How to create a custom GPT tutor
by Marcus Birkenkrahe
Source
This essay is based on the transcript of my impromptu video “How to create a custom GPT tutor” (YouTube), where I demonstrate the creation of a tutor for a course on operating systems. This idea came out of my “Coding in an AI World” course at Lyon College, spring 2026.
Introduction
Custom GPTs provide a way to transform a general-purpose AI conversation into a focused tool designed for a specific, repeatable task. Instead of navigating a broad dialogue in which the model continually offers suggestions and alternative directions, a custom GPT constrains the interaction to a defined workflow. This can be particularly useful in teaching and learning environments, where the goal is often structured practice rather than open-ended discussion.
The concept is not limited to a single platform. In ChatGPT these tools are called GPTs, while other systems such as Google Gemini use different names (for example, Gems). Regardless of the terminology, the underlying idea is the same: configure an AI model to behave like a specialized assistant tailored to a particular task.
Why Create a Custom GPT?
In a typical AI conversation, the system tends to propose new ideas or alternative directions at every step. While this flexibility is often useful, it can also be distracting when the user simply wants to complete a specific task.
Custom GPTs solve this problem by constraining the interaction. They are especially helpful when the task is:
- Repeatable
- Clearly defined
- Focused on a narrow goal
Examples include drills, quizzes, or structured exercises. Instead of engaging in long exploratory conversations, the AI performs the same type of task consistently and efficiently.
In educational settings, this makes custom GPTs particularly valuable as tutors that guide students through practice questions and exercises.
The Custom GPT Dashboard
When creating a custom GPT in ChatGPT, the interface presents three main areas:
-
Create Area: This is where the developer describes the purpose of the GPT and outlines the desired behavior. The system uses this description to generate an initial configuration.
-
Configuration Panel: The AI translates the description into structured settings such as the GPT’s name, description, instructions, and capabilities. These settings can be edited and refined.
-
Preview Panel: This area allows the developer to test the GPT interactively. Conversation starters and example prompts appear here so the behavior of the agent can be evaluated.
This three-part structure makes it possible to move quickly from idea to working prototype.
Defining the Role of the GPT
The most important step is defining the purpose of the agent clearly.
In the example demonstrated, the goal is to create a tutor for an undergraduate Operating Systems course (CSC420). The tutor is designed to:
- Drill students with practice questions
- Generate short quizzes
- Provide simple coding problems
- Draw exclusively on the course lecture material
Once the purpose is described, the system automatically generates an initial configuration, including the name of the agent, a description, and suggested conversation starters.
Conversation Starters
Conversation starters act as quick entry points for common interactions. For a tutoring system, these might include prompts such as:
- “Quiz me on a topic from the lecture material.”
- “Give me one practice question.”
- “Create a short operating systems quiz.”
- “Give me a simple coding problem to solve.”
These shortcuts make it easier for students to begin working with the tutor without having to formulate detailed prompts.
Refining the Tutor’s Behavior
The automatically generated instructions usually require further refinement. In particular, the instructor should define how the tutor evaluates student answers and provides feedback.
Typical adjustments include:
- Accept partially correct answers
- Provide brief feedback first
- Offer additional explanation only if requested
- Generate follow-up questions when students make mistakes
- Limit the number of follow-up questions
These refinements shape the learning interaction into a guided practice process rather than a long explanatory dialogue.
Providing Source Material
A crucial step in building an effective tutor is uploading the relevant course material.
For example, lecture slides or notes can be uploaded as PDF files. The GPT can then be instructed to generate questions only from those documents. This effectively creates a controlled sandbox in which the AI operates exclusively within the boundaries of the instructor’s material.
As the course progresses, additional files can be uploaded so that the tutor evolves with the curriculum.
This approach also eliminates much of the work traditionally involved in creating quizzes. Students can practice continuously in the AI environment instead of relying solely on static quizzes in a learning management system.
Selecting Capabilities
Custom GPTs can also be configured with specific capabilities depending on their intended use. Examples include:
- Web search for checking documentation or references
- Code interpreter for coding exercises or data analysis
- Image generation for visual explanations
- External app connections such as Google Drive
For a programming or systems course, enabling the code interpreter and web search can be particularly useful.
Testing the Tutor
After configuration, the tutor can be tested in the preview panel.
The instructor can ask the GPT to generate questions and then simulate student responses—both correct and incorrect—to observe how the tutor reacts. This stage often reveals issues that require adjustment, such as:
- Explanations that are too long
- Too many follow-up questions
- Unclear evaluation criteria
Iterative testing helps refine the behavior before the tutor is released to students.
Sharing the Custom GPT
Once the configuration is finalized, the GPT can be published and shared. ChatGPT provides several options:
- Private use
- Use within a workspace
- Shareable link for anyone with access
Students can then interact with the tutor directly through this link.
Conclusion
Custom GPTs offer a practical way to build specialized AI assistants tailored to specific workflows. In education, they can serve as tutors, practice systems, grading assistants, or feedback generators.
Because these agents can be tightly aligned with course materials and instructional goals, they often provide capabilities beyond those of traditional learning management systems. As these tools mature, they have the potential to reshape how instructors design practice activities and how students engage with course content.