Skip to Main Content

Generative AI

Effective Prompts

A prompt is the input (usually text) that provides the instructions to a GenAI tool (like ChatGPT) to generate a response.

The following are some general tips for what is now becoming known as 'prompt engineering' (UNESCO, 2023):

  • Keep it simple. Clear, straight forward language in a logical order. Avoid complex or ambiguous wording.
  • Include examples to illustrate the desired output or format for generate content.
  • Include context. This helps to generate relevant and meaningful responses.
  • Refine as needed. You may need to experiment with variations of your prompt to get the desired output.
  • Your word choice matters. How a prompt is written can generate biased or harmful responses. Review your word choice carefully to avoid inappropriate responses. Avoid using leading or loaded questions that may produce biased responses.

Review the limitations of GenAI so that you clearly understand what it can and cannot do. And always evaluate the output.

Typical categories of text prompts used in GenAI tools:

  • Can you write / explain / summarize / provide / analyze / compose / re-write...
  • Give me a table summarizing some of the key differences between...and...
  • Rephrase this sentence / Expand the idea based on the folllowing descriptors...
  • What are the leading causes / factors / impacts / definitions / benefits / challenges / issues of...
  • What would be a good image to use in my presentation on ....
  • How do I search / cite / code / convert .....
  • Summarize the plot of..inone short paragraph.
  • Create a list of fiveChatGPT features that can be used to promote...
  • I have a debate about... Can you help me find arguments in favor of this, as well as some counter arguments?

Image prompts tend to include more information including the objective, media, action, subject(s) and context for the desired output.

  • Objective examples: an image that showcases... / captures the essence of.../ resembles a scene like...
  • Media examples: painting, photograph, sketch, 3D rendering, illustration...
  • Action examples: standing, walking, sitting, reading a book...
  • Subject examples: dog, students, house, ...
  • Context examples: in a field, on a chair, in a classroom...

For more sophisticated output, you may choose to (a) provide more details about the subject or context of the image, (b) request a specific artistic style, or (c) further define the technical requirements.

  • (a) A photograph of a man wearing a navy shirt.
  • (b) A photograph of a man wearing a navy shirt in the style of cubism.
  • (c) A photograph of a man wearing a navy shirt in the style of cubism with a resolution of 150 DPI.

Image prompting can be a bit harder - test your hand at it with this Say What you See activity from Google Arts and Culture Experiments.

Adapted from "AI-Based Literature Review Tools" by Texas A&M University Libraries.

Prompt engineering is a new and developing skill set. If you are interested in learning more, here are some mini-courses and sites that might interest you:

chat loading...