CLEAR Framework for Prompt Engineering, designed to optimize interactions with AI language models like ChatGPT. The framework encompasses five core principles—Concise, Logical, Explicit, Adaptive, and Reflective (Lo, 2023).
Keep prompts short and to the point to avoid confusion and ensure the AI understands the request clearly. In short, have brevity and clarity in prompts.
Example:
Instead of requesting, “Please provide a comprehensive analysis of factors impacting the future of electronic vehicle industry in Canada”, use a concise prompt like, "Analyze the factors shaping the future of electric vehicles in Canada."
Structure prompts in a logical sequence to guide the AI through the desired process or thought pattern.
Example:
“List the steps to write a research paper, beginning with selecting a topic and ending with proofreading the final draft” is a logically structured question.
Be specific and detailed in the prompts to minimize ambiguity and improve the accuracy of the AI’s responses.
Example:
Instead of, “Tell me about elderly care,” an explicit prompt would be, “Provide a concise overview of different types of elderly care, along with their costs and benefits.”
Adjust prompts based on the context and the AI’s previous responses to refine and improve the interaction.
Example:
If asking, “What are some ways to conserve water?” leads to generic responses, try a more targeted and adaptive prompt like, “List household practices for conserving water and their potential impact.”
Review and reflect on the AI’s outputs to learn from the interactions and continuously improve prompt quality
Example:
After receiving AI-generated content on the eco-system of Humber River watershed in Toronto, evaluate the response's accuracy, relevance, and completeness. Use insights from the evaluation to refine future prompts, such as asking for more specific aspects such as biodiversity or conservation efforts.
Lo, L. S. (2023). The CLEAR path: A framework for enhancing information literacy through prompt engineering. The Journal of Academic Librarianship, 49(4). https://doi.org/10.1016/j.acalib.2023.102720
RISEN (Balmer, 2024) is another prompting mnemonic that has found some popularity in the domain of digital entrepreneurship and may offer some useful strategies to structure your prompts.
Ask the GenAI to take the role of a subject matter expert in the topic you are interested in.
Example:
"Take the role of a housing policy researcher who has spent several years investigating the housing crisis in Ontario."
Provide the GenAI with explicit instructions for the behaviour or task you want it to complete.
Example:
"Provide a comprehensive list of everything I should consider in an undergraduate study on the housing crisis in Ontario."
Provide a guiding structure or steps for the GenAI to generate its response.
Example:
"1. Start by listing all the possible causes of the housing crisis.
2. Then describe the impacts on the social determinants of health.
3. For each item, offer a leading solution from a policy perspective."
Provide an overall end goal to guide the GenAI’s focus.
Example:
"The solutions offered in the list will be used to identify gaps and opportunities for research on their effectiveness."
Establish constraints or key requirements for style, tone, or format for the GenAI’s output.
Example:
"Ensure the list does not exceed 500 words and utilizes an informed and professional tone. Ensure that each cause, impact, and solution are grouped together under the same numbered heading."
Balmer, K. (2024, March 5). You’re using ChatGPT wrong - upgrade prompts with this framework [Video]. YouTube. https://youtu.be/RlmSQiMW3nk
RISEN (Balmer, 2024) is another prompting mnemonic that has found some popularity in the domain of digital entrepreneurship and may offer some useful strategies to structure your prompts.
Specificity is key in any research endeavour. The same is true when engaging with genAI. When you're developing a prompt, try to be as specific as possible with your ask. Not only will that likely lead to more meaningful, relevant outputs, it may also limit the chances of inaccuracies in outputs. Another added bonus is that you have an opportunity to reflect on your topic or question at the start of the journey.
Example
If you’re looking for a good meal in Etobicoke, asking “What is the best restaurant in Toronto?” could return results from the entire city. Asking “What is the best restaurant in Etobicoke [or Rexdale]” gives you a better chance of finding what you’re looking for.
When prompting genAI, consider if asking the AI to take on a specific role might help focus the output you're seeking. The AI will try to take on that role as it develops the output.
Example
If you're need to learn about the origins of the universe, you could ask the genAI to explain that phenomena "acting as if it were a cosmologist."
GenAI can produce outputs in many formats (essays, reports, summaries, etc), and so specifying how you'd like the output presented may be helpful.
Example
If you'd like a to learn about the plot of a film to see if it's something you'd like to watch, you could ask the genAI to "provide a plot summary of the film Seven Samurai, without spoilers."
Try to be clear with genAIs about what you do or don't want in the output you receive. If you have specific dietary restrictions when asking for a particular recipe, include those.
Example
If you'd like a recipe for multigrain bread, but are allergic to buckwheat, you could ask to "provide a multigrain bread recipe without buckwheat as an ingredient, that also does not require a bread machine."
If you'd like the output you receive from the genAI platform to be similar to a specific style , you can share a sentence or paragraph as an example with the genAI platform.
Crucially though, do not infringe on copyright:
You can provide notes on specific tones and audience for the output you'd like to receive back.
Example:
"Could you explain the expansion of the universe in a comedic manner for an audience of first year undergraduate students?"
Prompt engineering, as with starting on research on a specific topic, is a learning process. You get better at it over time as you adapt, reflect, and continue to learn. In this case, working from an initial prompt you have thought of is a good start. Building on it with some of the strategies discussed above will help you augment and refine the prompt (and so the output) as well as your own skills.
Many genAI platforms ask for feedback already in different ways - whether that is typed response from you, a thumbs up/thumbs down, or more. You can provide feedback to help the platform learn. Consider engaging with the genAI platform as though it were a colleague or teammate that you're working together with on a project/topic.
As a starting point, you could even ask the genAI platform to create a prompt for you.
Example
If you have a basic idea of the topic you'd like to explore, you could ask the platform "What should I ask you to help me understand the significance of Joni's Mitchell's music?" Additionally, you could add something like "Tell me what else you need to do this" to help fill in gaps left by the AI.
Adapted from Harvard University's "Getting started with prompts for text-based Generative AI tools"