AI Teaching Strategies: Designing Your Carmen AI Assistant

Providing timely support to your students is not always easy, particularly in courses with large enrollments or limited opportunities for one-on-one interaction. Students may have recurring questions about assignments, course expectations, or where to find important resources when you are not available to provide immediate guidance.

The Carmen AI Assistant extends instructional support beyond the classroom. You can customize the assistant to answer routine questions, direct students to relevant course materials, or provide coaching aligned with specific instructional goals. Students will receive course-specific guidance when they need it, allowing them to spend less time searching for answers and more time learning.

This guide will walk you through the process of designing, implementing, and assessing an AI assistant that complements your teaching and supports student learning.

What is the Carmen AI Assistant?

AI assistants are increasingly being explored as tools to extend instructional support and enhance the learning experience in higher education. Research suggests that AI-mediated learning assistants can provide personalized guidance and timely help outside of scheduled class time (Wollny et al., 2021; Schaaff et al., 2025). When designed intentionally, AI assistants can support a variety of teaching and learning goals, such as guiding reflection, facilitating brainstorming, helping students study, or pointing them to relevant resources.

The Carmen AI Assistant is an instructor-configured AI chatbot that is integrated within your CarmenCanvas course. Unlike general-purpose generative AI tools such as ChatGPT, Microsoft Copilot, or Google Gemini, the Carmen AI Assistant is designed around your course and its instructional goals. As the instructor, you define its purpose, role, and behavior through a system prompt and determine which course materials it uses as a knowledge base. As a result, the assistant provides responses that are grounded in course content and directly aligned with your instructional goals and learning outcomes.

For more on how the AI assistant works—and how to decide if it's the right fit for your course—visit the Carmen AI Assistant Instructor Resource.

Elements of an AI Assistant

Before you begin to design your AI assistant, it helps to understand the components that comprise one. The Carmen AI Assistant is not a single, uniform tool that looks the same across all courses in which it’s used. Rather, it is guided by programming known as a system prompt that tells it how to interact with a user. A system prompt is composed of interconnected, customizable elements that define the assistant’s behavior, tone, persona, and constraints. If you use the Carmen AI Assistant in your course, you will design a system prompt that meets your specific needs.

First, it’s important to understand the core components that make up an AI assistant and the design decisions that shape its behavior. In the table below, each element represents a point of control where you can decide how the assistant behaves, what it prioritizes, and how students engage with it in your course.

For an AI assistant to function effectively, it’s important to consider how all of these elements work together. For example, if an assistant’s role is to promote questioning, reflection, and idea development, it may not behave as intended if it’s positioned during high-stakes assignments rather than low-stakes practice activities or if its knowledge base includes materials that encourage answer-giving rather than reflection and sense-making.

Designing Your AI Assistant

Enabling the Carmen AI Assistant is not as simple as clicking a button or flipping a switch. Because you have full control over the assistant’s configuration, you must first make a series of instructional choices that shape how it interacts with students. This section will help you make the key pedagogical decisions that guide the design and behavior of your AI assistant.

Determine the Purpose

The first step in designing your Carmen AI Assistant is to determine its purpose. Will it support reflection, homework help, concept review, debate, or some other task? 

To ensure it behaves consistently and supports learning as intended, it is important to tailor your AI assistant to one specific purpose. If you ask the AI assistant to help students with reflection and offer homework help, it might not do either of those things particularly well. Luckily, there is no limit to the number of assistants you can add to a course, beyond what might be useful or manageable for students. If you have more than one pedagogical goal, you can consider creating a separate assistant for each task.

Consider the following questions to determine the best purpose for your AI assistant:

• Are there particular concepts, assignments, or areas in the course students struggle with that an AI assistant could address? 
• Could students use more support with a particular skill your course requires (e.g., reflection, writing or communication, critical thinking, analysis)?
• Are there common or frequent student questions that an AI assistant could answer using course materials?
• Is there a particular instructional challenge the AI assistant could address—for example, offering students feedback on their work or providing differentiated instruction?

No matter what you choose, as you clarify the assistant’s purpose, consider how that purpose intersects with your learning outcomes or other course elements. It’s important that the assistant supports students’ progress toward completing assignments and achieving outcomes in ways that reinforce, rather than hinder, their learning. 

Define a Role

Once you have determined your Carmen AI Assistant’s purpose, you are ready to define its role. How will the assistant engage students in learning? 

While the purpose and role of an AI assistant are closely connected, they serve distinct functions. The purpose of an AI assistant is primarily a pedagogical decision, reflecting the learning goal the assistant is intended to support, whereas the role translates that pedagogical intent into technical behavior. It shapes how the assistant interacts with students, including its stance, tone, constraints, and patterns of response. An AI assistant can embody a wide variety of roles, from “coach” to “facilitator” to “Socratic questioner.” 

