Run two short sessions: 45 minutes on safe uses and a do-not-paste list, then 60 minutes of hands-on tasks with counter staff and pharmacists together, each person signing a one-page sheet at the end. Keep AI to drafting and tidying text, never to clinical decisions, and never to anything with a patient's details in it.
Pharmacy teams differ from a general office in one way that matters here: almost everything on the screen at the counter is patient information. A cheerful "try it on your emails" session, borrowed from a generic course, will not tell your staff what to do when a regular asks whether an interaction matters or when the owner's own rota spreadsheet is full of names. The plan below is for an owner or lead pharmacist to run in the shop, using the tool you already have, with rules written for a pharmacy.
The worked example is a three-site independent chain: one owner-pharmacist, four employed pharmacists, eight dispensing and counter staff, and two delivery drivers. Everyone follows the same do-not-paste list; the hands-on tasks differ by role. If you are earlier in the journey and still deciding where AI belongs in the business, the first 90 days of AI for an independent pharmacy owner covers that first.
What to settle before the first session
Choose one tool and one account type
Pick one assistant for the team and make it a work account, not a personal one. Business plans from the main vendors (ChatGPT Business, Claude Team, Microsoft 365 Copilot and Gemini in Workspace) say they do not train on business content by default, while consumer plans may need a setting changed. The business seat prices I checked on 4 November 2026 were $20 a user a month billed annually or $25 monthly for ChatGPT Business (from two seats), and $21 a user a month on annual billing for the Microsoft 365 Copilot Business add-on; if you already run Microsoft 365, Copilot Chat comes at no extra cost with eligible business plans. Do not buy seats yet. For the two sessions, two or three shared seats or a work-account free tier are enough, and the owner can decide who needs a licence afterwards.
Name the person who answers "can I paste this?"
In most independents this is the owner or the responsible pharmacist. They give a clear yes or no within the day. Put their name on the sign-off sheet.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowCheck the current guidance from your professional body
Your regulator and professional body publish guidance on technology and patient information that changes. I have not quoted any here. Before session one, spend fifteen minutes reading what applies to you, and add one line to the rules sheet that points at it.
Session one: safe uses and the do-not-paste list (45 minutes)
Everyone attends: pharmacists, counter staff, dispensing staff. Drivers can join the first 20 minutes. The point is to give the team a common picture before anybody types anything.
| Minutes | Content |
|---|---|
| 0-5 | Why the shop is doing this and what will not change: the pharmacist's judgement and the dispensing system stay in charge |
| 5-15 | What these tools are good at and where they fail: a live example of a confident wrong answer |
| 15-30 | The do-not-paste list, read aloud, with five "paste or not?" cases |
| 30-40 | Safe uses: three examples of text work that involves no patient data |
| 40-45 | Questions, and how to reach the person who answers data queries |
The do-not-paste list
Print this and fix it by the till and in the dispensary. It should fit on half a page.
NEVER PASTE INTO ANY AI TOOL, EVEN WITH A WORK ACCOUNT
- A patient's name, address, phone number, email or date of birth
- Prescription details, medicines, doses, conditions or allergies
- Anything from the dispensing or patient records system
- Delivery lists, round sheets or anything with patient addresses
- Staff personal or disciplinary information
- Supplier prices, contracts or commercial agreements
- Passwords, log-ins, or screenshots that show any of the above
OK TO PASTE:
- Text you wrote yourself that names no patient
- Public information (our opening hours, our website text, a supplier's
public brochure)
- Made-up examples ("a customer asks about...")
NOT SURE? Ask [name] first. Wait for the answer.Rule of thumb for borderline cases: if you could not leave the page on the counter in view of a customer, do not paste it. Screenshots are the sneaky route; staff think of "text", then paste an image of the screen. Say so in the session.
Five paste-or-not cases for the room
- "A customer wants a reply to her complaint about a late delivery; may I paste her email?" No. Remove her name and address first, or write the reply from a made-up version.
- "Can I paste our opening hours and ask for a Christmas notice?" Yes. It is public.
- "I want a plain-English version of a leaflet's wording for a patient." Yes if the leaflet is the public one, but the pharmacist checks the result against the original, and it is never used as advice.
- "Can I paste a rota with everyone's names to rebalance the shifts?" Use initials or letters. Staff names with hours are personal data too.
- "A patient asked whether two medicines clash, can I ask the AI?" No. That is a clinical question for the pharmacist, using the references your practice already trusts.
The fifth case matters most. Say out loud that the assistant is not a reference source and that a smooth answer is not evidence. The longer treatment is in which pharmacy tasks should never be handed to AI, which is worth setting as pre-reading for the pharmacists.
