Recently

See 184 more →

Prompt Engineering

• Updated Jun 16, 20253 min read

Task + Context + Exemplar + Persona + Format + Tone

In an ironic turn of events, I’ve found myself working with OpenAI at a contract capacity. And with any tech company, people are keen on using the product in their process religiously, while I’m kind of left behind still thinking about personalization sans machine learning.

I’m generally not a fan of ChatGPT because of how generic it sounds, but I recognize it can be the most powerful. Accessing Claude on my work computer seems blasphemous, so I’m making an effort to use ChatGPT as part of my process. Often, I find myself using ChatGPT and getting suboptimal results while coworkers are getting answers with more breadth and depth. I realize my prompts aren’t specific enough, and I’m not being specific about being specific. What does that even mean?

In a bathroom break, I searched for ways to prompt ChatGPT better , which led me to Jeff Su’s video. The formula is as follows, in order of importance:

  1. Task
    • Start with action/verb
    • Clearly articulate end goal
    • Could be one task, or multiple tasks
    • Give it a clear objective with needs and goals so you don’t get a generic summary
  2. Context
    • Getting just enough context
    • What is the user’s background?
    • What does success look like?
    • What environment are they in?
    • GPT gives you endless possibilities, so it’s most productive to ask these questions to limit it.
  3. Exemplars (Examples)
    • LLM research suggests that providing examples leads to more accurate results
    • Providing a specific structure or framework
    • Not necessary for every prompt, but it will improve the quality of output
  4. Persona
    • Who do you want AI to be?
    • Think of someone you wish you had instant access to
    • Real people could be used, if they were famous enough
  5. Format
    • How you envision the output
    • Common formats: emails, bullets, code blocks, paragraphs, mark down
  6. Tone
    • Casual, formal, witty, enthusiasm, pessimistic, etc
    • Tell it the feeling it’s going for

General Tips

  • Ask ChatGPT to help you be more specific with a prompt
  • Tell ChatGPT to bold changes to easily skim through what was changed
  • Build a prompt library?!
  • Use answer levelling for more quality ideas — it’s like giving chatGPT a frame of reference where level 1 is basic and level 2-3 (recommended sweet spot) is a more advanced
  • Use chain of thought prompting (breaking the task in manageable steps) to provide more accurate and consistent answers (eg., write me a cover letter, vs write a captivating hook for a first paragraph of a cover letter, now write the second to expand x, now write the closing paragraph to close it off)

Additional resources

title: "Top 5 ChatGPT Use Cases for Professionals"
image: "https://www.jeffsu.org/content/images/2021/03/RANDOM-PICS--2-.png"
description: "Hey friends - If you’re tired of generic ChatGPT tutorials that only scratch the surface, you’re going to love this. 😉 This post dives deep into five practical and actionable use cases, specifically designed for professionals like you. Learn how to leverage ChatGPT to enhance your productivity, streamline your workflows, and"
url: "https://www.jeffsu.org/top-5-chatgpt-use-cases-for-professionals/"

Graph View