Physicians spend more time on documentation than most people outside medicine realize. A full-time attending can spend two or more hours every clinical day writing and editing notes, composing referral letters, filling out prior authorization forms, generating patient education materials for complex discharges, and drafting committee reports for M&M or quality improvement. That time comes directly out of patient care, preparation, and recovery. It is not a new problem. But AI-assisted drafting has become genuinely useful for the documentation layer — and it is worth knowing exactly what it can and cannot do before you build it into your workflow.
This post covers the MD-specific documentation pipeline: SOAP and H&P note drafting, patient education handouts, specialist referral letters, prior authorization narratives, peer review and committee writing, and physician-to-physician correspondence. Two adjacent posts on this blog cover related ground: Claude AI for nurses and healthcare workers owns bedside documentation, nursing workflow, care coordination notes, and the documentation tasks specific to nursing practice; Claude AI for dentists owns oral health documentation, treatment plan writing, and dental-specific patient education. This post owns the physician pipeline — the H&P, the referral letter, the prior auth narrative, the M&M write-up. Those are distinct workflows from nursing documentation and dental practice, and they warrant a separate treatment.
This post is written for physicians, MDs, DOs, hospitalists, attending physicians, residents, and fellows. It is also directly applicable to physician-led practices managing their own documentation. Read the limits section first — this audience is rightly liability-conscious, and the use cases only make sense in the context of what Claude cannot do.
What Claude Cannot Do for Physicians
These are hard limits. Read them before using any prompt in this post.
- No access to EHR or EMR systems: Claude has no connection to Epic, Cerner, Meditech, Athenahealth, eClinicalWorks, DrChrono, or any other EHR/EMR. It cannot pull patient charts, write directly to the medical record, read encounter data, or interface with clinical documentation workflows. Every piece of information Claude works with must be manually provided in the prompt.
- No access to patient records, labs, imaging, or vitals: Claude sees only what you type. It has no access to lab values, imaging reports, vital signs, medication administration records, or any part of a patient's medical record. It cannot retrieve, summarize, or analyze actual patient data.
- No real-time clinical decision support or drug interaction checking: Claude is not a clinical decision support tool. It has no access to real-time drug interaction databases, formulary systems, allergy checking, dosing calculators, or point-of-care clinical tools. Do not rely on Claude for medication dosing, drug interactions, or clinical decision-making.
- No ICD-10 or CPT coding: Claude can reference diagnostic terminology, but it has no access to current ICD-10 or CPT code sets and cannot provide billing codes for clinical encounters, diagnoses, or procedures. Code assignment requires a qualified coder and/or your EHR coding tools.
- No prescription authority or medication verification: Claude cannot prescribe, verify prescriptions, confirm dosing accuracy, or check against a patient's current medication list. Any medication references in Claude's output must be independently verified by the prescribing physician.
- Not HIPAA-compliant for real patient data: Do not paste real patient names, dates of birth, medical record numbers, diagnoses tied to specific patients, or any other protected health information (PHI) into Claude. Use de-identified or fictional encounter details in every prompt. Claude provides structure; you supply verified, real clinical data at sign-off.
- No access to clinical guidelines or evidence databases: Claude cannot query UpToDate, DynaMed, ACP guidelines, specialty society guidelines, or any real-time clinical evidence source. It may reference guidelines from its training data — always verify clinical recommendations against current authoritative sources before relying on them.
- No insurance eligibility or coverage verification: Claude has no access to insurance eligibility systems, payer portals, formulary databases, or real-time prior authorization status. For authorization decisions, use your practice management system and payer-specific tools.
What Claude does for physicians: drafts the language, structure, and formatting of clinical documentation. The clinical content, medical judgment, verification, and sign-off are yours.