Safe uses to demonstrate
Demonstrate three, live, and use a different one for each role group.
- A shop notice. Prompt: "Write a 60-word notice for our shop door saying we will close at 1pm on Christmas Eve and reopen on 27 December. Friendly, no exclamation marks." The result is checked for dates and opening times against the diary.
- A plainer version of an internal process. Paste the shop's own handwritten procedure for opening up, and ask for a numbered checklist. A pharmacist checks every step against the real procedure; the AI may drop or add steps.
- A first draft of a job advert. The owner supplies the facts: hours, pay band, duties. The AI supplies the wording. The owner removes anything that promises more than the job offers.
A realistic illustrative sample output for the first one, with what you would fix:
Reply (illustrative):
"Season's greetings from all of us. Please note we will close at 1pm on
Christmas Eve and reopen as normal on 27 December. Please order any repeat
medicines early so we can have them ready for you before the break."
What to fix: check the dates. Check whether "as normal" is true (is
Boxing Day closed?). Delete the sentence on repeat medicines unless the
shop actually offers that service in that window. Never let the AI add
a promise about service you have not agreed.End of session one
Hand out the do-not-paste list and the sign-off sheet (below), and give one small task before session two: write down two repetitive writing jobs you do each week that contain no patient information. A counter assistant might write "Writing out the weekly order note" and "Replying to staff WhatsApp about shifts".
Session two: hands-on tasks, in pairs (60 minutes)
Run this two or three days after session one, with the same group. Put people in pairs: one pharmacist or experienced assistant with one newer person. Pairs keep each other honest about the do-not-paste list.
| Minutes | Content |
|---|---|
| 0-10 | Quick recap: three cases from the do-not-paste list; each pair gives one answer |
| 10-30 | Tasks A and B (below), done in pairs on laptops or the shop PC |
| 30-45 | Task C: check an output for mistakes |
| 45-55 | Share the best prompt from each pair; save to a shared document |
| 55-60 | Sign-off sheet and next steps |
Task A: draft a routine reply from a template
Use a made-up enquiry: "A customer asks whether we offer free delivery in a certain area." No real patient. The pair writes a prompt giving the shop's real delivery policy (typed in) and asks for a reply of under 80 words. The check: does the reply promise anything that is not in the policy? Here is a before and after of the prompt.
BEFORE: write a reply about free delivery
AFTER: Write a reply of under 80 words from a community pharmacy to a
customer asking about delivery. Use only these facts: delivery is free
for repeat prescriptions over a minimum order of [AMOUNT] within [AREA];
otherwise it costs [AMOUNT]; delivery days are [DAYS]. If you need a fact
I have not given, write [FACT NEEDED]. Friendly, plain English.Task B: turn a staff process into a one-page checklist
Choose a non-clinical process, such as opening the shop, handing over to the next shift, or checking the temperature log for the medicines fridge (the checklist, not the readings). Pharmacists often find this the most valuable task: the first draft shows gaps in what the team does and writes down. The pair then edits the checklist against how it is really done. See writing pharmacy SOPs with AI for a full workflow, and treat any AI-drafted SOP as a draft that the responsible pharmacist approves.
An illustrative first draft for a shift handover, with what the pair should fix:
SHIFT HANDOVER CHECKLIST (illustrative AI draft)
1. Check and sign the controlled drugs register.
2. List prescriptions waiting for collection.
3. Note any owings and the date each is due.
4. Check the fridge temperature log is filled in.
5. Pass on any customer callbacks promised today.
6. Lock the dispensary and set the alarm.
What to fix: step 1 is a guess at a routine the AI cannot know; use
the shop's own controlled drugs procedure wording, or delete it. Step 6 belongs to closing,
not handover. Missing: who takes over as responsible pharmacist, and
where the handover note is kept.Notice the pattern: the draft is tidy, plausible and partly wrong for this shop. A pharmacist who reads it for two minutes catches the gaps; a newer assistant who files it as the shop's procedure does not. That is why every AI-drafted checklist should carry the approving pharmacist's initials and a date before it goes into the procedures folder, and why the pair, not the tool, has the last word on each step.
Task C: find the mistake
Before the session the owner prepares three short AI-made texts, each with one planted fault, all built on made-up data:
| Text | Planted fault |
|---|---|
| A shop notice about opening hours | Wrong day for reopening |
| A summary of a public leaflet | A recommended maximum that the original does not state |
| A reply to a made-up complaint | Promises a refund policy the shop does not have |
Pairs have ten minutes to find all three. Then ask: what would have happened if nobody had checked? The leaflet fault is the one that makes people sit up, and it carries the lesson that a polished summary is not a source.
