A Framework for Beginners to Write Useful GPT Prompts I Learned from a Prompt Engineering Champion

Chris Lettieri
The Augmented Engineer
3 min readJan 19, 2024

Most beginners are using GPTs all wrong, and it’s time to change that.

AI is not magic, it is a tool we must learn to use correctly to get the results we want.

When I started using ChatGPT last year, I thought I was doing it right, but I wasn’t. Most beginners expect too much without providing enough details in their prompt. In general, you must provide context and be specific.

Here is one simple framework for how to write better prompts so that AI can be useful for you.

Provide Context and Specifics with the CO-STAR Framework

The number one mistake beginners make writing prompts is not providing enough instructions.

The two most important rules here are to provide the context and be specific with your ask. There are many ways to do this. I recently read Sheila Teo’s post on Medium about how she used the CO-STAR framework to win a competition.

Here is her CO-STAR framework with my suggestions of what questions you need to answer.

  • Context: What’s your role? What’s the domain of the world does GPT need to understand to give you a useful answer?
  • Objective: What task do you want the GPT to do?
  • Style: GPT’s generate text, what style do you want the text to be in?
  • Tone: What tone should the GPT take with it’s response?
  • Audience: Who is the message for? What level of expertise do they have?
  • Response: What specific output format do you want the response in?

From my experience, you must at least include the context, a specific task, and an audience to get a useful response.

Example Prompt Using the CO-STAR Framework

# CONTEXT #

I am a software engineer tasked with explaining the benefits and functionalities of our new cloud-based storage solution.

# OBJECTIVE #

Write a brief report that clearly outlines the advantages of adopting our cloud-based storage solution, focusing on aspects like security, scalability, and cost-effectiveness.

# STYLE #

The report should be in a professional, informative style, similar to white papers or industry reports that effectively communicate technical information to a non-technical audience.

# TONE #

Informative and convincing, aiming to build confidence in the technology among stakeholders who may not have a technical background.

# AUDIENCE #

The primary audience for this report is the executive team and department heads who are not familiar with technical jargon. The report should be accessible, avoiding technical language, and focusing on practical benefits.

# RESPONSE #

A well-structured report, with clear headings, concise explanations, and a summary of key benefits.

Writing a good prompt takes more effort on your part. But ChatGPT or any LLM is not magic. You need to think through the problem.

The more work you do up front, the more useful a response you’ll get.

--

--

Chris Lettieri
The Augmented Engineer

Trader turned data engineer, building deep neural networks on time series data. I like second brains and podcasting @ (https://bitsofchris.com).