The AI Prompt Formula Every Clinician Should Know

# EHR

# EMR

# documentation

# prompts

Vague prompts produce vague notes. Here is the four-part structure that gets EHR-ready output every time, plus the five mistakes to avoid.

Product Doximity Ask

Level Beginner

Time 5 min read

Specialty All clinicians

The difference between a useful AI output and a generic, placeholder-filled draft almost always comes down to the prompt. Vague prompts produce vague notes. Specific prompts produce output that is ready to copy into your EHR with minimal editing.

This is true regardless of which tool you use, but it is especially relevant in Doximity Ask, where the conversational format lets you refine your prompt mid-session and layer in clinical context across multiple turns.

The 4-Part EHR Prompt Formula

Every strong AI prompt for clinical documentation has four components. Missing any one of them is usually why output requires extensive editing.

01 Role and EHR

Specify your specialty and name your EHR. Example: "I am an internist using Epic EHR." This single addition changes how the output is structured, what section headers are used, and how the note is formatted.

02 Task

State what you need: note type, document type, or code lookup. Example: "Write a progress note" or "Generate MDM documentation" or "Look up ICD-10 code."

03 Clinical Context

Include age, sex, diagnosis, and relevant clinical details. Do not include identifying information (name, DOB, MRN). Example: "45yo male, T2DM, A1c 10.2, refusing metformin."

04 Format

Tell AI how to structure the output for your EHR. Example: "Use SOAP format. Order A&P by clinical acuity. Optimize for copy-paste into Epic."

What Vague vs. Specific Looks Like

Consider two prompts for the same clinical scenario: an 18-year-old admitted with cough and dyspnea on exertion.

Vague: "Write a SOAP note for a patient with cough and shortness of breath." Result: generic placeholders, no EHR formatting, requires extensive editing.

Specific: "I am a hospitalist using Epic. Write an H&P for an 18yo male admitted for cough and dyspnea on exertion, likely CAP vs. atypical pneumonia. Include MDM with problem, data, and risk elements. Order A&P by acuity. Optimize for moderate-complexity E/M billing." Result: MDM paragraph with all three elements, formatted for Epic, ready to copy.

The upgrade formula is simple: add your specialty, your EHR name, and your coding level goal to any prompt you already use.

5 Mistakes to Avoid

Most early frustrations with AI in clinical documentation trace back to a small set of avoidable mistakes.

Pasting AI output without reading it. AI can generate plausible-sounding but incorrect drug doses, lab values, or clinical facts. Every AI-generated note requires your review before signing.

Including PHI in a non-HIPAA-compliant tool. Sending identifiable patient data to a consumer AI tool (such as the free version of ChatGPT) is a HIPAA breach. Use de-identified clinical details, or use a HIPAA-compliant tool like Doximity Ask.

Writing vague prompts for complex notes. The more complex the note, the more specific your prompt needs to be. A hospitalist note for a patient with five active problems needs context, structure, and format instructions.

Never verifying coding suggestions. AI can suggest ICD-10 and CPT codes confidently and incorrectly. Always verify against your documentation before submitting.

Signing notes that "look fine" without reading. Rushing through AI-generated content is a legal and clinical risk. Read every sentence before signing.

Your First Week: A 5-Day Plan

The best way to build confidence with AI in clinical documentation is to start with one task, compare the output to your usual work, and expand from there. Here is a simple plan for your first week using Doximity Ask.

Day 1 (Monday): Use AI to draft one progress note from your bullet points. Compare the output to your usual note. Note what needed editing.

Day 2 (Tuesday): Ask AI for MDM documentation on your most complex patient. Review the three elements (problems, data, risk) against the visit.

Day 3 (Wednesday): Use AI to draft a discharge summary or consult note. Apply the 4-part prompt formula and notice the difference in output quality.

Day 4 (Thursday): Create one SmartPhrase (or equivalent) for your most common diagnosis. Load it into your EHR.

Day 5 (Friday): Look up the 5 ICD-10 codes you use most often. Save them as a quick-reference note.

KEY Your EHR remains the system of record. AI helps you move faster; it does not replace clinical judgment, legal responsibility, or your review of every note before it is signed.