Role-specific extras (use one per group, 10 minutes)
- Counter staff: turn the shop's own return-and-refund policy into a short script for the counter, with a polite "I will check with the pharmacist" line for anything health-related.
- Pharmacists: draft a staff memo about a new internal process, and compare its steps to the real one. Practise saying "no" to using it for anything clinical.
- Dispensary technicians: make a rota template with letters instead of names; check the hours add up using a calculator, because chat tools are unreliable at arithmetic.
- Delivery drivers: they do not need task work; give them the do-not-paste list and the rule that delivery sheets never go into a chat tool.
Three counter scenarios to role-play in session two
If the room has ten spare minutes, role-play. Each scenario below has the situation, what the member of staff should do, and what usually goes wrong. Use made-up customers throughout.
Scenario one: a regular asks what the shop thinks of an AI answer
A customer shows a phone with a chatbot reply about combining two over-the-counter products. The assistant should say, "I can't comment on that, but let me fetch the pharmacist", and hand over. What goes wrong: the assistant reads the reply, finds it sensible, and agrees. Practise the sentence aloud until it is automatic.
Scenario two: the owner asks for a quick reply to a supplier
The owner dictates a note to a supplier about a late order and asks an assistant to "get the AI to tidy it". The assistant pastes the note; it contains the supplier's account number and a negotiated price. The group spots both on the do-not-paste list, and rewrites the prompt using [ACCOUNT NO] and [PRICE], to be restored by hand. What goes wrong: the account number goes in first and the group notices afterwards. The lesson is to look at the text before you paste, every time.
Scenario three: someone wants to paste a screenshot
A newer assistant has a confusing screen from the dispensing system and wants the AI to explain it. The answer is no, and a better route: write down the question in your own words ("the system shows a warning I haven't seen; what does it usually mean?") and ask the pharmacist or the system's support line. What goes wrong: screenshots carry names in headers and sidebars that the person did not even notice.
Keeping locums and part-timers in the loop
Pharmacies rely on people who are not there every week, and they are where a policy tends to fail. Three small habits close the gap:
- Put the do-not-paste list in the locum pack beside the fire exits and the till code, and ask the locum to sign the sheet on their first shift.
- Keep one short recording (five minutes, screen and voice) of the owner walking through the list, so people who missed session one can watch it on their phone before they start.
- Ask the responsible pharmacist on duty to answer data questions that day, so the person named on the sheet is not the only route.
What to do if something is pasted by mistake
Mistakes will happen. Agree the steps now so people report them. First, stop and tell the data decider immediately; no blame for reporting. Second, delete the conversation in the tool and note the date, time and what was pasted (describe it; do not copy it into the note). Third, the owner or data protection lead decides whether any report to a regulator or to the patient is needed, and records the decision. Fourth, add the case, anonymised, to the next team briefing. People who report quickly should be thanked in front of the team, otherwise the next person hides it.
The sign-off sheet
A single sheet each person signs and dates keeps the owner able to show that staff were told. It does not remove the owner's responsibilities, but it makes the rules real. A filled-in version for one member of the team:
AI USE: STAFF SIGN-OFF
Name: [first name, surname] Role: Counter assistant Site: [site]
Date of session one: [date] Date of session two: [date]
I have read the do-not-paste list and I understand:
[x] I never put patient or staff personal details into any AI tool.
[x] I use the shop's work account only.
[x] I do not use AI for clinical questions or interactions.
[x] I check any text before it reaches a customer.
[x] If unsure, I ask [name] before pasting.
Signature: __________ Checked by: __________Keep the sheets in the staff file. Re-run the sign-off for new starters in their first week, and for everyone once a year or when you change the tool.
Two weeks later: check it stuck
A short, kind follow-up is better than an audit. For each person ask: "Show me one thing you used the tool for this fortnight." Look for two signs. First, the task contains no patient information. Second, they changed something in the output before using it. If someone shows a clean example, great. If someone cannot show anything, ask what stopped them; the answer is usually "I wasn't sure what I could paste", which means the list needs a clearer example. If you find any patient detail in a chat history, do not shame them; remove it, record it in the incident note, and ask your data protection lead whether anything must be reported.
A simple tracker
| Measure | Target at two weeks | Illustrative result |
|---|---|---|
| Staff who have signed the sheet | 14 of 14 | 13 (one locum outstanding) |
| Staff who used the tool for a non-patient task | At least half | 8 of 14 |
| Patient details found in chat history | None | None |
| Hours saved a week across the team (rough) | Any figure above 2 | About 4 |
Those results are made up to show the layout. The measure that matters is the third: zero patient details, every time.
