[Tutorial Kit] New Tools for Public Health: ChatGPT

Shamini V De Silva
1 day ago
5 min read

🎬 Video Timestamps

0:00 - Intro

1:10 - 5-Part Prompt

1:50 - Prompt Chaining

2:04 - Proof Loop

🎯 Challenge

Decide the 1 top trending public health topic in the U.S. using a combination of AI and traditional public health prioritization processes.

3 Learning Objectives

By the end of the session, participants will be able to:

Understand the basics of large language models (LLM) including the concepts of "prompt engineering", "prompt chaining" and "proof loops".
Engineer Prompts using the AI tool: 🧰 🛠️ ChatGPT.
Collaborate with colleagues to prioritize 1 public health topic.

Prerequisites. Beginner-friendly, no prior experience needed

Keywords and core concepts covered:

Large Language Models
Prompt engineering
Prompt chaining
Proof loops

Overview

Artificial intelligence (AI) is rapidly reshaping how public health professionals work—but the real challenge is integrating AI tools into a field built on collaboration, ethics, and shared decision-making. Large language models (LLMs) like ChatGPT can process vast amounts of information quickly, but they don’t “understand” context the way humans do. Knowing how to interact with AI is essential. This interaction includes prompt engineering, the process of honing detailed, structured inputs; prompt chaining, asking followup questions; and proof loops, double-checking all outputs.

7 STEPS

Step 1: Set Up Your Tools
Step 2: Run a Basic Prompt
Step 3: Create an Engineered Prompt
Step 4: Test Prompt Chaining
Step 5: Experiment with Roles and Goals
Step 6: Create a Proof Loop with Multiple Queries
Step 7: Reflect and Refine

Core Concepts

Let's focus on 3 important concepts for interacting effectively with LLM.

Prompt engineering,
prompt chaining,
proof loops

At its core, prompt engineering means going beyond simple Google-style queries.

Instead of typing “top public health topics,” you provide structured instructions: ROLE, GOAL, CONTEXT/CONSTRAINTS, OUTPUT FORMAT

5 parts of a structured prompt:

ROLE - assigning a job title
GOAL - clearly state the purpose of the query
CONTEXT - add additional background information
CONSTRAINT - add additional parameters, guidelines, or guardrails
OUTPUT FORMAT - how you want the results to appear

You can further improve results through prompt chaining, where you refine outputs through follow-up questions either before or after you run the query.

"Before you start, ask me any questions you need so I can give you more context. Be extremely comprehensive."

Once the output is shared and results are received, proof loops are an important step. This is where you validate results by reading the outputs, comparing multiple runs, and/or checking sources. These strategies are especially important in public health, where accuracy and interpretation matter.

Step-by-Step Walkthrough: Finding Trending Public Health Topics

Step 1: Set Up Your Tools

Open a large language model (e.g., ChatGPT; no login required for this activity) and a topic tracking sheet (such as Google Sheets). The sheet will help you log prompts and compare results across attempts.

HANDBOOK Prompts to Copy & Paste: Google Doc https://docs.google.com/document/d/1EJ8_ZERc0SHsU-EM-KTV7Zot8dzGTrqu0Gr_S4wxUMg/edit?usp=sharing
(Optional) Prompt Tracker Google Sheet (run multiple prompts and compare in a pivot table): https://docs.google.com/spreadsheets/d/1Nmo2K_1zS4fzjukwH14fgU-nlkOGUUYWMC1Fc5kZZ2w/edit?usp=drive_link

Step 2: Run a Basic Prompt

Start simple with a query like:

“What are the top trending topics in public health?”

Observe the output. You’ll likely get a broad, inconsistent list. While this is useful as a baseline, it is not very precise.

