Doximity's 2026 State of AI in Medicine report put AI adoption among US doctors at 63%, up from 47% a year earlier [1]. The reason is not glamorous: clinicians routinely spend around two hours on paperwork for every hour of patient care, and AI scribes and drafting tools cut charting time by as much as 75% [2]. One large group practice estimated AI scribes saved physicians roughly 15,000 hours in a single year [3]. The clinicians getting that time back are not the ones with the fanciest tools. They are the ones with a reusable prompt library tuned to the non-clinical work that eats their day.
This article collects 50 ready-to-use prompts for healthcare professionals, organized by workflow: clinical documentation, patient communication, patient education, evidence and literature review, coding and billing, practice administration, quality and safety, and professional development. Each uses [bracket placeholders] so you can adapt it. They work with ChatGPT, Claude, and Gemini, with notes where one model is clearly better for a task.
Healthcare workflow map: eight categories where prompts give clinicians their time back
Before you paste: two non-negotiable rules
1. Patient privacy is absolute. Consumer AI chatbots (the free tier of ChatGPT, standard Gemini, Claude.ai) retain prompts and may use them for training. Pasting any identifiable patient information, names, dates of birth, record numbers, images, rare-diagnosis details that could identify someone, breaches your duty of confidentiality and data-protection law (HIPAA, GDPR, and local equivalents). For real clinical work, use an enterprise-grade deployment with a contractual no-training commitment and the appropriate security and compliance controls, ideally one covered by a Business Associate Agreement or its local equivalent. Otherwise, de-identify completely before you prompt: strip every identifier, keep only the clinical shape.
2. AI is not a clinician. These prompts support documentation, communication, and administration. They do not make clinical decisions. Never use AI output for diagnosis, treatment selection, drug dosing, or direct patient advice without the independent judgment of a qualified clinician. Models fabricate doses, contraindications, guideline thresholds, and citations with complete confidence, and are often trained on guidance that has since changed. Every clinical fact, dose, and reference an AI returns must be verified against an authoritative source: current guidelines, the drug label, peer-reviewed literature, your institution's protocols. Treat AI as a drafting and thinking aid, never as a source of clinical truth.
Safe-use decision aid: what you can paste, what to de-identify, and what must be clinically verified
With those two guardrails in place, here are the 50 prompts.
1. Clinical Documentation
The biggest time sink and the safest place to start. AI drafts from the clinical facts you provide; you review and sign. It never invents findings.
1. Structure a consult note from my notes
Turn my shorthand into a structured consult note in [SOAP / problem-oriented] format: [paste de-identified notes]. Use only the information I gave you. If a standard section has no input, write "not documented" rather than inferring. Keep clinical language precise and concise. Flag anything that looks internally inconsistent for my review.
2. Draft a discharge summary
Draft a discharge summary from these de-identified inputs: [admission reason, course, key results, procedures, discharge meds, follow-up]. Structure: reason for admission, hospital course, significant findings, discharge medications, follow-up plan, and outstanding actions. Do not add any diagnosis, result, or medication I did not provide. Mark gaps as "to complete."
3. Write a referral letter
Write a referral letter from a [specialty] to a [specialty] for a patient with [de-identified clinical summary]. Include: reason for referral, relevant history, current management, the specific question I am asking the receiving clinician, and what I have already excluded. Professional, concise, no filler. Leave a clear placeholder for anything I still need to add.
4. Summarize a long record for handover
Summarize this de-identified record into a handover summary: [paste]. Structure: active problems, current treatment, pending results and actions, anticipated issues on the next shift. Cite the source line for each active problem. Keep it scannable for a clinician taking over in two minutes.
5. Convert dictation into a clean note
Clean up this dictated text into a polished clinical note without changing any clinical content: [paste de-identified dictation]. Fix grammar and structure, expand obvious abbreviations consistently, and keep my clinical meaning exactly. If a phrase is ambiguous, flag it rather than guessing what I meant.
6. Draft a procedure note template
Draft a reusable procedure-note template for [procedure]. Include the standard sections: indication, consent, technique, findings, complications, post-procedure instructions. Leave bracketed fields for the case-specific details I fill in each time. Keep it compliant with typical documentation expectations for this procedure.
7. Generate a problem list from a summary
From this de-identified clinical summary, propose a structured problem list with active and inactive problems: [paste]. Base it only on what is documented. For each problem, note the supporting line. Flag anything mentioned but not clearly characterized, so I can decide whether it belongs on the list.
