Introduction
Many of us would agree that a teacher’s most precious, limited resource is time. Using it thoughtfully and efficiently in the classroom is crucial for student learning, and maximizing planning time to prepare lessons is just as crucial and a great challenge.
Colleague AI, a specialized AI assistant, can be a powerful tool for planning lessons faster with greater depth, a rare win-win, assuming we have the skill to effectively prompt it. In this blog post we will explore three effective prompting strategies and also develop strategies for responding when AI does not generate accurate, relevant, and useful outputs.
Effective Strategy #1: Zero-Shot Prompting
Zero-shot prompting is the most direct strategy. You provide the AI with a role, offer instructions and context, and specify outputs. A well-crafted zero-shot prompt will result in an AI-generated response that meets your needs with no further inputs. The challenge with zero-shot prompts is being specific and comprehensive so that the AI has the necessary details to produce the response you desire. It is helpful to think of zero-shot prompts in terms of three components: the role, the tasks, and the output. Example zero-shot prompts are broken down into these components below:
Zero-Shot Use Case 1: Creating a Rubric
| Role | You are an expert teacher and instructional designer, skilled in creating engaging educational content that effectively achieves learning objectives. |
| Tasks | Your task is to design a grading rubric for a 6-grade science assessment to gauge students’ knowledge about light, one- direction mirror, and reflection, on a 3-point scale, using the “what, how, why” explanation framework, and embedding the UDL principles in the grading rubric. Universal Design for Learning (UDL) includes three key principles: Engagement (The”Why”of Learning): motivating learners and sustaining their interest. Representation (The “What” of Learning): offering information in more than one format, such as text, audio, video, and hands-on learning, to cater to different learning styles and preferences. Action and Expression (The “How” of Learning): giving learners various ways to express what they know, such as written assignments, oral presentations, multimedia projects, or other creative outputs. |
| Output | Output your grading rubric in a table. |
Zero-Shot Use Case 2: Building a Welcoming Classroom Culture, designed by Hayley Spohn
| Role | You are an experienced teacher and classroom management expert, adept at fostering a welcoming and inclusive classroom culture, even with students who may be reluctant to participate or show vulnerability. |
| Tasks | Your task is to design a set of beginning of the year activities that will help build a welcoming classroom culture for an 8th-grade class where students often present themselves as “too cool” to be vulnerable or actively participate. The strategies should set the foundation for a classroom culture where every student feels valued, connected, and an integral part of the community. |
| Output | List 10 activities that can be implemented in the classroom at the beginning of the year with a group of 8th graders “too cool” to be vulnerable or actively participate. Avoid any activities that have been over done or will be seen as “cringe” by the students. Include brief descriptions of how each strategy will help every student feels valued, connected, and an integral part of the community. |
Zero-Shot Use Case 3: Writing a Letter
| Role | You are an experienced school leader and expert communicator, proficient in crafting compelling messages that effectively convey information to school stakeholders. Your task is to write a school update letter for Washington State’s AI policy: https://ospi.k12.wa.us/student-success/resources-subject- area/human-centered-artificial-intelligence-schools. |
| Tasks | Include a brief introduction, a section on the Policy, a section for Important Dates, and a final paragraph/summary. Include the following information in your letter: Introduction: welcome educators back to school, remark on their summer experiences, express excitement for the upcoming school year Policy. Summary: explain the importance of AI policy, include the OSPI link, and summarize its key points in five sentences. Actions: school principals organize trainings to prepare educators to implement this policy in classroom instruction. Important Dates: list upcoming district professional development dates. |
| Output | Write the letter in an informative and friendly style. |
In general, the more clarity you have in your mind around the role, the tasks, and the output, the better the AI-generated response to a zero-shot prompt will be. Zero-shot prompting is a particularly good strategy for completing a well-defined task where you may already have a fair amount of expertise, but typing the prompt is substantially quicker than attending to all the details and writing the task on your own. The end result is a product of similar quality to what you could generate without AI, but created in a fraction of the time.
