AI Prompt Engineering Tips for Teachers

To build on the second grade science teacher example, you might type this next:

“Design a 60-minute science lesson about photosynthesis. Your objective should align with Next Generation Science Standard 2-LS2-1. The concept is new to students, although they have each planted a flower from a seed 2 weeks ago. The plants are just emerging from the soil. In the lesson, include an opening to capture student attention, identification of the outcome, and then an outline of the lesson. Make sure that the lesson is engaging and provides a hands-on learning experience for students.”

If you break down this example, you can see that several important details were added. You set a specific time limit of 60 minutes, you identified the task of designing a lesson, and you provided the objective for the lesson, including both the general topic of photosynthesis and the specific standard with which the lesson should be aligned. With your new details, you even explained the output format that you wanted from the AI. You said that there should be an opening attention-getter, an outcome, and an outline of the lesson. You also provided a little more background and context to help the AI better align to your specific needs.

The key is to remember that the more detailed you can be with your prompt, the better the AI can serve as an understanding and helpful planning partner.

This is a really important aspect of interacting with a generative AI chatbot, as this is where you can turn a below average response into an excellent one. Since AI is predicting content based on your wording and detail, it’s possible that the prompt language you chose doesn’t result in a perfect outcome. That’s to be expected. Much of prompt engineering is “thought engineering” and follow-up questioning.

Again, imagine that you are talking to another human colleague who might not have understood what you meant the first time. What would you do? You’d probably provide additional context and detail before asking a follow-up question. You should do the same thing with the chatbot. Follow up until you have what you need.

Remember, you can also extend the task if you think of something else you’d like generated. For instance, once you have a lesson about photosynthesis, you might follow up with something like this:

“Now, create a practice activity for this lesson with a formative assessment. Be sure to provide the answer key.”

If you’re just getting started and want a little more structure and guidance than what was outlined in the first three tips, you might want to start with a sample prompt framework. These are sample prompts with somewhat generic wording and then space holders for the specific details you want to add. Think of these like “fill in the blank” exercises.

Here’s an example from AI for Education for writing a strengths-based lesson plan prompt. The capitalized, bracketed content indicates fields to be completed by the user.

“Act as an experienced [GRADE LEVEL / CONTENT AREA] teacher, skilled at designing engaging learning activities for students of all ability levels. Develop a [CONTENT AREA] activity for [GRADE LEVEL] students who are developing an understanding of [CONCEPT]. They have a good understanding of [LIST PREVIOUSLY MASTERED CONTENT]. Their other skills include [LIST STUDENT STRENGTHS]. Suggest three to five activities that build on these strengths to help them develop their skills in [CONCEPT].”

There are a growing number of prompt libraries available online. Here are a few to get you started:

• GenAI Chatbot Prompt Library for Educators (AI for Education)
• AI Prompt Libraries for Educators (Eric Curts via Control Alt Achieve)
• Prompt Library (Dr. Ethan Mollick via More Useful Things: AI Resources)

In addition to prompt libraries, there are several AI tools that are specifically built for education and that have become very popular with teachers. These sites offer libraries of common teaching tasks that you can select. Tasks range from developing assessments to lesson design to communication.

By selecting a teaching task, the site asks you some questions about your needs and then does much of the prompt engineering in the background for you. The prompt it uses is shaped by both a preset structure as well as the information you have provided. These sites can offer guidance and help reduce the potentially overwhelming experience of staring at a blank prompt box in one of the major generative AI tools.

Three tools in particular have become popular in education circles:

• MagicSchool
• SchoolAI
• Brisk Teaching

There are several definitions for reverse prompt engineering. In its simplest form, you can think of reverse prompt engineering as asking the AI to help you generate your prompt.

For example, you might type something like this:

“What do you need to know to help me generate an excellent science lesson for a class of second graders?”

The AI will then provide you with a list of questions that you would answer in a follow-up message. This targeted information can help the AI generate your desired output, which is a lesson.

Another version of reverse prompt engineering would be to follow up your conversation string with the AI by asking:

“What else do you need to know?”

Just as before, you would then answer the AI’s questions or provide the requested details.

With this technique, you provide the AI an example of the type of response you’re seeking. That example is the one-shot prompt. For the science lesson example, you might paste in a copy of a successful lesson that you have developed and then tell the AI to use that lesson as a model.

Essentially, you are teaching the AI what you want for an output. It’s similar to showing a human collaborator an example of what you’re trying to produce. Having a strong model from which to work can greatly improve the outcome. It can be helpful to model AI interactions after human-to-human ones. Visualize the AI chatbot as your thought partner and have a professional conversation.

In this model, you not only give the AI a task but also instruct it to think through the steps needed to reach a solution, often providing examples of the reasoning process at the same time. Chain-of-thought prompts can take a bit longer to write, but you’ll be rewarded with a more detailed and accurate response.

For example, if you were trying to get the AI to explain how to solve a math problem, you might type something like this:

“Here is how to solve a math problem step-by-step:

1. “To solve 2x + 3 = 11, first subtract 3 from both sides to get 2x = 8.
2. Then, divide both sides by 2 to find that x = 4.”

Once you’ve added the model of how to work through this type of math problem, you then give the AI a new problem and ask it to solve that problem sharing the steps it used based on the model you provided.

“Now, solve this problem step-by-step: 3x – 5 = 10.”

You are essentially teaching the AI about what you want and how it should reason through the work. Again, this is a similar approach to one you might take with a colleague. It’s good teaching, but instead of teaching a person, you’re teaching the AI about what you are seeking.