Building Custom GPTs
From the course ChatGPT Projects & Custom GPTs
Custom GPTs are specialized ChatGPT assistants you configure for specific tasks. They combine instructions, knowledge, and capabilities into shareable AI tools.
Creating a Custom GPT
- Go to "Explore GPTs" → "Create"
- Use the GPT Builder (conversational) or Configure tab (manual)
- Define name, description, and instructions
- Upload knowledge files
- Enable capabilities (web browsing, DALL-E, code interpreter)
- Set conversation starters
Instruction Writing
Write clear, specific system instructions:
You are a senior Python code reviewer. When given code:
1. Check for PEP 8 compliance
2. Identify potential bugs and edge cases
3. Suggest performance optimizations
4. Rate code quality on a scale of 1-10
5. Provide corrected code with comments
Always explain WHY a change is recommended, not just what to change.
Never suggest changes that would break existing functionality without noting it.
Knowledge Files
Upload domain-specific documents:
- Product documentation for support GPTs
- Style guides for writing GPTs
- Technical specs for engineering GPTs
- Training materials for educational GPTs
- FAQs and playbooks for operational GPTs
Capabilities
| Capability | Use Case |
|---|---|
| Web Browsing | Research, fact-checking, current events |
| DALL-E | Image generation, visual content |
| Code Interpreter | Data analysis, calculations, file processing |
Actions (API Connections)
Connect Custom GPTs to external services:
- Fetch data from your company's API
- Create records in CRM systems
- Send notifications via webhooks
- Query databases for real-time information
- Integrate with project management tools
Named Example: Zapier's GPT Action Library
Zapier built one of the largest libraries of Custom GPT Actions, enabling GPTs to trigger 6,000+ app automations — creating Trello cards, sending Slack messages, or adding rows to Google Sheets — directly from a conversation. Their "Zapier Natural Language Actions" API is a blueprint for how to connect Custom GPTs to any REST API without requiring users to leave ChatGPT.
// Example Custom GPT Action schema snippet
{
"openapi": "3.0.0",
"info": { "title": "My CRM API", "version": "1.0" },
"paths": {
"/contacts": {
"post": {
"summary": "Create a new contact",
"requestBody": {
"content": {
"application/json": {
"schema": {
"type": "object",
"properties": {
"name": { "type": "string" },
"email": { "type": "string" }
}
}
}
}
}
}
}
}
}
Common Pitfall: Custom GPT knowledge files are chunked and retrieved with basic semantic search — they don't give the GPT perfect recall of every sentence. For critical facts (pricing, legal terms, product specs), include them directly in the system prompt rather than burying them in a knowledge file where retrieval is unreliable.
Key Takeaways
- •Custom GPTs combine instructions, knowledge files, and capabilities
- •Detailed system instructions produce more consistent, specialized behavior
- •Knowledge files add domain expertise without retraining
- •Actions connect GPTs to external APIs for real-world functionality
Progress is kept in this browser only.