This article explains what Custom GPTs are, why they are useful, how to create and share one in ChatGPT Edu, and important limitations and considerations.
Detailed Information
What Are Custom GPTs?
Custom GPTs are specialized versions of ChatGPT that you configure for a specific purpose. Think of them as AI assistants that have been given a focused job description: you define what they know, how they should behave, what files they can reference, and what tools they can use.
Once created, a Custom GPT can be used repeatedly by you or shared with colleagues, providing consistent, tailored responses without having to re-explain context every time.
Custom GPTs are available to ChatGPT Edu users and can be built entirely through a conversational interface. No coding is required.
Why Build a Custom GPT?
In a standard ChatGPT conversation, you start from scratch each time. You might find yourself repeatedly pasting the same instructions, uploading the same reference documents, or explaining the same context.
A Custom GPT eliminates that overhead by building your instructions and knowledge into a reusable assistant.
Custom GPTs are particularly valuable when you have a task that:
- Is performed frequently
- Follows a consistent pattern
- Relies on specific reference material
- Needs to be handed off to others who may not know how to prompt effectively on their own
How to Create a Custom GPT
Step 1: Open the GPT Builder
From the ChatGPT sidebar, click Explore GPTs, then click Create in the upper-right corner.
This opens the GPT Builder, which has two tabs:
- Create: Build your GPT conversationally by describing what you want. ChatGPT will help scaffold it.
- Configure: Control each setting directly.
Step 2: Define the Basics
In the Configure tab, set the following:
- Name: A clear, descriptive name for your GPT, such as "Syllabus Reviewer" or "Grant Budget Helper."
- Description: A short summary of what the GPT does. This is visible to anyone you share it with.
- Instructions: Detailed instructions that tell the GPT who it is, what it should do, how it should respond, and any rules or constraints it should follow.
Instructions are the most important field. Be specific. The more precise your instructions are, the more consistent the GPT's behavior will be.
Step 3: Add Knowledge Files
You can upload reference documents that the GPT will use to inform its responses. Supported examples include:
- PDFs
- Word documents
- Spreadsheets
- Text files
For example, you might upload a department style guide, a policy manual, a dataset, or a collection of frequently asked questions.
The GPT will draw on these files when answering questions, grounding its responses in your specific content rather than relying solely on its general training.
Step 4: Enable Capabilities
Choose which built-in tools your GPT can use:
- Web Browsing: Allows the GPT to search the internet for current information.
- Code Interpreter / Data Analysis: Lets the GPT write and execute Python code, analyze uploaded data, and generate charts.
- Image Generation: Enables the GPT to create images from text descriptions.
Enable only the capabilities your GPT actually needs for its intended purpose.
Step 5: Select a Model
ChatGPT Edu users can choose from multiple models when building a Custom GPT. Each model has different strengths. The most current model that OpenAI offers, as well as the Instant versions of these models, can be used.
Step 6: Add Conversation Starters
Conversation starters are prewritten prompts that appear when someone opens your GPT. They help users understand what the GPT can do and provide an easy way to get started.
For example, a syllabus reviewer GPT might include:
- "Review my syllabus for accessibility issues."
- "Suggest improvements to my learning objectives."
Step 7: Save and Share
When you save your GPT, choose its visibility:
- Only me: Private, for your personal use.
- Anyone with a link: Shareable with specific colleagues or teams within your organization.
GPTs shared through a link are accessible to other ChatGPT Edu users in your institution who have the link.
Tips for Writing Effective Instructions
The Instructions field is where the real customization happens. Use the following guidelines to produce more reliable results.
Be Explicit About the Role
Start by telling the GPT what it is.
For example:
You are a writing tutor who helps undergraduate students improve their academic essays. You never write content for the student. Instead, you ask guiding questions and suggest specific improvements.
Define Boundaries
Specify what the GPT should and should not do. If it should only answer questions about a particular topic, say so. If it should decline requests outside its scope, include that instruction.
Describe the Output Format
If you want responses in a particular structure, describe that format in the instructions.
Examples include:
- Bulleted feedback
- A scored rubric
- A specific report format
- A table with defined columns
- A step-by-step checklist
Reference Your Knowledge Files
If you have uploaded documents, tell the GPT when and how to use them.
For example:
When answering questions about university policy, always refer to the uploaded Employee Handbook PDF first before using general knowledge.
Include Examples
If there is a specific tone, style, or response pattern you want, include a brief example in the instructions so the GPT can pattern-match.