Step 3: Create an Engineered Prompt

Now refine your input by adding structure:

ROLE: “Act as a public health analyst”
GOAL: “Identify the top 5 trending topics in public health”
CONTEXT: “Focus on topics in the news and public discussion”
CONSTRAINTS: “In the U.S. over the last 30 days”
OUTPUT FORMAT: “Return the results in a table with brief explanation”

Run this prompt and compare it to the basic version. You should see more focused, structured, and relevant results.

Step 4: Test Prompt Chaining

Enhance your prompt by adding the following to the end:

“Before you start, ask me any questions you need, so I can give you more context. Be extremely comprehensive.”

The model may ask about scope (e.g., global vs. local, social media vs. academic sources). Your responses will shape a more tailored output. This step often changes the results significantly.

Step 5: Experiment with Roles and Goals

Now it is time to iterate, test, and refine your prompt. As you edit, be as specific as possible. Engineered prompts can be long.

Be as specific as possible in your prompt.

INCLUDE:
- ROLE: Assign a role to set the perspective
  - Example:
    - “Act as a senior public health consultant…”
    - “You are a policy expert…”
- GOAL: Task and end result. Use action words
  - Example:
    - "Write," "Summarize," "Analyze," "Create.”
- CONTEXT: Provide background information for the goals/
  - Example:
    - "Prioritize traditional news outlets over social media”
- CONSTRAINTS: What are guidelines and guardrails?
  - Example:
    - "List the top 5"
    - "Reference only peer-reviewed articles from PubMed”
- OUTPUT FORMAT Define how you want the result to be formatted
  - Examples:
    - “table with the headers: topic | reference url | hashtags | description,”
    - Format: “bulleted list,” “policy brief,” “presentation script.”
    - Length: “brief,” “300 words,” “30 minutes.”
ADD THIS (OPTIONAL):
- TONE: Match language and emotion.
  - Examples: "Scientific and professional”; “lighthearted and exuberant”; “compassionate and calming”
- EXAMPLES: What are you looking for? Share examples with the LLM of to provide additional context.

Try modifying the ROLE and perspective:

Social media analyst → emphasizes viral, social topics
Policy advisor → focuses on government priorities and writing policy briefs
Researcher → highlights evidence-based trends

Note how each role produces different insights, revealing how framing influences outcomes.

Please refer to THIS HANDBOOK for more guidance and examples of prompts.

Step 6: Create a Proof Loop with Multiple Queries

Because each query is likely to produce slightly different results and there is no clear 1 right answer, we recommend doing multiple queries and compare themes.

Record your final prompt and the five topics it generates in the Prompt Tracker Google Sheet.
Categorize each topic (e.g., infectious disease, mental health, policy).
Repeat the process (or compare your results to the results of others) and look for recurring themes across multiple prompts.
We recommend counting the number of times a theme or topic recurs across multiple queries. For an example on how to track topics, see Prompt Tracker Google Sheet, Tab: ‘SEE PIVOT Table’ tab. This table rolls up the different categories by performing a frequency count and organizes the most consistently mentioned topics (i.e., most reliable “trending” topics) in a bar chart.

Step 7: Reflect and Refine

In the case of trending public health topics, there’s no single “correct” answer. The best results depend on your goals and the goals of your team. Use iteration to repeatedly refine prompts until the output aligns with your values, goals, and needs.

By combining structured prompting, iterative testing, and validation, you can turn AI into a powerful decision-support tool, while still centering your work in critical thinking and public health expertise.

Submit A Screenshot of Your Prompt for a Certificate

Note: We review projects once a month - typically at the end of the month.

Instructor

Tracy Flood MD PhD

President and Co-founder

BroadStreet Institute

Dr. Flood has over a decade experience in foraging for community health data, as well as in data visualization, report writing, mapping, and data design for software. She is passionate about empowering community change, increasing data literacy, and turning data into actions that will have long-term impact. She has worked with over 2,000 interns as President of BroadStreet.

Connect with Tracy on LinkedIn

AI Consultant

Isha Baxi B.Eng, M.Binf