Drug information answers with citations you then verify
Summarizing information is the single most common AI use pharmacists report — 53% of chatbot users in a 2025 survey — but the same literature shows why raw chatbot answers cannot be the final word: one study found ChatGPT answered only 10 of 39 real medication questions satisfactorily and fabricated references. The workable pattern is a structured prompt that forces the model to separate what it knows from what it is guessing, feeding a verification step the pharmacist was going to do anyway.
You are a drug information pharmacist. Answer this question for a practicing pharmacist, not a patient. Question: {{question}} Setting and relevant de-identified context: {{context}} Format: 1. Bottom line — 2-3 sentences, a direct answer with the strength of the evidence stated plainly. 2. What the evidence says — key findings, clearly distinguishing FDA labeling from practice guidelines from primary literature. 3. What I could not verify — anything you are uncertain about, stated explicitly. 4. Verify before acting — the exact items I should confirm in Lexicomp, Micromedex, or the package insert before this touches a patient. Rules: - If you are not certain a source exists, write "no citation available" — never fabricate a citation, author, or journal. I will check every reference you give. - Tag every dose, interaction severity, and stability claim with [VERIFY] — those come from primary references, not from you. - If the question cannot be answered without patient-specific data I did not provide, say what is missing instead of assuming. - Under 400 words.
Fill in your details and the prompt updates live — then copy.
Bottom line: Yes — calcium carbonate reduces levothyroxine absorption, and standard practice is to separate the doses by several hours [VERIFY: exact interval per Lexicomp]. This is a well-documented, manageable interaction, not a reason to stop either product. What the evidence says: FDA labeling for levothyroxine lists calcium carbonate among agents that decrease absorption and advises separating administration [VERIFY: labeling wording]. Small prospective studies showed reduced T4 levels with co-administration. What I could not verify: whether newer calcium salts differ meaningfully. Verify before acting: the recommended separation interval and any patient-specific absorption issues.
The full workflow
- Ask the question with setting and context, not just the drug names
- Check every [VERIFY] tag against Lexicomp, Micromedex, or the package insert
- Spot-check any citation the model offers — fabricated references are common
- Document your answer from the primary reference, not from the chat transcript
Watch out for
A chatbot is never a primary drug information reference. In published testing it missed a real Paxlovid-verapamil interaction and answered most medication questions unsatisfactorily — verify everything against Lexicomp, Micromedex, or the labeling.
Fabricated citations are the signature failure: studies found nearly half of AI-generated medical references do not exist. Check each one before it enters your documentation.
HIPAA: strip patient identifiers from the context — a de-identified scenario answers the question just as well.
Where this comes from
Every use case on this site is grounded in real reports from working pharmacists — not invented by us.