An Ivan Allen College communicator went to prompt engineering school with Assistant Professor Yeqing Kong. In this article, she shares what she learned.
By Stephanie N. Kadel
If you’ve spent even an hour or two on ChatGPT or another generative AI model, you know that getting it to generate the content you want can be challenging and even downright frustrating.
Prompt engineering is the process of crafting and refining a specific, detailed prompt — one that will get you the response you need from a generative AI model. This kind of “coding in English” is a complex and tricky process. Fortunately, our faculty at the Ivan Allen College of Liberal Arts at Georgia Tech are engaged in teaching and research in this exciting emerging field.
I met with Assistant Professor Yeqing Kong in the School of Literature, Media, and Communication to talk about prompt engineering. She shared three approaches to crafting a prompt that she has collected from leading experts. Kong will be teaching these methods, and much more, in her new Fall 2024 course, “Responsible AI for Communication.”
“This new course aims to equip students with the skills to integrate AI tools into the content creation process in ways that are both effective and ethical,” says Kong.
After we spoke, Kong invited me to try crafting a prompt using the rhetorical approach, the C.R.E.A.T.E. framework, and the structured approach to prompt engineering. These methods, as well as my results, are described below. I hope you’ll try these approaches on a prompt of your own!
Where I Started Out
For the purposes of this article, I decided to see if I could get ChatGPT to write a short social media post for this article. Here is my first attempt, without any of Kong’s methods applied:
Now let’s apply some of the prompt engineering methods Kong shared with me.
Method No. 1: The Rhetorical Approach
The rhetorical approach to prompt engineering was developed by Sébastien Bauer, an academic at the Universitat Autònoma de Barcelona. This method involves describing your main claim — “THE most crucial point,” as he puts it — and then your rhetorical situation. As part of this approach, the prompt may include descriptions of:
- The audience.
- The context. Where will the audience read this?
- The author and ethos. This might include your role or credentials.
- Pathos. What should your audience feel or believe?
- Logos. What logical points do you want to emphasize?
- Arrangement. For example, chronological or alphabetical.
- Style and delivery. For example, “Limit your response to 100 words and avoid using the second person.”
Since I didn’t really have an explicit or challenging point to support in my social media post, I described the rhetorical situation in my prompt.
If you get a response that’s off the mark, don’t give up! Revise and resubmit. In all, I went through about eight iterations before I got to the final version of this prompt. (For those interested, details are in the sidebar.)
Refining a Prompt
Prompt engineering is definitely a refinement process, and I got better at it as I went along. Identifying and addressing information that the AI finds confusing or extraneous is key.
For example, I told ChatGPT that I’m Gen X in one version, in hopes that it would infer something about my social media style. Instead, it included this odd sentence about being “a proud Gen X member” and made the post sound like a sales pitch. (“Don’t miss out!”)
I took out the Gen X bit and told it the post is for LinkedIn, which solved these issues. But it still wasn’t quite what I wanted, so I kept going. The final result (pictured in the article) was much closer to what I wanted.
Method No. 2: The C.R.E.A.T.E. Framework
In the C.R.E.A.T.E. approach developed by AI consultant and author Dave Birss, prompts are framed by addressing the AI as “you.” The acronym stands for:
- Character – Describe the role the AI is to assume. Kong says you can add in aspirational elements like “You are an experienced writer who crafts concise text without filler words or jargon”
- Request – Clearly and specifically define the request. “I want you to…”
- Examples – Give examples if you have them
- Additions – Refine the task. Describe a point of view to consider or a style to use
- Type of Output – For example, a 100-word summary or a chronologically organized bio
- Extras – Any further information you wish to provide, including reference text
The resulting prompt is a little like an instructor’s assignment, as you can see:
More information on the C.R.E.A.T.E. approach is available in Birss’ LinkedIn Learning course.
Method No. 3: The Structured Approach
The structured approach was developed by Lance Cummings, an AI content specialist and an associate professor at the University of North Carolina Wilmington.
The formula for this approach, as Cummings describes it in “The Anatomy of a Prompt,” is as follows:
- Start with a role and a goal. Broadly describe the overarching aim and purpose
- Provide context and background
- Clearly define the task. Make your expectations explicit and detailed
- Provide any reference content the AI will need to complete the task
To try out Cummings’ approach, I pasted the draft content of this article (up to this point), following this text:
ChatGPT responded:
I plan to post the response I got using Method #1 (the rhetorical approach), but I'm going to add in a couple of the emojis I got with the response above.
“You don’t need to stick to just one of these methods,” Kong adds. “You can draw elements from all three. Crafting a prompt is about finding the most effective and clear way to present the language model with its task, while providing context, details, and reference material as needed.”
Generative AI Models and the Liberal Arts
In early 2023, it seemed half the world was enthralled — and the other half appalled — by ChatGPT. Some thought generative AI, which creates original content in response to a prompt, would be especially bad news for the liberal arts. A year or so later, Kong disagrees.
“Generative AI has increased the demand for the very skills we worried it would replace, such as writing, editing, and critical thinking,” says Kong. “These liberal arts skills are also the key to prompt engineering.”