8. Draft a chronology of events
Build a clinical chronology from these de-identified notes: [paste]. For each entry: date, event, source line. Sort ascending. Flag any gap where something should logically have been documented but is missing. This is for my review and care coordination, not a legal record.
Want to know how effective your prompts are? Prompt Score analyzes them on 6 criteria.
Where clarity changes whether a patient understands and follows the plan. AI is strong at a first draft in language a patient will actually read. You confirm every clinical statement.
9. Explain a condition in plain language
Explain [condition] to a patient with no medical background, at roughly a [reading level]. Cover: what it is, why it happens, what we will do, and what they should watch for. Use one everyday analogy. Avoid alarming language. End with when to seek urgent help. I will verify every clinical statement before sharing.
10. Write post-visit instructions
Draft post-visit instructions for a patient after [visit type / procedure]. Include: what was done in plain terms, medications and how to take them [I will fill in and verify], activity and diet guidance, warning signs that need urgent attention, and the follow-up plan. Warm, clear, short sentences. Leave clinical specifics as placeholders for me to complete.
11. Translate a result into a patient message
Draft a message to a patient explaining a [normal / stable / non-urgent abnormal] result, in reassuring but honest language: [paste de-identified result context]. Say what it means, what happens next, and whether they need to do anything. Keep it under 150 words. Do not state any value or interpretation I have not given you.
12. Adapt a message for a different audience
Rewrite this patient explanation for [an anxious patient / a caregiver / a teenager / a patient who speaks English as a second language]: [paste]. Keep every clinical fact identical; change only tone, vocabulary, and length to fit the audience.
13. Draft a difficult-conversation framework
Help me prepare to discuss [difficult topic, e.g. a new chronic diagnosis] with a patient. Give me a structure, not a script: how to open, how to check understanding, the questions they are likely to ask, and language to avoid. This is communication coaching; all clinical content stays mine.
14. Write an appointment-reminder sequence
Write three short reminder messages for an upcoming [appointment / screening / vaccination]: one at [2 weeks], one at [3 days], one at [day before]. Each says exactly what the patient needs to bring or prepare. Friendly, brief, no jargon. Under 100 words each.
15. Answer a common patient FAQ
Draft a clear answer to the common patient question: "[question]." Audience: [patient type]. Keep it accurate, plain, and short, and end with a line on when this does not apply and they should call. I will review for clinical accuracy before publishing.
3. Patient Education Materials
Reusable assets that save the same explanation a hundred times. AI drafts; a clinician verifies every fact before it reaches a patient.
16. Create a condition leaflet
Draft a one-page patient leaflet on [condition / treatment]. Sections: what it is, what to expect, self-care, warning signs, and where to get help. Plain language, friendly tone, bullet points where useful. Leave every dose, figure, or threshold as a placeholder for me to fill and verify.
17. Build a pre-procedure checklist for patients
Write a patient-facing preparation checklist for [procedure]: what to do in the days before, the day before, and the morning of, including medications to pause or continue [placeholders for me], fasting, and what to bring. Clear, reassuring, scannable.
18. Draft a medication information sheet
Draft a patient-friendly information sheet for [medication], with placeholders for the specifics I will confirm: what it is for, how to take it, common side effects, what to do if a dose is missed, and when to seek help. I will verify every clinical detail against the current label before use.
19. Simplify an existing leaflet
Rewrite this patient material at a lower reading level without losing accuracy: [paste]. Shorten sentences, replace jargon with plain words, and keep all clinical meaning. Flag any sentence where simplifying risks changing the meaning so I can check it.
20. Create an FAQ for a condition
Generate a patient FAQ for [condition]: the 10 questions patients most often ask, with short, plain-language answers. Leave any number, dose, or threshold as a placeholder. End each answer with a clear "call us if..." line. I will verify before publishing.
21. Draft lifestyle guidance
Draft general, non-individualized lifestyle guidance for patients with [condition]: diet, activity, and self-monitoring, in plain language. Make clear this is general information and not a substitute for personal medical advice. I will review and tailor for each patient.
4. Evidence & Literature Review
AI compresses and compares what you have already gathered. It does not replace the search, and every clinical claim it summarizes must be checked against the source.
22. Summarize a paper for a journal club
Summarize this study for a journal club: [paste abstract / paper]. Structure: question, design, population, intervention, primary outcome and effect size, key limitations, and how much it should change practice. Note the level of evidence. Flag anything I should verify against the full text before relying on it.
