Build Your Own AI Assistant: A Practical Guide to Custom GPTs for Educators Featuring Insights from Dr. James Basham, Dr. Matthew Marino, and Dr. Trey Vasquez
Authors: Philip Garza and Yerin Seung; info@ciddl.org
Artificial intelligence does not have to be a black box. In a recent CIDDL webinar, Drs. James Basham, Trey Vasquez from the University of Kansas, and Matthew Marino from the University of Central Florida walked educators through one of the most practical applications of large language models available today: building a custom generative pre-trained transformer, or GPT, trained on information you already trust.
The key insight driving their conversation was simple. When you ask a general AI tool a question, it searches the open web and returns whatever it finds. When you build a custom GPT, you control the corpus. The model goes to your documents, your policies, your research, and answers from there.
[embedyt] https://www.youtube.com/watch?v=ncIP11d4rlY[/embedyt]
What Is a Custom GPT, Exactly?
The word "GPT" gets used interchangeably with ChatGPT, but the panel was quick to clarify that a generative pre-trained transformer is a category, not a brand. Whether you use OpenAI, Google Gemini, Claude, or another platform, the same general approach applies: give the model a name, a set of instructions, and a knowledge base, and it will draw from those resources when users ask questions.
Dr. Basham demonstrated this live with the High Leverage Practices GPT, or HLPGPT, a tool he built for a course he teaches. The HLP GPT is trained on publicly available research tied to the 22 High Leverage Practices in special education. Students and educators can ask it questions about instructional strategies, and it returns responses grounded in that specific research base, with direct references to the sources it used. Thousands of users have interacted with the tool since its creation.
Why This Matters for Special Educators and Teacher Preparation
Dr. Marino offered a scenario educators will recognize immediately. A special education teacher receives an email from a parent asking about district policy. Traditionally, the teacher opens the district website, searches through dozens of PDF documents the school board has posted over the years, and tries to locate the relevant passage. A custom GPT changes that workflow entirely. Load those documents into the knowledge base, and the same question gets a direct, cited answer in seconds.
The applications extend well beyond policy lookup. The panel described uses in human resources, budget planning, and leadership contexts where institutional policies shape everyday decisions but rarely sit at anyone's fingertips. Dr. Basham shared examples from his own teaching: a GPT trained on a course syllabus to serve as a teaching assistant, and another trained on a case study for a multi-tiered systems of support course. Dr. Marino pointed to K-12 teachers loading open-source textbooks and reference materials so students can ask questions at any level of understanding and receive responses tailored to how they learn best.
That last application carries real implications for differentiated instruction. A student who did not grasp a concept during direct instruction can return to the GPT and ask again, phrased differently, as many times as needed.
How to Build One
Both major platforms follow a similar setup process. In Google Gemini, custom tools are called Gems. In OpenAI, they are called GPTs. In either case, the builder provides four basic elements: a name, a description, instructions for how the tool should behave, and a knowledge base of uploaded files or linked resources.
The instructions deserve particular attention. Dr. Basham shared a lesson from the HLP GPT: after receiving user feedback, he realized the tool was not handling a specific practice correctly. He added a targeted instruction addressing HLP 12 directly, tested it, and the problem was resolved. Instructions function like both a personality and a rulebook. If the tool is not performing as expected, refining those instructions is the first step.
The panel offered two additional strategies worth keeping close. When the tool gives a response you know is wrong, write the response you wanted, return it to the GPT, and tell it directly to learn from this and remember it. That phrasing embeds the correction into the tool's memory and adjusts subsequent responses. And if you are struggling to write an effective prompt, ask the AI itself. Describe what you are trying to accomplish and ask the model to help refine the prompt. The resulting prompt is almost always more thorough and produces a better outcome than the original.
Critical Considerations Before You Start
The panel was consistent on privacy and security. Dr. Vasquez emphasized not to put personally identifiable information, student records, or anything governed by the Family Educational Rights and Privacy Act (FERPA) into a free or personal AI account. If your institution has an enterprise license with a data sharing agreement in place, check with your IT department to understand what is permissible under those terms before uploading anything.
For K-12 teachers, Dr. Vasquez also recommended confirming administrator support before introducing custom GPTs in the classroom. For faculty in higher education preparing future teachers, the argument was stronger: building AI literacy and fluency in preservice teachers is not optional. These tools are part of the professional environment students are entering, and teacher preparation programs have a responsibility to address them directly.
Start Small, Iterate Often
The strongest takeaway from the webinar was the importance of iteration. A custom GPT improves through use. The first version will not be perfect, and stepping away when it is not is the one mistake worth avoiding. Refine the instructions, adjust the knowledge base, test with colleagues, and keep going.
The panel also suggested taking the next step with students: teach them to build their own. When students construct a GPT around information meaningful to their own challenges and goals, they engage with the tool in a fundamentally different way and, more often than not, discover applications their teachers had not considered.
The resources to build these tools are already available. The research base, the policy documents, and the course materials are already there. A custom GPT makes it easier for you and the students you serve to access all of it.
Join the conversation in our community!
CIDDL is committed to providing high-quality resources to support the increasing knowledge, adoption, and use of a range of educational technologies that can be used for educators, related services, or leadership preparation programs. For more resources, including videos and blogs, subscribe to our newsletter and follow us on YouTube, Facebook, and LinkedIn. The most important part of our CIDDL community is YOU. Join our community and share the innovative ways you are using technology, ask a question about technology integration, or participate in our bi-weekly live AI Community Chats. We look forward to seeing you in our community!