6 Use Cases for Claude AI in Physician Practice
1. SOAP and H&P Clinical Note Drafting
The documentation burden per patient encounter is well understood by every physician reading this. SOAP notes, H&P notes, progress notes — the structure is standardized, but generating complete, well-organized note language from encounter shorthand takes time. Claude takes a bulleted encounter summary (chief complaint, HPI, exam findings, assessment, plan — no real patient data) and produces a complete structured note with all standard sections, vitals and lab placeholders clearly marked, and numbered problem lists in the Assessment. It is a first draft. You review, edit with verified clinical data, and sign.
Draft a complete SOAP note (or H&P if indicated below) for a clinical encounter
using the details I provide. Use no real patient data — this is a documentation
structure exercise using de-identified or fictional encounter details.
Output format:
SUBJECTIVE
Chief Complaint: [one line]
HPI: [narrative paragraph — onset, location, duration, character, associated
symptoms, relieving/aggravating factors, radiation, timing]
Review of Systems: [pertinent positives and negatives by system]
Medications: [PHYSICIAN TO INSERT — do not fabricate]
Allergies: [PHYSICIAN TO INSERT]
PMH / PSH / FH / SH: [structured list with physician-to-complete placeholders
for specific details]
OBJECTIVE
Vital Signs: [INSERT ACTUAL VITALS — temperature, BP, HR, RR, SpO2, weight]
Physical Exam: [organ system by system — findings as provided, mark
[PHYSICIAN TO VERIFY] on any exam findings not explicitly provided]
Relevant Labs / Imaging: [INSERT RESULTS — do not fabricate values]
ASSESSMENT
Numbered problem list with working diagnosis or differential for each problem.
Format each as: #[N]. [Problem / Diagnosis] — [brief supporting rationale
based on subjective and objective]
Mark each diagnosis [WORKING DIAGNOSIS — PHYSICIAN TO CONFIRM]
PLAN
Per problem, numbered to match Assessment:
#[N]. [Problem]
Workup: [labs, imaging, consults ordered or planned]
Medications: [PHYSICIAN TO VERIFY ALL DOSING AND INTERACTIONS]
Referrals: [specify]
Patient education: [key points discussed]
Follow-up: [timeframe and triggers for earlier return]
Note type: [SOAP / H&P]
Encounter type: [outpatient office visit / inpatient admission / ED / follow-up]
Specialty / clinical context: [e.g., general internal medicine / cardiology / urgent care]
Chief complaint: [one to two sentences — no real patient identifiers]
HPI details: [de-identified encounter details]
Exam findings: [as available — mark any gaps]
Working assessment: [your working diagnoses or differential]
Plan elements: [what you intend to order, prescribe, refer, or follow]
[VERIFY ALL CLINICAL DETAILS BEFORE SIGNING]
[DO NOT PASTE REAL PATIENT DATA INTO CLAUDE]
[PHYSICIAN MUST REVIEW AND SIGN BEFORE ENTERING INTO EHR]
[ALL MEDICATION REFERENCES MUST BE INDEPENDENTLY VERIFIED BY THE
PRESCRIBING PHYSICIAN]
The output gives you a fully structured note with clear placeholders wherever actual clinical data must be inserted by the physician. It is not a chart-ready document — it is a structured draft that eliminates the blank-page problem and lets you focus review time on clinical content rather than formatting. Every note goes through physician review and sign-off before it enters the EHR. That requirement does not change because a draft was faster to produce.
2. Patient Education Materials
Explaining a new diagnosis, a procedure, or a complex medication regimen to a patient in the 8 minutes available at end of a busy clinic visit is a known challenge. A clear, jargon-free handout the patient can take home, reference later, and share with a family caregiver extends the impact of that conversation significantly. Writing those handouts from scratch for every condition is not realistic. Claude takes a diagnosis or procedure name, a reading level preference, and the key clinical points you want covered — and produces a structured, plain-language patient handout ready for your review.
Write a patient education handout for the diagnosis or procedure described below.
Do not include any real patient information — this is a general educational template.