The table below provides examples of specific AI assistant roles, how they might behave in a course, and the skills they might help students develop. Your own assistant may take on a different role or function depending on what is most useful for your students and your course.

As with the purpose, your AI assistant should serve only one role to maximize its effectiveness. To determine the role your assistant should play, think through how it can best support your pedagogical purpose and other course elements.

Consider the following questions as you select a role for your AI assistant:

• At what cognitive level (recall, analysis, synthesis) should the assistant operate for this course? Use the verbs in your learning outcomes to guide your response.
• Should the assistant respond to student prompts with direct answers and explanations or encourage critical thinking by posing guiding and clarifying questions? Use your assignments and learning outcomes to determine what might be most beneficial to students.
• Should the assistant model expert thinking or reasoning processes? Consider whether you want students to be able to see how the assistant arrived at a conclusion by making its reasoning explicit. 
• Should the assistant encourage students to engage in further reflection, justification, or exploration before offering feedback? Reflection and justification before feedback can reinforce deeper learning. 
• What student requests should the assistant redirect or decline to answer to preserve productive struggle? Productive struggle—working through challenges to construct meaning and deepen learning—can be undermined if an AI assistant provides answers too quickly or too completely.

Map Out Boundaries 

Defining your Carmen AI Assistant's boundaries is just as important as defining its purpose and role. Clear constraints on the assistant’s behavior reinforce alignment with learning outcomes and academic integrity expectations and help to prevent misuse or overreliance. Boundaries also help ensure that the AI assistant complements—rather than replaces—course activities, assignments, and students’ own efforts. 

To establish effective boundaries, you could specify that your assistant:

• Provides prompts and guidance rather than complete answers.
• Helps students plan assignments and reflect on learning but does not generate final submissions.
• Redirects students to course materials instead of directly answering questions.
• Is available only when students need targeted support, such as before introducing a challenging concept or during a complex assignment.

Making boundaries explicit in your AI assistant’s design—and, when appropriate, visible to students—reinforces its role as a learning support aligned with your pedagogical goals and students' needs.

Setting Up Your AI Assistant

Once you have made the key pedagogical decisions about your AI assistant’s design, you're ready to configure it for use in your course. This section will guide you through writing the system prompt, defining the assistant's knowledge base, and positioning it where it can best support student learning. Together, these steps translate your design decisions into an AI assistant that is ready for students to use.

Write the System Prompt

You have defined a purpose, role, and boundaries for your AI assistant. Now it's time to bring those decisions together in a clear system prompt.

The system prompt tells the assistant how to behave, interact with students, and operate within the boundaries you have established. There is no single “right” way to compose a system prompt, and you will likely adjust and refine your prompt as you test it and gain experience using it in your course. The following guidelines can help you create an effective starting point that aligns with your instructional goals.

One useful approach to writing a system prompt is the RTRI framework: Role, Task, Requirements, and Instructions. The decisions you made when designing your assistant fit naturally into each of these components.

The following resource from the University of Toronto provides more information on the RTRI prompt and technical writing tips: Developing an Effective System Prompt.

Another effective way to write a system prompt is to use generative AI as a drafting tool. Provide the AI with explicit design inputs—such as the assistant’s role, purpose, boundaries, learning outcomes, or example interactions—and ask it to generate an initial system prompt. 

Treat the AI-generated prompt as a starting point rather than a finished product. First, review the AI output to ensure it aligns with your instructional goals and the assistant’s intended role, task, and boundaries. Then revise the language to improve clarity and precision. Finally, test the prompt using realistic interactions, assess its performance, and refine it as needed.

Define the Knowledge Base

Your AI assistant uses the system prompt to guide its interactions with students, but it also needs a knowledge base—a designated source of information it can use to support student learning. The more targeted and relevant the knowledge base is to your system prompt, the more effectively the assistant can support students. By default, the Carmen AI Assistant has access to your Carmen files, which serve as its knowledge base. 

Review the course files in your Carmen shell to determine whether they provide the information your AI assistant needs to function as designed. Will the available files enable the assistant to complete the tasks described in your system prompt? Do they provide the information needed to support student learning? If the assistant needs additional context to function effectively, you can upload a Word, PowerPoint, PDF, or text file containing the relevant information.  

Depending on your assistant’s purpose and role, you might ensure your Carmen files include: 

• The syllabus and other documents addressing course policies for a course Q&A assistant.
• Readings, reference materials, or other resources for a Socratic tutor or study assistant.
• Assignment prompts and rubrics for an AI coach that provides feedback on student work.
• Scenarios, personas, or role-playing activities for a performance-based assistant.

