Prompt Writing 101: A Framework for Ask with Dipu Patel

Hello, my name is Dipu Patel, and I've been a PA for 26 years. I've spent many of those years working at the intersection of clinical practice and AI in healthcare, teaching clinicians and students how to use AI tools without losing what makes our work ours.

In the next few minutes, I'm going to share a prompt engineering framework that I teach clinicians and students.

Before we begin, let's talk about why prompt engineering matters. AI isn't magic or a replacement for clinical judgment. It's a drafting assistant. Like any assistant, if you give it vague instructions, you'll get vague answers. A structured framework helps you communicate more clearly so the AI can produce more useful drafts while you remain responsible for reviewing and applying your clinical reasoning.

The framework I teach consists of five parts: role, task, context, format, and constraints.

The role gives the AI perspective. The task explains what you want it to do. The context provides the patient information and relevant clinical details. The format specifies how the response should be organized, such as a SOAP note or patient discharge instructions. Finally, the constraints define what information the AI can use, the tone, and any word limits.

Let's see this framework in action using Doximity Ask.

I also want to point out that Doximity Ask is HIPAA compliant, as indicated by the badge in the upper right corner.

Our first example is a family medicine follow up visit.

Using the role, task, context, format, and constraints framework, I tell the AI that it is assisting a family medicine clinician. I ask it to draft a concise follow up SOAP note for a 50 year old woman with hypertension and type 2 diabetes using only the information I provide. I specify that the output should be in plain text and that it should not invent labs, findings, or diagnoses.

When I click Ask, Doximity analyzes the prompt and generates a concise SOAP note with the sections I requested. It includes the patient's medications, blood pressure readings, vital signs, labs, and follow up plan exactly as instructed.

This is a great example of how a well structured prompt produces a concise draft that I can quickly review and edit.

Now let's look at a pediatric example using templates.

From the Templates section in Doximity Ask, I create a reusable pediatric documentation template. The template follows the same framework but includes more detailed instructions specific to pediatric outpatient notes.

I define universal documentation rules, pediatric specific considerations, note sections, and clarification mode. Clarification mode allows Doximity Ask to generate a draft note first, then identify up to five important questions that would improve safety, counseling, or follow up.

After saving the template, I insert a pediatric case involving a seven year old with fever, cough, and congestion.

Using the template, Doximity Ask generates a structured outpatient note with the chief complaint, history of present illness, review of systems, physical exam, assessment, and plan.

It then provides clarification questions, including hydration status, work of breathing, sick contacts, medication allergies, and caregiver understanding.

After I answer those questions, Doximity Ask automatically updates the note, incorporating the additional information into the history, assessment, and plan.

The result is a more complete draft that I can review, edit, and then copy into my electronic health record while maintaining control over the final documentation.