Reading level target: [6th grade / 8th grade / general adult (10th grade equivalent)]
Structure the handout with the following sections:
WHAT IS [CONDITION / PROCEDURE]?
Plain-language explanation — no jargon. Define any medical terms used.
WHAT CAUSES IT?
Common causes or risk factors — concise, non-alarming where appropriate.
WHAT YOUR DOCTOR FOUND
[PHYSICIAN TO COMPLETE — describe relevant exam findings or test results
in plain language. Do not fabricate clinical findings.]
WHAT YOUR TREATMENT PLAN INVOLVES
[PHYSICIAN TO COMPLETE — describe specific treatment plan in plain language.
Claude provides structure; physician inserts actual plan.]
WHAT TO WATCH FOR — WHEN TO CALL YOUR DOCTOR
Red flag symptoms that warrant an urgent call or return visit.
Specify: "Call us if you experience..." and "Go to the ER immediately if..."
WHEN TO FOLLOW UP
[PHYSICIAN TO INSERT — timeframe and scheduling instructions]
QUESTIONS TO ASK AT YOUR NEXT VISIT
3–5 suggested patient questions that promote engagement with the care plan.
Condition or procedure: [name — e.g., Type 2 Diabetes / Knee Arthroscopy /
Atrial Fibrillation / New Anticoagulation Therapy]
Key clinical points to cover: [list 4–5 specific points you want the handout
to address — e.g., lifestyle modifications, medication timing, dietary changes,
activity restrictions, monitoring requirements]
Anything to emphasize or de-emphasize: [optional — e.g., "emphasize that this
is manageable with adherence" or "avoid alarming language about prognosis"]
[VERIFY AGAINST CURRENT CLINICAL GUIDELINES BEFORE DISTRIBUTING]
[REVIEW BEFORE DISTRIBUTING TO PATIENTS]
[DO NOT USE AS A SUBSTITUTE FOR THE INFORMED CONSENT CONVERSATION]
[PHYSICIAN TO COMPLETE ALL [PHYSICIAN TO COMPLETE] FIELDS WITH
ACTUAL CLINICAL DETAILS]
The physician education handout prompt pairs well with complex discharge conversations and new diagnosis counseling. Preparing the handout before the patient encounter — 10 minutes of prompt work the day before — means the conversation can be more focused, and the patient leaves with something specific to read rather than a generic printout. If you are looking for prompts that translate directly to clinical workflow efficiency, patient education drafting is consistently one of the highest-leverage applications.
3. Specialist Referral Letter
A well-written referral letter communicates the clinical question clearly, provides the relevant history concisely, includes current medications and allergies, and makes the urgency level explicit. A poorly written referral generates a callback, delays the appointment, and occasionally results in the specialist seeing the patient without the information they needed. The structure of a good referral letter is not complicated — executing it consistently under time pressure is what creates friction. Claude takes your patient summary (de-identified) and produces a complete, formally structured referral letter in the format receiving specialists prefer.
Draft a formal specialist referral letter using the information below.
Use de-identified or placeholder patient information — no real patient PHI.
Structure the letter with the following sections:
HEADER: [Date | Your name, credentials, and practice | Referring to: specialist
name and specialty]
PATIENT DEMOGRAPHICS: [Patient initials or placeholder | DOB placeholder |
Insurance: PHYSICIAN TO INSERT | Primary diagnosis code: PHYSICIAN TO INSERT]
DEAR DR. [SPECIALIST]:
REASON FOR REFERRAL: One to two sentence statement of why you are referring
and what clinical question you want the specialist to address.
RELEVANT HISTORY: Concise clinical background — onset, relevant prior workup,
pertinent PMH, surgical history relevant to the referral.
PERTINENT EXAM AND TEST FINDINGS: Key findings that prompted the referral —
physical exam, lab values [PHYSICIAN TO INSERT ACTUAL VALUES],
imaging findings [PHYSICIAN TO INSERT ACTUAL RESULTS].