Effective Strategy #2: One or Few-Shot Learning
One- or few-shot learning involves providing the AI with a context and one or more examples of the desired output. AI determines how to craft its response through the examples you provide. This technique might be particularly helpful when you have a clear idea of how you want an output to be formatted but you are not sure how to describe it in a few sentences or a paragraph. Some examples of one-shot prompts are shown below.
One-Shot Use Case 1: Do Now Questions, designed by Hayley Spohn
| Context | My ninth-grade students often struggle with understanding the purpose and effect of figurative language, such as similes and metaphors, each year due to a foundational knowledge gap around these concepts. I want to create a set of ‘Do Now’ questions as a way to embed spiral review around the topic, using examples that are fun, relatable, and accessible for multilingual learners. |
| Example | For example, after presenting the simile ‘The sun is like a big yellow ball,’ ask, ‘Explain how this simile helps us visualize the sun and enhances our understanding of the scene.’ |
| Output | Generate 10 similar questions that prompt ninth graders to explore how figurative language like similes, metaphors, and personification contribute to imagery, tone, and meaning in the text. |
One-Shot Use Case 2: Standards-based Rubrics Generation
| Context | Write a standards-based rubric for 3.OA.B.5 and the assessment item “what unknown number makes the equation true: 8×6=6x()” by following the example in the attached image. |
| Example |  |
| Output | Output in a table. |
The difference between one-shot prompts and few-shot prompts are the number of examples given; more examples means it is now a few-shot prompt. Also, the order of context-examples-output is not critical for AI to comprehend the prompt. If images are included as examples in the prompt it is likely much easier to write the context and output and then attach the images at the end of the prompt.
Effective Strategy #3: Chain of Thought
The Chain of Thought approach can be helpful when you want AI to break down a complex task into smaller, more manageable steps or complete a complex task with multiple steps or inputs. This prompting strategy typically combines components of zero-shot and one/few-shot prompting but can feel a bit more conversational in nature.
Below are a couple examples of Chain of Thought prompts. Full text of AI responses are omitted because of their length and detail.
Chain of Thought Use Case 1: Multi-steps word problem solving
| Human Prompt | AI Response | |
| Round 1 | Solve this word problem step by step: John has 5 apples. He gives 2 to his friend and buys 3 more from the store. How many apples does John have now? | AI responds with an accurate answer but steps are not clear enough. |
| Round 2 | Organize your work into the following steps:Step 1: Understand the initial number of apples John has. Step 2: Calculate how many apples John has after giving some away. Step 3: Add the number of apples John bought. Step 4: Provide the final answer. | AI reorganizes work into the steps given. |
Notice that the prompt in round 1 provided the task (the word problem) and an output (step-by-step solution,) but it provides no context. The AI has limited guidance and generates an answer that does not quite hit the mark. The prompt in round 2 can be thought of as either a more detailed output or providing context; either way, the additional details give the AI what it needed to provide the desired response.
Chain of Thought Use Case 2: Write an Individualized Education Plan (IEP)
| Human Prompt | AI Response | |
| Round 1 | How would you write a useful and informative Individualized Education Plan (IEP)? | AI responds with relevant knowledge for IEPs. |
| Round 2 | Charlie, a third grader, has an emerging concept of multiplication. She can find answers to multiplication problems using a multiplication chart and using blocks but does not fully understand why and how to multiply. She typically can concentrate only 10 minutes at a time. As the special education teacher, you have the option to add modifiers to the goal to make it more achievable or more meaningful for the student. Please create an IEP for her. | AI generates an IEP plan around fluently multiplying, moving from concrete (blocks) to representations (diagrams) to abstract (using algorithms and memorized facts.) |
| Round 3 | This is a good start. Could you create a year-long IEP plan for Charlie, including essential components that my district asks for: name of the student, grade level, subject area, baseline assessment, modified learning goals, specific support strategies, and IEP team. | AI takes its plan from Round 2 and reformats it with the added requirements. |
| Round 4 | Today, I am teaching this CCSSM standard: Fluently multiply and divide within 100, using strategies such as the relationship between multiplication and division CCSS.Math.Content.3.OA.C.7. Please propose several specific ways to modify this CCSS into IEP goals for this student Charlie. | AI modifies its response in Round 3 with a focus on the given standard. |
| Round 5 | Great. Based on your modified goals, please offer me specific teaching strategies that I can support Charlie today in this given lesson. | AI lists strategies and explains how they connect to the modifications it provided in its response in Round 4. |
Notice how this is a rather sophisticated and nuanced use of generative AI that flows more like a conversation. In Round 1, there is no attempt at producing an output that will be used in the classroom; instead the prompt is designed to find out what the AI already knows about IEPs to determine how much knowledge we need to give it, or even if it is up to the task. In this case, the teacher determines the AI has enough knowledge, so the prompt in round 2 provides student information. One very important note here: always avoid providing personally identifiable student information, such as a full name, date of birth, and student ID number, to any generative AI! There is no need for AI to have this information to provide a useful response and it is important to protect your student’s identity.
