Learn how to build custom GPTs that can perform specific tasks, follow detailed instructions, and provide tailored responses for your needs.
GPTs are used to create custom versions of ChatGPT that can be tailored for specific tasks.
→ Think of GPTs as personal assistants with specific, repeatable behaviors that are taught by crafting instructions.
→ GPTs require no programming skills and can easily incorporate specific academic content or institutional knowledge relevant to your courses or research areas.
→ GPTs can integrate seamlessly with educational tools and platforms, enhancing your teaching workflows and administrative efficiency.
Chat
→ Chat provides quick, general responses to a broad variety of questions without tailored institutional context, often requiring additional follow-up.
→ Chat has general knowledge without specific context, requiring further interaction and prompting.
→ Chat is ideal for immediate, everyday queries without requiring any setup.
GPTs
→ GPTs act as accelerators, helping automate repetitive tasks such as generating course outlines, providing student feedback, or summarizing research materials efficiently and consistently.
→ GPTs act as enablers, helping you tackle complex tasks without specialized skills, providing expert-level assistance in areas like data analysis, coding, or content creation.
Screenshot of the GPT configuration page
Name
A well-chosen name helps users easily identify the assistant's role.
Description
A good description sets clear expectations and helps the GPT stay focused on its intended tasks during interactions.
Poor description: Helpful for professors!
Stronger description: This GPT assists faculty in navigating university policy documents and internal guidelines. For detailed budgeting inquiries, please refer to Budget Assistant GPT.
Instructions
An effective custom GPT has clear, direct, and helpful instructions, think of them as having a conversation with your assistant.
Simple, but effective examples of instructions:
Analyzing budget reports for school departments:
This GPT analyzes departmental or school-wide budget reports and highlights any major discrepancies or spending trends. It provides a bullet-point summary of these findings, making it easy for administrators to spot key areas of concern or opportunity.
Assisting with IT and classroom technology troubleshooting:
This GPT reviews classroom technology setups (e.g., projector connections, learning management system configurations) and identifies potential issues. It responds with a list of those issues, along with a short explanation of each one, helping IT staff and faculty resolve problems quickly.
Supporting students in Introduction to Psychology
You are a teaching assistant for the Introduction to Psychology course (PSY101) at [University Name]. Your primary goal is to support students by answering course-related questions, explaining key concepts, and providing study resources while adhering strictly to the course syllabus and materials provided.
The best instructions add content and detailed steps, refer to knowledge files, and specify the response format of the GPT. A framework for more detailed, crafted instructions:
## **Context**
Provide a brief background that explains the role and purpose of the Custom GPT. The context sets the stage for how the assistant will understand user inputs and structure its outputs. Be clear about the users’ level of expertise and the assistant's specific tasks.
## **Instructions**
List clear, numbered steps that the GPT should follow in sequence. Use sub-bullets for additional details or explanations within each step.
1. Ask the user to upload the relevant data file (CSV or Excel format).
- Ensure the file contains key marketing metrics such as impressions, clicks, and conversion data.
- If the user uploads an incorrect file format, provide a clear message explaining the acceptable formats.
2. Analyze the data for key metrics:
- Calculate conversion rate (CR), cost per acquisition (CPA), and return on ad spend (ROAS).
- Present these metrics in a structured table format.
3. Suggest follow-up analyses based on the initial insights:
- Recommend deeper analysis of channels with the highest ROI.
4. Confirm with the user after each step and ask if they’d like additional analyses or reports.
## **Response Format**
Define how the assistant should present its responses. Different tasks may require different formats, so provide clear guidelines for various scenarios.
## **Reference to Knowledge**
If your GPT requires specific knowledge files or external references, list them here.
Conversation Starters
Conversation starters guide users by showcasing the assistant's capabilities and providing examples of interaction. These starters are presented on the "chat" interface within a GPT and serve as sample prompts.
Sample conversations starters for a Lecture Recall GPT:
Summarize last week's lecture.
Explain key points from today's class.
List main concepts covered this month.
Provide review questions from recent lectures.
Knowledge
Knowledge helps the GPT understand unique terminologies, workflows, or internal documents of your organization, making its outputs more contextually relevant. You can upload up to 20 files that are 512MB each within a GPT.
To maximize the effectiveness of your GPT, it's important to properly prepare and structure your knowledge files.
PDF files are often more effective and reliable than PowerPoint presentations or Word documents.
It’s highly recommended to only upload text files that do not have multiple columns.
Capabilities
Capabilities are checkboxes that define a GPT's technical powers and the scope of tasks it can handle:
Web Search
Canvas
DALL·E Image Generation
Code Interpreter & Data Analysis
If you only want your GPT to reference internal documentation that has been pre-uploaded as Knowledge (i.e. you are building an HR Helper that references company policies), be sure to leave Web Search unchecked.
Actions
Actions allow your GPT to perform task outside of the ChatGPT UI via API integrations. GPTs can perform specialized tasks such as fetching data, triggering workflows, or updating external systems.
OpenAPI's standard spec defines the GPTs action
Authorizations supported:
No Auth
API Key/Secret
OAuth (recommended)