Claude handles long full-text papers well within its context window; for several papers at once, Gemini's structured comparison output is often cleaner.
23. Compare two guidelines
Compare these two guideline summaries on [topic]: [paste]. Produce a table of where they agree, where they differ, and the practical implication of each difference. End with the open questions a clinician would still need to resolve. I will confirm the current version of each guideline.
The techniques you're reading about work. Test your prompts now with Prompt Score and see your score in real time.
I want to research [clinical question] structured as PICO: [population, intervention, comparison, outcome]. Draft a search strategy: key terms, synonyms, and Boolean structure for [PubMed / other], plus inclusion and exclusion criteria. I will run the search and appraise the results myself.
25. Extract the practice-relevant points
From these de-identified study summaries on [topic]: [paste], extract only the points that would actually change day-to-day practice, separated from the academically interesting but not actionable. For each, note the strength of evidence and what to verify before applying.
26. Draft a critical appraisal
Help me critically appraise this study: [paste abstract / methods]. Walk through risk of bias, sample size and power, confounding, generalizability, and conflicts of interest. Pose the questions a skeptical reviewer would ask. This structures my appraisal; the judgment is mine.
27. Translate evidence into a one-paragraph brief
Turn this evidence summary into a one-paragraph brief for busy colleagues: [paste]. State the bottom line first, then the strength of evidence, then the caveat. Under 120 words. No overstatement; if the evidence is weak, say so.
5. Coding, Billing & Reimbursement
Administrative accuracy that protects revenue and compliance. AI drafts the documentation framing; a coder or clinician verifies every code.
28. Suggest documentation to support a code
For an encounter I am documenting as [service / complexity level], list the documentation elements typically expected to support it, so I can check my note is complete. Frame as a documentation checklist, not a coding decision. I will confirm against current coding rules and our compliance policy.
29. Explain a denied claim
Explain this claim denial in plain terms and list the likely documentation or coding reasons: [paste de-identified denial]. For each likely reason, what to check and what a corrected resubmission would need. I will verify against the payer's current rules.
30. Draft an appeal letter
Draft an appeal letter for a denied [service] claim. Inputs: [de-identified denial reason, the clinical justification]. Structure: the service provided, the medical necessity as documented, the specific policy criteria met, and the request. Professional and factual. Leave clinical specifics as placeholders for me to confirm.
31. Build a coding-documentation checklist
Create a documentation checklist for [common encounter type] so the note reliably supports correct coding: the elements to capture, the common omissions that cause downcoding or denials, and a final yes/no review. I will align it with current coding standards.
32. Draft a prior-authorization request
Draft a prior-authorization request for [service / medication] based on these de-identified clinical facts: [paste]. Include: the clinical rationale, the criteria the payer typically requires, and the supporting documentation to attach. I will verify the payer's current criteria and all clinical details.
33. Summarize an encounter for billing
Summarize this de-identified encounter note into the elements a coder needs: presenting problem, history and exam scope as documented, complexity drivers, and services performed. Do not assign codes; surface what is documented so the coder can. Note anything missing that would affect coding.
6. Practice & Clinic Administration
The operational work that runs the clinic and rarely gets credit. Low clinical risk, high time savings.
34. Draft a clinic policy
Draft a clear internal policy for [process, e.g. handling abnormal-result callbacks / no-shows / prescription refills]. Include: purpose, scope, step-by-step procedure, who is responsible, the record to keep, and the escalation point. One page a new staff member can follow.
35. Optimize a clinic schedule template
Given this clinic's appointment mix and constraints: [describe visit types, durations, staffing, hours], propose a daily schedule template that reduces bottlenecks and idle time, with buffer for urgent slots and admin. Explain the trade-offs of the layout you propose.
36. Write a staff communication
Draft an internal message to clinic staff about [change: new process / system / policy]. Explain what is changing, why, what each role needs to do, and from when. Clear and respectful of people's time. Under 200 words. End with who to ask with questions.
37. Draft a patient-recall workflow
Design a recall workflow for [cohort, e.g. patients due for a chronic-disease review]. Include: how to identify the cohort, the message sequence, how to track responses, and how to escalate non-responders. Keep patient-facing messages plain and non-alarming.
