Skip to Main Content

AI and Language Models (LLM)

CRISPE is an acronym for one of the methods that can be used to make instructions clearer and more effective for AI models. Skilled researchers have developed the CRISPE Method, a formulated plan to construct prompts that lead Generative AI whether it be ChatGPT or CoPilot to produce the desired content. Here’s a brief explanation of CRISPE.

Understanding the nuances of prompt creation and practicing good prompt design techniques are important components of AI literacy.

Objective

Needs Improvement

Revised Prompt

C-Context

Set the stage for what you need. Provide all relevant background information that frames the task.. Also, tell it who the audience is.

“Explain the French Revolution.”

(No detail — could result in a massive, unfocused history dump.)

 “You are helping a high school student prepare for an exam. Explain the French Revolution with a focus on causes, not outcomes.”

R-Role

Define the persona the AL should take. This helps keep the language and complexity of the material appropriate. Are you an undergraduate or an instructor? Is this for an English class or biology? Are you preparing this for a class, for a journal, or for a job?

“Teach me about descriptive statistics.”
(No role specified — could result in overly complex or overly simple answers.)

“I’m a high school senior who needs to understand mean, mode and median so that I can build scatterplots for my statistics class.”

I-Instruction

be clear about what you want AI to do. Break it down what you’re doing, how you’re doing it and what you hope to achieve.

 “Explain mitosis”

(Unclear what to explain, to what depth, or for what purpose.)

 “Help me understand what mitosis is and each of its stages as it relates to the development of cancer for my college biology class.”

S-Specification

how do you want the output delivered? Bullet points, paragraphs, number of words, tone of voice. Is it casual or technical?

“Describe World War II.”
(Far too broad — could take pages.)

 “Provide dates, events and their description to build a timeline of WW2 focusing only on the Pacific theater.”

P-Performance

Set expectations for how the response should be delivered (tone, format, depth) What, if any, quality standards should be met?

“What are the stages of project management?”
(No performance guidance — might come back as a plain list or dense paragraph.)

“Write the stages of project management in a numbered list, with a one-sentence practical example for each stage.”

E – Examples → Show the AI what kind of output you want.

Provide any complete examples that you have. You might reference websites or articles.

 

Created with assistance from ChatGPT.

The CRISPE framework is just one model for designing effective prompts for generative AI tools. Other models include the PROMPT Design Framework (from Sarah Hartman-Caverly at Penn State University Libraries) and the CREATE Framework (from Dave Birss, a Generative AI Educator for LinkedIn Learning).

More tips for creating effective prompts for generative AI tools:

  • Try—and try again! Like doing research, prompting AI is an iterative process. Even well designed prompts may need refinement as you learn the limitations of each AI tool. 
  • Assign a role to the AI. Giving an AI tool a role to play provides it with context, sets expectations, and guides its behavior toward more relevant and tailored responses. For example, if you're looking for help brainstorming keywords for your research topic, you might prompt the AI tool to "act as an academic librarian." Or if you want to practice for a job interview, you may prompt it to "act as a hiring manager."
  • Give examplesShare example responses or templates with an AI tool to guide how it generates content.
  • Specify what not to include. In addition to prompting an AI tool for what you want it to include in its response, you can also prompt it to avoid specific content, language, or approaches.
  • Adjust the AI's temperature. In the context of AI, temperature is a parameter that controls the randomness of an AI model's output. For many free generative AI tools, the temperature is preset by developers to ensure a balance between creativity and predictability in responses. However, some AI tools allow you to change their temperature. Look for options that allow you to generate "more precise," "more creative," or "more balanced" responses.
  • Ask the AI to explain its processes. Not sure how an AI tool arrived at a response? Ask it for more information about how it got there, including its step-by-step reasoning, the factors it considered, and the sources it consulted.