Based on: Liu J, Liu F, Wang C, Liu S. Prompt Engineering in Clinical Practice: Tutorial for Clinicians. Journal of Medical Internet Research. 2025;27:e72644. DOI:10.2196/72644. (JMIR)

---

Artificial Intelligence Is Only as Good as the Question We Ask

Large language models (LLMs) have rapidly become part of everyday clinical practice. They summarize patient encounters, generate documentation, explain guidelines, assist with literature reviews, and even support differential diagnosis generation.

Yet one fact is becoming increasingly clear:

The quality of an AI-generated response is determined largely by the quality of the prompt.

This concept—known as prompt engineering—is emerging as one of the most important digital competencies for modern clinicians. Rather than treating AI as a search engine, physicians must learn to communicate with it as a structured clinical collaborator.

The excellent tutorial by Liu et al. provides a practical framework for clinicians to build better prompts that improve accuracy, relevance, efficiency, and safety while maintaining physician oversight. (PMC)

---

What Is Prompt Engineering?

Prompt engineering is the process of designing instructions that help an AI model understand:

• your objective
• the clinical context
• the desired reasoning process
• the preferred output format
• the limitations it should follow

Think of it this way:

Poor prompt

“Tell me about heart failure.”

↓

Generic answer.

Better prompt

“You are a heart failure cardiologist. Summarize the 2022 AHA/ACC/HFSA Heart Failure Guideline for a PGY-2 Internal Medicine resident. Focus on HFrEF guideline-directed medical therapy, medication sequencing, contraindications, and monitoring. Present the response as a table.”

↓

Focused, clinically useful answer.

---

The Core Prompting Styles

1. Zero-Shot Prompting

The simplest form.

No examples are provided.

Example:

What are the causes of hyponatremia?

Best for:

• quick facts
• definitions
• guideline lookups
• medical education

---

2. One-Shot Prompting

Provide one example before asking the model to complete a similar task.

Example:

Here is an example discharge summary…

Now generate one for this patient.

Useful for:

• discharge summaries
• patient instructions
• clinic letters
• operative reports

---

3. Few-Shot Prompting

Instead of one example, provide several.

Example:

Example 1
SOAP note...

Example 2
SOAP note...

Generate another note using the same style.

Benefits:

✔ More consistent formatting

✔ Better documentation quality

✔ Reduced hallucinations

---

4. Role Prompting

Assign the AI a professional role.

Examples:

“You are a cardiologist.”

“You are a radiologist.”

“You are an infectious disease consultant.”

“You are an evidence-based medicine educator.”

Role prompting helps the model tailor vocabulary, reasoning, and recommendations to the intended audience.

---

5. Contextual Prompting

Clinical context dramatically improves output.

Instead of asking:

How should hypertension be treated?

Provide context:

A 72-year-old man with diabetes, CKD stage 3, resistant hypertension, bradycardia, and heart failure with reduced ejection fraction presents for medication optimization.

Now the response becomes patient-specific instead of generic.

---

6. Chain-of-Thought Prompting

For complex clinical reasoning, instruct the model to reason through the problem step by step.

Example:

Explain your diagnostic reasoning step-by-step before giving the final differential diagnosis.

This approach improves performance on complicated clinical tasks, although clinicians should independently verify every recommendation before applying it to patient care. (PMC)

---

7. Self-Consistency Prompting

Ask the model to generate multiple independent reasoning paths before selecting the most consistent answer.

This is particularly valuable for:

• diagnostic dilemmas
• guideline interpretation
• evidence synthesis

---

8. Meta Prompting

Instead of directly asking the clinical question…

Ask the AI to improve the prompt itself.

Example:

Improve this prompt to obtain the most accurate evidence-based answer.

This often produces substantially better responses.

---

The Anatomy of a High-Quality Clinical Prompt

An effective clinical prompt typically includes five components:

1. Role

Who is the AI?

You are a board-certified cardiologist.

---

2. Objective

What should it accomplish?

Summarize the patient’s heart failure management.

---

3. Clinical Context

Include relevant information.

• age
• symptoms
• laboratory values
• medications
• imaging
• comorbidities

---

4. Constraints

Limit the response.

Examples:

• use only ACC/AHA guidelines
• avoid speculation
• identify uncertainty
• cite supporting evidence

---

5. Output Format

Specify exactly how the answer should appear.

Examples:

✔ SOAP note

✔ Bullet points

✔ Table

✔ Differential diagnosis

✔ Patient education handout

✔ Clinical decision tree

---

Clinical Applications

Prompt engineering has applications throughout healthcare.

Examples include:

• Clinical documentation
• Medical decision support
• Differential diagnosis generation
• Evidence synthesis
• Guideline interpretation
• Coding assistance
• Patient education
• Prior authorization summaries
• Literature review
• Quality improvement initiatives

The tutorial emphasizes that prompt engineering is not limited to documentation—it can support many cognitive workflows when used responsibly and with clinician oversight. (PMC)

---

Best Practices for Clinicians

✔ Be specific.

✔ Provide sufficient clinical context.

✔ Clearly define the desired output.

✔ Ask the AI to explain uncertainty.

✔ Request evidence when appropriate.

✔ Iteratively refine prompts.

✔ Never upload protected health information into systems that are not approved for clinical use.

✔ Always validate AI-generated content before incorporating it into patient care.

---

The Future of Clinical AI

Prompt engineering is becoming as important as literature searching or evidence appraisal.

Tomorrow’s clinicians will not simply know how to use AI.

They will know how to communicate with AI effectively.

Those who master prompt engineering will obtain more accurate clinical documentation, more relevant educational content, better research assistance, and more efficient workflows—while continuing to apply the critical thinking, judgment, and accountability that only trained healthcare professionals can provide.

---

Key Takeaway

Prompt engineering is more than a technical skill—it is rapidly becoming a core clinical competency. As highlighted by Liu et al., combining structured prompts with thoughtful clinician oversight allows AI to enhance documentation, education, evidence synthesis, and decision support while preserving patient safety and professional judgment. AI should augment—not replace—clinical reasoning. (JMIR)

---

Reference

Liu J, Liu F, Wang C, Liu S. Prompt Engineering in Clinical Practice: Tutorial for Clinicians. Journal of Medical Internet Research. 2025;27:e72644. DOI:10.2196/72644. (JMIR)

This article paraphrases and synthesizes concepts from the original publication and does not reproduce its copyrighted text.