CURRENT MEDICATIONS: [PHYSICIAN TO INSERT — Claude will not fabricate
a medication list]
ALLERGIES: [PHYSICIAN TO INSERT]
CLINICAL QUESTION FOR THE SPECIALIST: Explicit statement of what you want
the specialist to evaluate, recommend, or manage.
URGENCY: [Routine (schedule within [timeframe]) / Semi-urgent (within 2 weeks) /
Urgent (within 48–72 hours) / Emergent — specify]
ADDITIONAL NOTES: [Optional — any specific requests, preferred follow-up
communication method, or context the specialist should know]
CONTACT: [Your practice phone, fax, and preferred contact for specialist response]
Closing and signature block.
Referring specialty / context: [e.g., Internal Medicine referring to Cardiology]
Receiving specialty: [e.g., Cardiology / Pulmonology / Neurology]
De-identified reason for referral: [brief clinical description — no real PHI]
Relevant history highlights: [key clinical points]
Key findings to include: [use placeholders for actual values]
Urgency: [routine / semi-urgent / urgent]
Specific question for the specialist: [what you want them to address]
[VERIFY ALL CLINICAL DETAILS BEFORE SENDING]
[PHYSICIAN TO INSERT ALL LAB VALUES, IMAGING RESULTS, AND MEDICATION LIST]
[REVIEW WITH REFERRING PHYSICIAN BEFORE TRANSMITTING]
The referral letter prompt eliminates the structural work and lets you focus on the clinical substance. The format above — especially the explicit "Clinical Question for the Specialist" section — produces letters that tend to reduce callback volume because the specialist knows exactly what is being asked before they open the chart. Pharmacists managing transitions of care have a related but distinct documentation workflow; the referral letter format here is specific to physician-to-specialist communication.
4. Prior Authorization Documentation
Prior authorization narratives are among the most time-consuming and frustrating documentation tasks in clinical practice. The clinical case for a medication or procedure is usually straightforward. Getting it into the structured narrative format that moves through insurance review requires specific language: medical necessity statement, diagnostic background, failed alternatives where applicable, supporting clinical criteria, urgency notation. Claude takes your clinical rationale and produces a complete prior auth narrative structured for payer review.
Draft a prior authorization narrative for the medication or procedure described below.
Use de-identified clinical information — no real patient PHI.
Structure the narrative as follows:
PATIENT PROFILE (DE-IDENTIFIED): [Age range / gender / relevant diagnoses —
no identifying information]
REQUESTED MEDICATION OR PROCEDURE: [Name, dose/frequency if medication,
or procedure description]
DIAGNOSIS AND ICD-10 CODE: [Diagnosis — PHYSICIAN TO CONFIRM ICD-10 CODE]
MEDICAL NECESSITY STATEMENT:
Clear statement that the requested treatment is medically necessary for this
patient based on diagnosis, clinical findings, and treatment goals.
Use objective clinical language appropriate for payer review.
CLINICAL BACKGROUND:
Concise clinical history supporting the request — relevant diagnosis timeline,
pertinent exam findings [PHYSICIAN TO INSERT ACTUAL FINDINGS],
relevant lab or imaging results [PHYSICIAN TO INSERT ACTUAL VALUES].
PRIOR AND FAILED ALTERNATIVES (if applicable):
Structured list of treatments already attempted, duration, reason for
discontinuation or failure. Format: [Treatment] — [Duration] — [Outcome/Reason
for change]. Mark [NOT APPLICABLE — FIRST-LINE TREATMENT] if no alternatives
are required.
SUPPORTING CLINICAL CRITERIA:
Reference applicable clinical guidelines or payer criteria language supporting
the request. Mark all citations [VERIFY AGAINST CURRENT PAYER CRITERIA AND
CLINICAL GUIDELINES — DO NOT SUBMIT WITHOUT VERIFICATION].
