This post is part of our series “An Educator’s Guide to AI,” written by FACTS team member and educator Kevin Donohue.
When I was a beginning teacher, I found Harry and Rosemary Wong’s The First Days of School to be extremely helpful. Stepping into a classroom of 34 fifth graders only half my age, I was unprepared until the Wongs’ focus on professionalism, preparation, procedures, and communication helped guide me to establish clear expectations.
This guided my career as I moved into middle school and then administration. A clear set of goals, with support to reach them, benefits students, teachers, parents, and staff. The more clearly we delineate what we are looking for and provide models or examples, the better results we get. This same approach is immensely helpful for using artificial intelligence well.
When we first interact with a large language model (LLM), such as ChatGPT, Gemini, or Claude, we see a standard input textbox. We might resort to our usual search queries, as if we were using Google or another search engine (shout-out to Ask Jeeves!). However, due to the construction of their model, the accumulation of their training data, and the tool’s neural networks, the skill of prompting is the key to unlocking better results.
Fortunately, following the advice of the Wongs and other strong pedagogical approaches prepares educators to prompt effectively. The more detailed instructions we provide to the LLM, the clearer the results we receive. Just as when we give a writing assignment to students, we want to ensure we detail the length, topic, basic construction, steps along the way, rubric for assessment, and even provide examples. Doing these for our AI tools will get our results closer to an A.
An AI Prompting Recipe
There are many different guides to prompting. In the education realm, I recommend several that I will link to at the bottom of this post. Most guides center on five basic concepts: context, task, instructions, parameters, and input. Let’s walk through each of these as we seek to cook up a good prompt.
-
Context is the background information for the generative LLM. Think of a student who walks into the classroom tardy. We quickly provide resources and information so that they are not lost and get caught up. For AI, we do something similar by providing a role or job for the AI and the essential information relevant to the task. For example, we might ask the AI to be a pedagogical coach, speech language therapist, or a social media influencer. We also want to provide the LLM with information about our audience.
-
Task is the heart of our prompt recipe. This is the specific job or duty we want the AI to do. Do we need help with drafting an email, planning a meeting, summarizing research, or analyzing demographic data? This is the one sentence that best directs the AI to do what we want. This is what we approach the blank interface box with first when writing our prompt.
-
Instructions add the flavor to the task, enhancing a simple request into one with greater depth and clarity. Rather than just asking a student to write a paper, we ask for a five-paragraph essay. Similarly, rather than asking AI for an email, we ask for a professional three- to five-sentence response that links to the relevant handbook sections. Remember, just like with students, more is better, and unlike with students, the AI usually reads all the directions.
-
Parameters are the constraints on the task’s result. Like students who turn in pages of text when asked for a simple answer, AI tends to expand when it isn’t necessary. Parameters help limit the results. It’s also helpful to include what you don’t want (or negative prompting). Because AI can generate new responses based on previous ones, when we add additional instructions and parameters, AI can generate a response closer to what we desire. This is a fantastic way to refine AI’s output.
-
Input is the last key aspect of a good prompt. Input can be additional data, a sample, or needed information to complete the task. For students, this is where we point them to specific sources or sections of their textbook. For AI, this might include previous email communications, the handbook itself as a reference, and de-personalized communications. Remember, the data you share with an AI is no longer yours. Be smart and respectful about what you type and upload.
Putting these ingredients together can make our prompting stronger. For example, here is a sample prompt drawn from AIforEducation:
You are an expert speech-language pathologist, highly skilled in supporting and enriching students’ speech and language growth and development. Your task is to create a list of synonyms for a third-grade student. The list should include at least 15 words, with five synonyms for each. The words should be interesting and challenging, yet appropriate for third-graders.
Breaking the prompt down:
-
Context: You are an expert speech-language pathologist, highly skilled in supporting and enriching students’ speech and language growth and development.
-
Task: Your task is to create a list of synonyms
-
Instructions: The words should be interesting and challenging yet appropriate for third-graders.
-
Parameters: The list should include at least 15 words, with five synonyms for each.
-
Input: for a third-grade student
You can already see additional areas where more context, instructions, parameters, or input could create a better list of synonyms. You could strengthen this prompt further by listing gender ratios, student interests or challenges, recent books or articles read, etc. By following these guidelines, you’ll be cooking with AI in no time!
In addition to these five fundamental aspects, I like to append two more parts to a prompt.
-
An appeal: encouraging deep thinking and stressing the importance of the result tends to lead to stronger results.
-
A request for questions: by adding a phrase like, “Prior to generating a response, ask any clarification questions that will provide a richer response.” This can help catch details missing from the context and input you might have missed.
Additional Resources for Better AI Prompting
If you aren’t sure where to start, consulting a prompt library like the one on AIforEducation or an embedded AI tool like MagicSchool, SchoolAI, Brisk, or Google Gemini is a great way to get started with training wheels for prompting. Explore even more tools in our blog post about accessible AI for educators.
Dan Fitzpatrick offers a different framework with his PREPARE model, highlighting more specifics for each prompt aspect. Caitlin Tucker likewise offers a simple model with REFINE. Finally, Leon Furze digs into some of the more advanced features of prompting in this blog post, Process > Prompts.
Like mastering teaching, learning to write an effective AI prompt takes time. I am a much better teacher today, not only because I followed the advice of mentors like the Wongs, but because I used that advice day in and day out for years with different students, subjects, and places. Building prompts over time can help strengthen the skill and enable you to use AI efficiently and effectively.
About the Author
Kevin Donohue is the Leadership Coaching Manager for FACTS, where he also delivers customizable professional development and coaching in artificial intelligence for every group of stakeholders at a school. A Tampa native, Kevin joined FACTS after thirteen years teaching and leading in Catholic schools in Los Angeles, San Diego, and Boston. He lives in Arkansas with his wife, a professor of philosophy, and four children. This piece was written by him but checked by ChatGPT and the amazing humans at FACTS.
Kevin Donohue
Recent Blogs
- Testing the Waters: How to Try AI Without Sinking
- Campus and School Management Systems: What They Are, Why They Matter, and How Schools Should Evaluate Them
- What Makes Daily Activity Reports a Game-Changer for Parent Engagement?
- What’s New in Scheduling: A Closer Look at Phase 2 Enhancements
- What’s New and Next at FACTS: Q3/Q4 Product Roadmap Update