Position the Assistant in Your Course

Now that you have developed a system prompt and reviewed your knowledge base, you're ready for the next important design decision: how to position the AI assistant in your course. As you decide where to embed your assistant, consider the points in your course where it will best support your instructional goals and student learning. Effective AI assistants have the greatest impact when their role, timing, and visibility align with the learning process they are designed to support—not simply when they are available everywhere (Beale, 2026). 

Position the assistant at moments of need that best align with your pedagogical intent. This could be near deadlines to support planning or self-checks, during multi-step assignments, or when students encounter challenging concepts or points of confusion. Instead of asking, “Where should the assistant live?” ask, “What learning behavior do I want the assistant to support?” This shift in thinking ensures the assistant’s placement reinforces your course design rather than unintentionally undermining it. 

You can position the assistant strategically to: 

• Support orientation and comprehension. Embed the assistant near the syllabus page, assignment overviews, or complex instructions when students are becoming familiar with expectations, terminology, or processes.
• Scaffold learning during challenging tasks. Position the assistant near assignments students commonly struggle with to prompt analysis, reflection, or decision-making (rather than supplying direct answers).
• Preserve productive struggle. Delay the assistant’s availability during graded problem-solving or synthesis tasks, positioning it as a reflective feedback tool after students have attempted the work on their own. 

There is no one correct position for your AI assistant. Keep its purpose and your course learning outcomes in mind as you determine where and when it will best support student learning.

Establishing Student Support

Like any new technology, the Carmen AI Assistant can present a learning curve for students. Providing clear guidance and support can help students understand the assistants’ purpose, use it responsibly, and get the most from it.  

There are many ways you can support students to effectively use the Carmen AI Assistant.

Image

Assessing Your AI Assistant

Just as you assess student learning in accordance with your learning outcomes, you should assess the effectiveness of the Carmen AI Assistant in accordance with your instructional goals. This will enable you to determine where and whether design tweaks or further student support is needed.

Evaluate the Assistant

Assessment of your AI assistant does not need to be formal or high‑stakes. Focus on gathering evidence of how the assistant is being used and whether it is supporting student learning as you intended. 

Consider the following approaches:

• Revisit the assistant’s original purpose. Did the assistant meaningfully support the learning goal it was designed to support (e.g., reflection, concept review, access to course information)? Where did it fall short? 
  • Where you look for evidence depends on the assistant’s purpose but could include evaluating whether there were improvements in students’ knowledge of course policies, understanding of course concepts, skills the tutor was meant to support (e.g., writing, reflection), or performance on exams or assignments. .
• Observe patterns of student use. Are students engaging with the assistant when planned? Are there points in the course when usage spikes or drops off, suggesting issues with placement or timing?
• Look for signs of pedagogical alignment or drift. Are students using the assistant in ways that align with its intended role, or is it being used as a substitute for course material or as a shortcut for completing learning activities?
• Solicit student feedback. Brief surveys, reflection prompts, or informal check‑ins can reveal how students perceive the assistant’s usefulness, clarity, tone, and boundaries.
• Reflect on instructional impact. Has the assistant reduced the frequency of common student questions, surfaced misconceptions, or allowed you to spend more time supporting students in meaningful ways during class time and office hours?

Assessing the AI assistant should be an ongoing process rather than a one-time effort. Use what you learn to identify what’s working, where improvements are needed, and how the assistant can better support your pedagogical goals and student learning.

Refine the Assistant

If your AI assistant is not performing in the manner you’d hoped, diagnose the issues and adjust the assistant accordingly. You may find that small, intentional refinements can significantly improve how the assistant supports student learning. Because you control the assistant’s configuration, these changes can be made incrementally without requiring a full redesign.

Consider making changes in the following areas:

• Refine the assistant’s purpose or scope. If the assistant is trying to do too much, narrow its focus to a single learning goal or instructional task. A more focused purpose often leads to more consistent and effective responses. 
• Adjust the system prompt. If the assistant is behaving in unexpected or unhelpful ways, revise the system prompt to clarify its role, tone, instructional stance, or boundaries. For example, you might emphasize guiding questions over explanations or reinforce limitations around answer-giving.
• Update the knowledge base. Add, remove, or curate your Carmen files to ensure the assistant is drawing from relevant, accurate, and current course materials. A more targeted knowledge base often leads to more useful and aligned responses.
• Reconsider placement and timing. If students are not using the assistant as intended—or not using it at all—or you see minimal benefits to learning, revisit where and when it is positioned in your course. You might embed the assistant in different modules, surface it more explicitly near assignments or deadlines, or reference it more intentionally in weekly course communications.
• Clarify guidance for students. If students seem unsure about when and how to use the assistant, revise how you introduce and explain its purpose. Provide transparent expectations for use, more explicit instructions, additional scaffolding, or examples of appropriate use.

As your course evolves and you learn more about how students use the assistant, continue making adjustments that keep it aligned with your instructional goals and responsive to students' needs.