Use Case Examples
Academic and Teaching
Syllabus Builder: A GPT configured with your department's syllabus template, university policies on academic integrity, and accessibility guidelines. Faculty upload a draft syllabus, and the GPT reviews it for completeness, suggests improvements to learning objectives, flags missing required policy language, and formats it according to the template.
Assignment Feedback Assistant: A GPT loaded with a grading rubric and assignment instructions. Faculty paste in student work, and the GPT provides structured feedback aligned to the rubric, identifying strengths, areas for improvement, and specific suggestions. The GPT is instructed to provide feedback only and never rewrite the student's work.
Study Guide Generator: A GPT that takes a set of lecture notes, textbook chapters, or slide decks and generates review questions, key concept summaries, and practice problems. Configured with Study Mode principles, it can quiz students interactively rather than simply providing answers.
Research Literature Summarizer: A GPT with instructions to summarize uploaded academic papers in a structured format: research question, methodology, key findings, limitations, and relevance. This can help faculty or graduate students manage large reading loads during literature review phases.
Administrative and Operational
Policy FAQ Bot: A GPT loaded with HR policy documents, employee handbooks, or departmental procedures. Staff can ask natural-language questions such as "How do I request FMLA leave?" or "What is the travel reimbursement process?" and receive answers grounded in the actual policy documents rather than generic responses.
Meeting Notes Summarizer: A GPT configured to take raw meeting notes or transcripts and produce structured summaries with action items, decisions made, and follow-up owners. Instructions specify the output format your team prefers.
Job Description Drafter: A GPT loaded with existing job descriptions, classification guidelines, and institutional formatting standards. Hiring managers describe the role in plain language, and the GPT produces a properly formatted draft that aligns with HR requirements.
Email and Communication Drafter: A GPT configured with your department's tone, common communication templates, and brand guidelines. It helps staff draft professional emails, announcements, and other communications that maintain a consistent institutional voice.
Technical and IT
Documentation Writer: A GPT configured with your team's documentation standards and loaded with existing examples. Describe a process or system, and the GPT produces a formatted knowledge base article, runbook, or how-to guide that matches your team's style.
Code Review Helper: A GPT with instructions to review code snippets for common issues, including security vulnerabilities, style violations, and performance concerns, and suggest improvements. It can be configured with your team's coding standards or style guides.
Troubleshooting Assistant: A GPT loaded with known-issue documentation, common error messages, and resolution steps. Help desk staff or end users describe a problem in plain language and receive targeted troubleshooting guidance drawn from your support knowledge base.
Student Services
Advising Resource Guide: A GPT loaded with degree requirements, course catalogs, and advising frequently asked questions. It helps advisors quickly look up prerequisite chains, credit requirements, and common student questions. It can be configured to always recommend that students confirm details with their official advisor.
Financial Aid FAQ: A GPT with financial aid policies, deadlines, and common questions loaded as reference material. It provides quick answers to routine questions while directing students to the financial aid office for complex or case-specific situations.
Campus Resource Navigator: A GPT that helps students find the right campus resource for their needs, including tutoring centers, counseling services, career services, accessibility accommodations, and more. It can be loaded with a directory of services and configured to ask clarifying questions before directing students to a resource.
Limitations and Considerations
- They are not infallible. Even with good instructions and reference documents, GPTs can still produce incorrect or incomplete information. Users should always verify critical outputs.
- Knowledge files have size limits. Very large document sets may not be fully ingested. If the GPT needs to reference extensive material, prioritize the most important documents and keep them well organized.
- They do not learn from conversations. A Custom GPT does not update its knowledge based on user interactions. If your reference material changes, update the uploaded files.
- Data sensitivity matters. Follow university data-governance policies when deciding what to upload to a Custom GPT. Do not upload data classified as sensitive or restricted unless institutional policy explicitly permits it.
- Sharing is link-based. Anyone in your ChatGPT Edu instance who has the link can use a shared GPT. There is no fine-grained permission model, so consider what information is embedded in the GPT before sharing it widely.
Getting Started
To begin building your first Custom GPT, log in to ChatGPT Edu, click Explore GPTs in the sidebar, and select Create.
Start simple. Pick a single, well-defined task and iterate from there. Once you are comfortable with the basics, you can build more sophisticated GPTs with multiple knowledge files, specific output formats, and detailed behavioral instructions.
For questions about ChatGPT Edu access or assistance with building Custom GPTs, refer to the related articles in this knowledge base or contact the IT Help Desk.
Environment
- University of Maine System
- ChatGPT Edu
- Custom GPTs
- Eligible UMS students, faculty, and staff
- Supported desktop and mobile web browsers