for referral (5) · the exact sentence we say
at the counter (3) · questions (2).

No pathophysiology that does not change the decision. No slides:
a script I can read from.

If you are missing information about how we work, ask me instead
of assuming it.

Two things you have to supply

Your protocol. The model does not know whether your pharmacy refers, rings the surgery, or writes it on the patient record. If you do not tell it, it will invent a reasonable one — and the team will walk away with a procedure that is not yours.

And the clinical review. A team session turns into how things are done from the next day on: one wrong fact in there gets repeated at the counter for months. Check what you are going to say the same way you would check any answer — lesson 3 of Level 1 — and for anything about medicines, against the product information.

A full example, and what is wrong with it

With the prompt above — the one with the real problem inside — you get a twenty-minute session with its structure and its script. And it is usable: take it to tomorrow's meeting and it works. This example is about the three things to change first, and why none of them shows when you read it.

This is the part that matters of what comes back — the decision and the counter phrasing.

What it replies

The decision (5 min). Systemic nasal decongestants are contraindicated in hypertensive patients because of their vasoconstrictor effect. Faced with anybody who reports taking antihypertensives, do not dispense: offer saline and refer to their GP. Record the intervention in the patient record and notify the health centre so that it is on file.

The phrasing (3 min). "You see, decongestants containing pseudoephedrine produce systemic vasoconstriction, which can raise your blood pressure and is why they are inadvisable in your case; I would recommend using saline and consulting your doctor if the congestion persists."

What you need to know (8 min). Pseudoephedrine is an alpha-adrenergic agonist producing vasoconstriction of the nasal mucosa. Its systemic absorption entails stimulation of alpha-1 receptors in vascular smooth muscle, with the consequent rise in peripheral resistance…

And now, what is wrong with that reply:

It has invented your protocol, and it sounds reasonable. "Record the intervention in the record and notify the health centre" — do you? In most pharmacies, no: there is no record to write it in and phoning the health centre about this is not what happens. It is not a clinical error, it is a fake procedure, and it is the most expensive one in this lesson because a team session becomes "how we do things here" from the next day on. What it has handed you is the average protocol of an average pharmacy, again.
"Contraindicated" with no nuance is an absolute that will let you down. The caution is real and the criterion of not dispensing without more information is defensible. But put that way — contraindicated, always, in any hypertensive — the team will have it contradicted by a doctor, by a leaflet, or by the patient who has taken it twenty times. And when that happens it is not only that sentence that falls: the whole session falls, and the next one with it. A criterion with its nuance — uncontrolled blood pressure, brief use, what to ask first — survives an argument.
The "phrasing" cannot be said at a counter. Say it out loud: forty words, featuring "systemic vasoconstriction" and "inadvisable in your case", and by the time you reach the end there are two more people waiting. You asked for the exact phrasing and got a paragraph from a leaflet. It is the piece that decides whether the session is any use the next day and the one that comes out worst by default, because the model writes to be read, not to be said. Asking for the size fixes it: "the phrasing, in under twenty words and with no technical word in it".
And the eight minutes went on physiology. You forbade it in the prompt and there are the alpha-1 receptors. It is not disobedience: "what you need to know" in a training text means, statistically, mechanism, and eight minutes is a lot to fill with anything else. What actually changes the decision is three lines — what to ask, what you can offer, when to refer — and that leaves five minutes spare. That is where the real cases you brought belong.

Pasting in your protocol as it actually is, even as one sentence, fixes three of the four: "we do not record anything here, we offer an alternative and if in doubt we call the doctor". With that it stops inventing, and the rest of the session adjusts itself to how you work.

The fourth — the clinical nuance — no prompt fixes, and it is the one you have to supply: what gets said in a team session gets repeated at the counter for months, so it is checked beforehand, with the product information in front of you, like any other answer from Level 1 lesson 3.

And if your pharmacy is not like that

If there are two of you in total. Then "training" is a big word for what is needed, and the twenty-minute format is overkill. What works is one thing a week, three minutes, while the drawers get restocked: a case that happened, what was decided, and the phrasing. Ask the model for it that way — "a three-minute item, one case only, no introduction" — and you will see it change completely. Twelve short items a quarter beat one long session postponed three times.

If you get new staff every few months. There what pays is not a session: it is turning each one into a one-side sheet that gets filed and reused. Ask for it in the same request — "and give me the half-page version for the folder as well" — because doing it afterwards costs twice as much. In a year you have the induction manual nobody has time to write, and written about what actually happens in your pharmacy rather than about a syllabus.

If you are the employee, not the owner. It works the same and needs nobody's permission: anybody can run three-minute items, and turning up with "this happened to me, I looked this up, and I think we should do this" is the fastest way to change anything. The caution is the usual one and it goes first: anything clinical gets checked before it is told, because the credit being spent there is yours.

If the topic is commercial rather than clinical. A campaign, a new category, how to offer a service. There is far less verifying here — no product information to check against — and the model does well, but the other problem shows up: it will hand you the average pitch, which is everybody's. Give it your numbers — what sold, what was left, what people ask you — and have it start from there. And mind what gets said to the customer: the rules on claims are the same as in the previous lesson, even spoken.

When it does not work first time

I get an enormous session and I do not know what to cut.

Cut by a rule, not by feel: out goes anything that does not change a decision at the counter. Mechanism of action, epidemiology and the drug's history all fall out under that criterion, and nothing is missed. And if you would rather not decide yourself, ask it this way: "tell me, for each section, which specific decision it changes; delete any that changes none". It does that well, and the result is something you can argue with.

The team liked it and a week later nobody is applying it.

That is normal and it is almost never the session's fault: the trigger is missing. A session changes what people do when somebody knows the exact moment they have to remember — "when somebody asks for a decongestant, ask whether they take anything for blood pressure" is a trigger; "bear interactions in mind" is not. Ask explicitly "at what precise moment at the counter does this apply", and if it is not clear to the model, it will not be clear to your team either.

I have no cases written down to start from.

You do have them, just not in a notebook. Look at this month's returns and at what you had to go and check: those are the two places where what went wrong is already recorded, and both exist already. Three of each gives you a quarter's worth of topics. And to start collecting them effortlessly: a note on your phone, one line every time somebody says "let me look that up". In two weeks there is more material than you can get through.

What if I prepare the session and say something wrong?

That is why the clinical part gets checked first, and why it is worth saying out loud where each thing comes from: "I looked this up in the product information", "I could not confirm this one". Put that way, a doubt of yours teaches more than a certainty — the team learns the method, which is what gets repeated afterwards. And if an error turns up, it gets corrected in two minutes at the next one; the damage from being wrong once is vastly smaller than from never running a session.

Before you call this learned

I start from something that happened at my counter, not from a topic.
I say how many people there are and what training they have.
I ask for the exact sentence to say, not just the reasoning.
I give it our protocol, or I ask it to ask.
I verify the clinical content before teaching it.

← Previous: patient-facing material Back to the school →