URGENCY NOTATION: [Routine / Expedited review requested — reason: specify]
ORDERING PHYSICIAN ATTESTATION BLOCK: [Structure only — physician signs]
Requested item: [medication name + dose/frequency / procedure name]
Diagnosis: [de-identified clinical description]
Clinical rationale: [why this treatment is appropriate for this patient —
de-identified]
Failed alternatives: [list if applicable — drug name, duration, reason failed]
Urgency: [routine / expedited]
Payer: [optional — if known, helps Claude use appropriate framing language]
[VERIFY ALL CLINICAL CONTENT AGAINST ACTUAL PATIENT RECORDS BEFORE SUBMITTING]
[VERIFY AGAINST PAYER-SPECIFIC CRITERIA — CRITERIA VARY BY PAYER AND PLAN]
[DO NOT SUBMIT WITHOUT PHYSICIAN REVIEW AND SIGNATURE]
[ICD-10 AND CPT CODES MUST BE CONFIRMED BY PHYSICIAN OR CODING STAFF]
The prior auth narrative prompt is most useful when paired with a clear statement of your clinical rationale upfront. The clearer your rationale in the prompt, the more persuasive the medical necessity language in the output. The structure — medical necessity, failed alternatives, supporting criteria, urgency — is the format that payer reviewers are working from on their end. Matching that structure reduces the back-and-forth that slows approvals.
5. Peer Review and Committee Writing
Morbidity and mortality conference write-ups, peer review case summaries, and quality improvement committee reports require a specific kind of prose: objective, chronological, blame-free, and scrupulously free of conclusions that haven't been through the review process. Writing that kind of report under time pressure — especially when you are also the attending involved in the case — is genuinely difficult. Claude takes structured case inputs and produces a complete committee-ready document in appropriate language.
Draft a peer review case summary / M&M conference write-up / quality improvement
committee report using the case information below.
Use de-identified case details — remove all identifying patient information before
providing inputs.
Specify document type: [Peer Review Case Summary / M&M Conference Write-Up /
QI Committee Report]
Structure the output as follows:
CASE SUMMARY (DE-IDENTIFIED):
Patient profile (de-identified): age range, gender, relevant diagnoses.
Presenting complaint and reason for encounter.
CLINICAL TIMELINE:
Chronological narrative of care events — admission/presentation through
discharge or outcome. Objective, factual, no editorializing.
Format: [Date/Time] — [Event / Clinical action / Finding]
CARE DECISIONS REVIEWED:
Structured list of the specific decisions or events under review. Objective
description only — what was decided, ordered, or done, and when.
No conclusions, no judgment language.
OUTCOME:
Factual statement of clinical outcome.
FINDINGS AND OBSERVATIONS:
Neutral, objective observations for committee discussion.
Format: [Observation] — [Relevant context]. Do not state conclusions.
Mark: [FOR COMMITTEE DISCUSSION — NOT A FINDING OF FAULT OR ERROR]
RECOMMENDED DISCUSSION POINTS:
3–5 questions or topics for the committee to address. Framed as open questions,
not predetermined conclusions.
SYSTEMS / PROCESS CONSIDERATIONS:
Any systems, workflow, or process factors relevant to committee review.
Objective only.
Document type: [specify]
De-identified case description: [clinical summary — remove all PHI before
providing this input]
Timeline of relevant events: [chronological list]
Specific events or decisions under review: [what the committee is reviewing]
Outcome: [factual clinical outcome]
[REVIEW WITH DEPARTMENT CHIEF OR QUALITY OFFICER BEFORE SUBMITTING]
[ALL IDENTIFYING INFORMATION MUST BE REMOVED BEFORE PROVIDING INPUT TO CLAUDE]
[DO NOT INCLUDE IDENTIFYING INFORMATION IN COMMITTEE SUBMISSIONS WITHOUT
FORMAL DE-IDENTIFICATION REVIEW PER INSTITUTIONAL POLICY]
[COMMITTEE DOCUMENT — NOT FOR DISTRIBUTION OUTSIDE PROTECTED REVIEW PROCESS]
The peer review and M&M prompt is particularly useful because the hardest part of that writing — maintaining objective, blame-free language while covering difficult clinical events — is where Claude's structured output discipline is most valuable. The output is a first draft for department chief or quality officer review. It is not a final committee document, and it does not leave the draft stage without that review step.
