# Somebody Already Asked It About Their Medicine — the opening of AI for Pharmacists

> Somebody Already Asked It About Their Medicine — the opening of AI for Pharmacists, free to read in full. 2,184 words on what AI does and does not do in this trade.

*Free to read in full. 2,184 words, the first section of AI on the Job: Pharmacists Edition.*

The first time this technology showed up in your pharmacy, it was in somebody else’s hand.

A woman at the window, phone turned around so you can see it, asking if this is right. Half a page of calm, well-organized text about the medication she is collecting, friendlier in tone than the leaflet stapled to the bag. Most of it is correct. One sentence in the middle is wrong in a way that matters, in the same even voice as the true parts, so she believed all of it equally on the drive over.

You sorted it out in under a minute, between verifications, and nobody wrote it down anywhere. That was the technology, seen accurately, before anybody sold you a thing: fluent, useful in places, and unable to tell you which of its own sentences is the bad one.

Both of the loud stories you have heard since then are that moment with one half deleted.

The first keeps the confidence and drops the error. It says the counter is finished. Counting is machine work already, and a good share of what your patients take was filled by a warehouse three states away and dropped in a mailbox. Verification, the story goes, is pattern matching, and pattern matching is what these models do.

The second story wears a lanyard. It turns up at a trade show, or in a webinar invitation that got past your filter, and it takes your prior authorizations, your refill outreach and your phone lines. More sync patients by spring, fewer abandoned scripts, a discount that ends Sunday. Both stories live in a year nobody has worked yet.

Neither one moves anything in front of you tomorrow. The verification bin is still deep, the 75 on the inhaler is still a 75, and the man whose prior auth you started on the fourteenth still has no medicine.

The five years are not this book’s business. The prior auth is.

Take the first story. A list finishes it, and everything on the list is something you did before lunch.

It cannot verify a fill. The last check is a licensed person holding the hard copy against the label against the thing actually in the vial, and deciding. Machines count, sort and flag, and have for years. The decision at the end lands on a name and a license number, because accountability has to land somewhere findable.

It cannot see the patient standing there. It cannot notice that the woman collecting her husband’s tablets mentions he has been sleeping a lot lately, on her way out, to the floor, in the voice people use when they are not sure it is worth saying. That sentence is the whole intervention, and it arrives once, sideways, only if somebody is there to catch it.

It cannot catch the interaction that actually matters. Your system fires alerts all day and you clear most of them, because most are noise. The one that counts is usually the one the profile does not contain: the supplement off your own shelf, the tablet borrowed from a sister who had some left. You get there by asking. No model has met your patient.

It cannot see the patient who is not taking it. Non-adherence produces no document. It shows up as thirty-eight days between refills on a thirty-day supply, as an inhaler that lasts a season, as a face you have not seen since March. An absence is only visible to somebody expecting a person.

It cannot counsel. Counseling is not the recitation. It is watching a face while you say the thing and saying it a second way when it plainly did not land. It is realizing halfway through that this is not a question about the medication at all — it is a question about money, and the person is deciding whether to fill it. What you say next settles that.

It cannot draw up a vaccine. It cannot hold for nine minutes and tell another clinician calmly that you are not comfortable with this one and here is why. It cannot refuse to fill.

And it is not a drug reference. Of everything here, that is the fence that matters most. Ask it about a dose, an interaction, a place in therapy, and the answer comes back fast and well-structured, in the same steady tone whether it is right, wrong, or three label revisions out of date. Nothing it knows carries a date and nothing it invented carries a mark. Your monograph and your clinical-decision resources are curated, dated and accountable; a chat window is none of the three. Ninety seconds in the real reference is the difference. Chapter 10 turns that into a rule with your license attached.

So: no. Not the check, not the counseling, not the judgment, not the reference, not the license. That is the answer to the first story.

The worry underneath is not a silly one. A benefit manager already decides what your patients pay, where the prescription may go, and whether you are in the network, without ever touching a tablet. So the shift worth watching is not a robot at the register. It is who explains things to your patients, and who they believe. Chapters 7 and 9 are built on that.

Now the half that is not reassuring, because a book claiming nothing was changing would fall apart by Chapter 4.

Something is changing. Not the check, not the counseling. Every word that surrounds them.

Nobody mentions the writing in pharmacy school. You write the prior auth package, then chase it. You write the appeal, built around criteria that live in a PDF on the plan’s site. You write the missed-refill message that has to inform without scolding, and the handout that repeats what you already said, because she was holding a toddler and her keys. You write the wholesaler dispute with a clock on it, the schedule, the SOP, the onboarding for the tech who started Monday, and the reply to the one-star review about a decision you did not make. You write the flu campaign in August or there is no campaign in September. And if you are employed, the one you never write is the resume, which still says you verified prescriptions accurately.

None of it is pharmacy, and none of it is paid. It happens after the door is locked, in the ten minutes you meant to eat, or not at all — and “not at all” carries a price too, paid later by somebody else.