38. Prepare a team huddle agenda
Build a 10-minute daily huddle agenda for a [clinic type]: the standing items, the data to glance at, and a slot for safety flags. Tight enough to actually finish in 10 minutes. Add the three things teams most often forget to cover.
39. Draft an onboarding checklist for new staff
Create an onboarding checklist for a new [role] in our [setting]. Group by: before day one, first week, first month. Include systems access, mandatory training, key policies, and the people to meet. Make each item a yes/no to track.
7. Quality, Safety & Protocols
Improvement work that depends on structure. AI drafts the framework; clinical and safety judgment stays with the team.
40. Draft a protocol from a guideline
Turn this guideline summary into a local clinical protocol skeleton for [setting]: [paste]. Structure: scope, steps, responsibilities, decision points, and documentation. Leave every clinical threshold, dose, and criterion as a placeholder for the clinical lead to set and verify. Mark where local adaptation is needed.
41. Structure an incident analysis
Help me structure a review of this de-identified safety event: [paste]. Use a systems lens: contributing factors across people, process, equipment, and environment, not individual blame. Produce the questions to ask and a template for findings and actions. The conclusions are the team's.
42. Build a safety checklist
Draft a safety checklist for [process / procedure] in the style of a pre-procedure timeout: the critical checks, who confirms each, and the stop conditions. Keep it short enough to be used every time, not filed and forgotten.
43. Draft an audit plan
Design a clinical audit for [topic]: the standard and its source [I will confirm], the measurable criteria, the sample and method, and how to present findings for change. Keep the criteria objective and tied to the standard.
44. Summarize audit results into an action plan
Turn these audit results into an action plan: [paste de-identified results]. For each gap: the issue, the likely cause, the proposed change, the owner, and how we will know it worked. Prioritize by patient-safety impact.
45. Draft a root-cause-analysis prompt set
Give me a structured set of questions to run a root-cause analysis on [type of recurring problem], moving from symptom to system cause. Group the questions by stage so a team can work through them in a meeting. Neutral, non-blaming language.
8. Professional Development & Teaching
AI is a patient tutor and a tireless teaching assistant. These prompts make it earn its keep.
46. Brief me on developments in my field
Brief me on significant developments in [specialty / topic] over the [last quarter]. For each: what changed, why it matters in practice, and one source to read to go deeper. If something is more hype than evidence, say so. I will verify before changing any practice.
47. Teach me an adjacent topic
I work in [your field]. I need to understand [adjacent topic] well enough to recognize when it is relevant and refer appropriately. Teach it in one sitting: the core concept, what matters in practice, the red flags, and the signals that mean "this needs a specialist."
48. Create a teaching case outline
Draft a teaching case outline on [topic] for [learner level], using a fully fictional, non-identifiable scenario. Include: learning objectives, the case progression with discussion points, the key teaching messages, and questions to check understanding. Keep it clearly fictional and educational.
49. Turn CME notes into a summary
I took these rough notes at a [CME / conference] session: [paste]. Turn them into a clean summary: five key takeaways, the open questions I still have, and a short list of changes to consider in my practice, each marked as "to verify" until I confirm it.
50. Draft a presentation outline
Help me outline a [grand rounds / team teaching] talk on [topic] for [audience], using only non-identifiable examples. Give me: a clear structure, the three messages the audience should leave with, the evidence I need to gather for each, and two discussion questions. I will supply and verify the clinical content.
How to actually use these
A prompt you run once and discard is worth little. A prompt that becomes part of your weekly routine compounds. Three moves separate clinicians who get marginal value from AI from those who get real hours back:
Save them. Every prompt that returns a usable draft goes into a personal prompt library tagged by setting and task. Next time you tune rather than rewrite.
Score them. When a prompt returns something weak, the problem is almost always the prompt, not the model. Use the six criteria that make a good prompt to rewrite: clarity, specificity, context, structure, examples, role. One iteration usually doubles output quality.
Version them. The first time a prompt works cleanly across several encounters, it becomes a template. Pin the version that works and keep the old ones, so you can see what changed when a guideline or a model updates. This is prompts as infrastructure applied to clinical work, which is exactly where it belongs.
The clinicians saving the most time are not using more AI. They are using fewer, better, reusable prompts for the documentation and administration that drain their day, they protect patient data without exception, and they verify every clinical fact before it touches a patient.
Sources
[1] Doximity, 2026 State of AI in Medicine: AI adoption among US doctors rose to 63%, up from 47% a year earlier. https://www.doximity.com/