6. Physician-to-Physician and Administrative Correspondence
Physician correspondence covers a range of document types, each with different format and tone requirements: clinical update letters to PCPs after specialist visits, cross-cover handoff notes, responses to medical records requests, disability and FMLA support letters, and department or hospital administrative memos. Running each of these as a separate prompt is one approach; the prompt below generates structured templates for all five types in a single session, which is more efficient when you are working through a documentation backlog.
Generate physician correspondence templates for all five document types listed below.
Use de-identified or placeholder information throughout — no real patient PHI.
For each template, produce a complete, ready-to-edit structure with physician-to-complete
placeholders clearly marked.
---
TEMPLATE 1: CLINICAL UPDATE LETTER TO PCP (POST-SPECIALIST VISIT)
Format: formal letter on practice letterhead structure
Sections: date / to / from / re: [patient initials + DOB placeholder] /
reason for visit / findings summary / diagnosis or differential /
treatment plan / medications added or changed [PHYSICIAN TO VERIFY] /
follow-up plan / specific action requested of PCP (if any) / closing
Tone: collegial, concise, clinical
---
TEMPLATE 2: CROSS-COVER / HANDOFF NOTE
Format: structured clinical handoff — not a letter
Sections: patient identifier [PLACEHOLDER] / primary team / covering physician /
active diagnoses / relevant overnight events / pending items (labs, consults,
procedures) / medications requiring attention / anticipated issues and contingency
guidance / escalation criteria (call vs. come in) / contact information
Tone: direct, specific, action-oriented
---
TEMPLATE 3: RESPONSE TO MEDICAL RECORDS REQUEST (STRUCTURE ONLY)
Format: formal letter — structure and language only, no actual records
Sections: date / requestor information / patient identifier [PLACEHOLDER] /
acknowledgment of request / scope of records being provided [PHYSICIAN TO
COMPLETE — DO NOT FABRICATE RECORD CONTENTS] / any limitations or redactions
with rationale / HIPAA release confirmation / contact for questions
[DO NOT FABRICATE OR INCLUDE ANY ACTUAL MEDICAL RECORD CONTENT]
Tone: formal, compliant, precise
---
TEMPLATE 4: LETTER SUPPORTING DISABILITY OR FMLA CLAIM
Format: formal physician support letter
Sections: date / to: [insurer / employer / FMLA administrator] / re: [patient
initials + DOB placeholder] / physician attestation of clinical relationship /
[PHYSICIAN TO COMPLETE: diagnosis and functional limitations — Claude provides
structure only, will not fabricate clinical opinions] /
[PHYSICIAN TO COMPLETE: expected duration of limitation] /
[PHYSICIAN TO COMPLETE: specific work or activity restrictions] /
physician signature block
[PHYSICIAN MUST COMPLETE ALL CLINICAL OPINION FIELDS — CLAUDE PROVIDES
STRUCTURE ONLY. DO NOT SUBMIT WITH PLACEHOLDER TEXT.]