Also notice how in later rounds, starting with round 3, the teacher tells the AI if the previous response was good or not. Affirmations help AI to follow the same line of thinking. That is particularly helpful in the remaining prompts in this example as the teacher guides the AI to generalize what it has already created to slightly different contexts. Round 3 asks the AI to develop a longer-term plan based on what it has already made, and rounds 4 and 5 ask the AI to zoom in on a particular lesson. These are not attempts to have the AI revise or clarify its work, but rather asks for the AI to build on its work and expand it into other contexts relevant to this teacher.
These use cases look great, but when I prompt AI the responses I get aren’t good. Any tips?
AI, like any tool, takes some time and practice to learn how to work with well. It is also fallible and not the best tool for every task. Give yourself the grace to be learner, much like you give grace to your students! With a little time, practice, perseverance, and thoughtful reflection, you will find places where AI helps you maximize your time and enhance your practice in meaningful, satisfying ways.
Most disappointing AI responses lack accuracy, relevance, or usefulness. Use the following suggestions to assist you with identifying and fixing potential issues of these dimensions.
Ways to address accuracy issues (when the response has inaccuracies or gives false information)
- Cross-reference with reliable sources. You can even cross-reference between two different AI models. Colleague AI offers “Clare” and “Charlie” —two AI assistants. It is possible a different AI will be more accurate.
- Ask Charlie (one of the AI assistants on Colleague AI) to provide citations and url links so that you can verify its sources and pinpoint the cause of the inaccuracy.
- Look for gaps in the AI’s existing logic. Chain of Thought prompting is helpful. Ask AI to break down the steps for you or share what it knows about the topic, and then fill in its gaps before having it try again on making your desired output.
Ways to address relevance issues (when the response doesn’t match the original ask or task)
- Review your original prompt and look for more details about your context that you can add.
- Rephrase your questions to be more targeted, much like you may rephrase and clarify instructions for a task in class when a student is confused or stuck.
Ways to address usefulness issues (when the response is difficult to follow or understand)
- Experiment with different output formats. Be specific about the length, tone, reading and vocabulary level of your students, table/text/or image, etc.
- Define AI’s role and offer AI details about your context.
- Ask AI to summarize or synthesize key insights if the AI responses are too long. This can include asking it to summarize into a specific small number (such as 3 or 5) of takeaways.
- Ask AI to translate the response into practical instructional strategies.
Conclusion
You can harness the power of Colleague AI to streamline your workflows, enhance student learning, and better support your classroom community by working on your ability to use these prompting strategies and staying mindful of potential issues. It may feel unnatural and inefficient at first, and that can be frustrating when the challenges eat up the limited time you have. Remember that sometimes we go slow at first to go fast later, just like we in our classrooms.
Hopefully this post provides strategies and tips that will help you on your journey with AI and get to the point where you are saving time and enhancing your practice more quickly with fewer challenges along the way. If there is a final key takeaway I can offer, it is to approach chatting with an AI assistant just as you would a real human colleague. Engage with AI in a natural, back-and-forth dialogue, and don’t be afraid to ask questions or bounce ideas off of either Charlie or Claire. They are here as virtual Colleagues to support you in your important work.