When a customer or patient asks about it
Staff will be asked, "Do you use AI?" Give them a straight answer to say: "We use it for things like drafting notices and tidying our own paperwork. We do not put patient information into it, and a pharmacist checks anything that matters." Honest, short and true. If you later add AI to anything patients see, such as a phone line or chat, say so on the service itself; the wording in telling patients you use AI note-taking adapts well. If a phone or chat tool is on your list, AI agents for pharmacies: phone, chat or refill requests first sets out the order to try them.
Using AI to build the induction for new counter staff
Once the sessions have run, the cheapest improvement is to fold the do-not-paste list into the induction pack. Building induction packs for counter staff with AI shows how to draft the pack, and the AI-use page becomes one page of it. A new starter then meets the rules on day one, before anyone has a chance to learn by mistake.
If you want help choosing a tool and tuning the sessions to your pharmacy's systems, that is the sort of thing my AI implementation consultation covers. The plan above can be run without it, with under two hours of the team's time in one week, and a printed list.
Questions pharmacy owners ask before running the sessions
Can staff use AI for anything clinical?
Not as a source for clinical decisions. A pharmacist's judgement, the dispensing system and the reference sources your profession already uses stay in charge. AI is for drafting and tidying text, such as rotas, notices and routine replies, and the pharmacist remains responsible for anything touching a patient's care.
Is a pharmacy chatbot account safe for patient details?
Treat patient details as never pasted unless your data protection lead has approved a specific tool and setting in writing. The simplest safe rule is the do-not-paste list in this plan: no names, dates of birth, prescriptions, conditions or contact details in any chat tool.
How long should each session be?
Session one runs 45 minutes and session two 60. Run them before opening or after closing, in the same week if you can, so what people hear on Tuesday is still fresh when they try it on Thursday. Pay or time off in lieu for anyone attending outside their hours.
Do locums and part-timers need to attend?
Yes, or they should read the one-page sheet and sign it before their next shift. A single person pasting a patient's details into a free tool undoes the whole plan, so the sign-off sheet matters more than perfect attendance.
Further reads
- How Independent Pharmacies Use AI to Shorten Phone Queues — Using AI to shorten phone queues, once staff are trained.
- Pharmacy Automation Companies: A Buyer's Checklist Before You Sign — Questions to ask before signing with an automation vendor.
- AI Software for a Community Pharmacy: Patient Reminders and Delivery Rounds — Reminders and delivery rounds, with the data limits in mind.
- Is It Safe to Put Customer Data Into ChatGPT? — The data question behind the do-not-paste list.
- How Much Time and Money Does AI Staff Training Take? — Budget the hours for training a bigger team.
- AI Training Programme for a Company: A Four-Week Staff Rollout Plan — A four-week version for bigger teams.
- AI for Mental Health Providers: A Roll-Out Plan for Group Practices — A seven-step roll-out sequence for a group therapy practice, with consent wording, an approved-tool list and a 90-day review.
- Best Pharmacy Dispensing Software: Does Any of It Have Real AI? — Six pass-or-fail tests that separate genuine AI from rules-based automation in dispensing systems, with a scored worked example.
- Independent Pharmacy Marketing: Flu and Travel Campaigns With AI — A seasonal marketing calendar for an independent pharmacy, with AI-drafted posts, medicines advertising guardrails and a sign-off step.
- Pharmacy Near Miss Log: Finding Patterns With AI, Not Blame — How an independent pharmacy can turn a near miss log into themes and actions with AI, using de-identified entries and a blame-free team review.
- How AI Helps Pharmacists Streamline Routine Tasks: A Two-Week Audit — A two-week task log for an independent pharmacy, sorted into hand over, assist and never, with counter and dispensary examples of each.
- Is There an AI Tool for Pharmacists? What US and UK Teams Can Use — Yes: ChatGPT for Clinicians for verified US pharmacists, general business-plan tools elsewhere. What to use them for and what to keep out.
- What Is Automation in Pharmacy? Dispensing Robots vs AI Office Workflows — Pharmacy automation splits into dispensing hardware and office workflows. See what each does and what a small independent can automate without a robot.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: ChatGPT Business, Claude, Google Workspace and Microsoft 365 Copilot Business pricing pages and Microsoft's Copilot privacy documentation (read 4 November 2026). No clinical guidance is quoted; check your professional body's current guidance on AI.