[REVIEW WITH RISK MANAGEMENT IF REQUIRED BY INSTITUTION]
Tone: formal, clinically grounded, liability-appropriate
---
TEMPLATE 5: DEPARTMENT OR HOSPITAL ADMINISTRATIVE MEMO
Format: standard memo
Sections: to / from / date / re / background (2–3 sentences) /
current situation or issue / proposed action or recommendation /
timeline / requested response or decision / contact
Tone: professional, direct, administrative
---
For each template, mark every field requiring physician completion with:
[PHYSICIAN TO COMPLETE: description of what is needed]
The FMLA and disability letter template warrants specific emphasis: Claude provides structure and language scaffolding, and nothing else. The clinical opinions — diagnosis, functional limitations, duration, work restrictions — must be completed by the treating physician. The template output with placeholder text is not a submittable document. That flag is in the prompt and should be treated as a hard stop.
Why Claude Over ChatGPT for Physician Documentation
The Claude vs. ChatGPT comparison covers the general case. For physician documentation specifically, a few points are worth calling out directly.
Projects keep clinical context persistent. Claude's Projects feature lets you create a dedicated workspace per condition cluster, per service, or per documentation type — for example, a "Cardiology Outpatient" project or a "Prior Auth Documentation" project. Within a project, Claude maintains context across the session, so you are not re-establishing clinical terminology and format preferences with every new note.
The full H&P fits in one session. Claude's context window is large enough to hold a complete H&P, a prior records summary, a current medication list, and a plan — all at once. For complex patients, that means you can paste the clinical background once and generate multiple downstream documents (referral letter, patient education handout, prior auth narrative) without losing context.
Structured output discipline. SOAP format, referral letter headers, numbered problem lists, CSI-style committee report structures — Claude applies consistent formatting across every output. That matters when notes are going into EHR templates that expect specific structure, or when referral letters represent the practice to a specialist community.
Conservative on fabricated clinical citations. ChatGPT has a documented tendency to invent drug dosages, ICD codes, and guideline references with confidence. Claude is more conservative on fabricated citations — though it is not immune, and every clinical reference still requires independent verification. For physician documentation, that difference in behavior is meaningful.
No internet access is a feature, not a limitation — in this context. Claude cannot check real-time drug interactions, current formularies, or live clinical guidelines. That is a real limit, and it is why this post does not recommend Claude for clinical decision support. For documentation tasks, where you are drafting structure and language based on your own clinical judgment, the absence of internet access is less relevant than the structured drafting capability.
4 Practical Tips for Physicians Using Claude
1. One Project per condition cluster, not per patient. Set up Projects organized around documentation type or clinical context: "Internal Medicine SOAP Notes," "Prior Auth Cardiology," "Patient Education Handouts." This keeps Claude's context calibrated to the clinical register you are working in without creating per-patient data hygiene issues.
2. Never paste real patient data. This bears repeating because the temptation is real when you are moving fast. Use de-identified or fictional encounter details in every prompt. Claude drafts the structure and language; you insert the verified, real clinical data at sign-off. That workflow is also what makes the output defensible — you are not relying on Claude's interpretation of a real patient record.
3. Use the patient education prompt before complex discharge conversations. Preparing a plain-language handout before a new diagnosis conversation, a complex discharge, or a procedure consent takes 10 minutes. It tends to reduce the follow-up call burden from patients and families who did not retain the verbal explanation. Mental health counselors use a similar approach for psychoeducation materials — the underlying principle is the same regardless of specialty.
4. Every output is a first draft. Claude produces structured, well-formatted documentation language. It does not produce chart-ready, legally defensible clinical documents. Every note, letter, or report produced using these prompts requires physician review, clinical content verification, and sign-off before it enters a medical record, leaves the practice, or gets submitted to a payer or committee. The review step is not optional and is not abbreviated because the first draft was faster to produce.
Get the Complete Claude Playbook
The prompts in this post are a working foundation for physician documentation. The Complete Claude Playbook ($27) is the step-by-step guide to getting the most out of Claude AI across your entire workflow — from setting up Projects correctly to building reusable prompt systems for documentation, correspondence, and administrative tasks. It covers what to do when outputs need refinement, how to maintain consistent clinical register across a session, and how to build a documentation workflow that compounds over time.