That pile is what these tools are honestly good at. Not deciding whether the therapy is right. The four paragraphs explaining to a frightened person why her plan wants her to fail two other drugs first, in words that sound neither like a form letter nor a lecture.

What they are bad at, since the booth will not raise it: they invent, and the made-up parts sound exactly like the rest. Ask for a plan’s prior auth criteria and back comes a tidy list belonging to no plan on earth. Ask about your state’s rules and you get another state’s, or nobody’s. It has not read your payer sheet and it has never been in your store.

So every workflow in here cuts the job in the same place. The facts are yours: clinical facts from your references, plan facts from the payer sheet and the denial letter, patient details from your own system. The sentences are its share. Where the cut is easy to fumble, a WATCH OUT box spells out the check, which takes about sixty seconds.

The commercial part, said once. Patients rarely leave over price. They leave over the fourth trip and the third time nobody told them anything. The pharmacy whose package arrives complete gets the approval days sooner, and the office that gets one clean page instead of three phone calls starts pointing people your way. None of that is new. What changed is what an explanation costs. That used to be an evening. It is now about ten minutes.

Take the second week of January, which every community pharmacy knows and nobody outside one believes.

The plan year turned over on the first. Deductibles reset, formularies moved, and none of it reached your patients gradually. By nine thirty the bin is deep and four claims sit rejected, one of them a 75 on a maintenance inhaler nobody has ever asked a question about before.

The man it belongs to is at the window while you read the screen. You explain it properly, which takes four minutes you did not have, and it works: he leaves annoyed at his plan instead of at you, which is most of the art of this job. Then you do it again at eleven, and twice after lunch.

The prior auth starts at 6:40, after the door is locked. Gathering, organizing, the cover note, the ask to submit today rather than Friday. Forty minutes, and it gets approved.

The second one you do faster and thinner, mostly blanks with nothing around them, and the office faxes back Wednesday asking for the part you left out, which costs three days. The third does not happen. It goes on a sticky note, and on Friday the patient calls to ask where his medicine is.

The day-seven call to the woman who started something new last Wednesday does not happen either, and nobody will ever count that one, because a call that did not happen leaves no record and neither does a bottle that only gets filled once.

The failure there is not a failure of skill. You explained beautifully at nine thirty, and the package you built at 6:40 was better than the office would have assembled alone. The failure is arithmetic: the good version costs forty minutes, January produced nine of them, and there is one pharmacist and a queue.

So, plainly, what you are holding.

It is a working manual for pointing a general-purpose chat tool, whichever of ChatGPT, Claude or Gemini you settle on, at the written half of pharmacy practice. Reject triage, prior auth packages, appeals. The counter explanation, the copay half-sheet, the counseling reinforcement handout, the day-seven call script, the med-sync pitch. Wholesaler and benefit-manager correspondence, reconciliation folders, SOPs, schedules. Review replies written with HIPAA discipline, and the resume that finally counts your interventions. There are 150 prompts, numbered, all repeated in a library at the back and findable in ten seconds with somebody waiting.

Every one of the 150 was tried in all three tools, and any that worked in only one of them was dropped. Nothing here is a prediction about what these things may eventually do behind a counter. It is only what survived the testing.

What it is not, said again because in this profession it is the whole ballgame: not a drug reference, not a clinical decision aid, not a second opinion on therapy, and never the place a question about a particular patient goes. It stays clear of controlled substances. Patient information does not enter a chat window at all — every prompt is built on brackets and de-identified structure, and the real details go in inside your own systems, afterwards. Anywhere the tool comes near something expensive, a RULES CORNER marks the line. Chapter 10 collects them, and it belongs in week one, not week six.

None of it requires you to be technical. It requires a phone, an email address and thirty minutes, which is all Chapter 2 needs. Chapter 3 is the one skill that decides whether any of it works, and you have had that skill since the first shift you ran with a new technician. You say the situation. You say what finished looks like. You say what comes back to you for a decision. A chat window takes the same briefing.

A word about scale, then you can start.

None of this is life-changing. It changes the hour after the door is locked, and inside a month, most of the week’s evenings. About eleven hours in an ordinary week, which is an honest number and not a free one: your first week costs time instead of returning any, which is what learning a tool does. Week three is usually where it starts paying you back.

Most pharmacists put a book like this down holding two things, usually the prior auth package and the reject card. That is not a shortfall, that is the design. Two you actually keep are worth more than twelve you meant to.

Nobody is coming for the counter. The last check still belongs to a person with a license and tired eyes, and the counseling still belongs to whoever is standing there when the real question comes out sideways. What this book offers is smaller than either story you were sold, and considerably more use: the prior auth out the same afternoon the reject appeared, the explanation printed instead of improvised, and a door you lock at closing time.

The next page is the arithmetic — which hours, and which chapter they come out of.

---

The rest is twelve chapters, 150 numbered copy-paste prompts, a glossary and a thirty-day plan.

- Book: ../books/016-ai-on-the-job-pharmacists-edition.html
- All 150 prompt titles: ../prompts/016-ai-on-the-job-pharmacists-edition.html

---

*AI on the Job Press · Greenlight Publishing · 2027 edition*