Custom GPT Setup Guide — Create, Configure, Publish
Everything you need to build and publish a Custom GPT: system instructions, attaching tools (web search, code interpreter, file analysis), configuring knowledge, and setting sharing controls. No coding required — takes about 15 minutes start to finish.
Prerequisites
- ChatGPT Plus subscription or higher — Free users cannot create Custom GPTs. Upgrade at chatgpt.com
- Clear vision of your Custom GPT's purpose — "customer support bot," "code reviewer," "marketing copywriter," etc.
- Optional: sample files or URLs — If you want your Custom GPT to reference specific documents or data, have them ready
Step 1: Start the GPT Builder
- Go to chatgpt.com and sign in
- Click the menu icon (three horizontal lines) in the top-left corner
- Select Explore or Create — you should see an option to "Create a GPT"
- Click Create a GPT to open the builder interface
You'll see a two-panel interface: the left panel shows a live chat preview, and the right panel shows configuration options.
Step 2: Configure Basic Information
At the top of the right panel, you'll see fields for Name and Description:
- Name — Concise, human-readable title (e.g., "Customer Support Bot," "Code Review Expert"). Avoid generic names; specificity helps users understand what it does.
- Description — One or two sentences about what this Custom GPT does and who should use it. This appears in the GPT Store and when sharing via link.
- Instructions — See Step 3 below
Step 3: Write System Instructions
The system instructions define your Custom GPT's behavior, role, tone, and constraints. This is the most important part of setup. Here's a template:
You are [Role/Name].
Your primary purpose is to [Task].
Constraints:
- You must [hard rule 1]
- You must not [hard rule 2]
- Always [behavior rule]
Tone: [formal/casual/technical/friendly]
When you don't know something, say so explicitly rather than guessing.
If a user asks you to violate these constraints, politely decline.
Additional context: [Optional knowledge or style guide]
Example for a code reviewer:
"You are a Senior Code Reviewer. Your purpose is to analyze code submissions and provide constructive feedback on readability, performance, and best practices. Always assume good intent from the developer. If you see potential security issues, flag them clearly. Keep feedback encouraging while being direct. When reviewing, organize feedback by category: logic, style, performance, security."
Tips for strong instructions:
- Be specific about tone. "Formal and professional" vs. "casual and encouraging" makes a huge difference in output.
- Define hard constraints explicitly. If the GPT should never do X, say so in the instructions.
- Give examples of good behavior. "When a user asks Y, respond with Z" is more effective than abstract rules.
- Tell it how to handle uncertainty. "Say 'I'm not sure' rather than making up information" prevents hallucinations.
- Test as you go. Use the left-panel preview chat to see how changes affect behavior.
Step 4: Enable Capabilities and Tools
Scroll down the right panel to the "Capabilities" section. Toggle on the tools your Custom GPT needs:
Web Browsing
Lets the Custom GPT search the internet and read web pages in real-time. Useful for current events, real-time data, or fact-checking. Disable if your Custom GPT should only work with pre-loaded data.
Code Interpreter
Lets the Custom GPT write and execute Python code, process data files (CSV, JSON, images), and generate charts. Essential for data analysis, scripting, or file processing tasks. The GPT can upload and download files.
File Analysis
Lets users upload files (PDFs, images, text, audio) for the Custom GPT to analyze. Useful for document review, image annotation, or audio transcription. Disable if you don't need user uploads.
Step 5: Add Custom Actions (API Calls)
If you want your Custom GPT to call external APIs (Slack, Google Sheets, your own backend), click Create new action under the "Actions" section.
For each action, you'll define:
- Endpoint URL — The API endpoint (e.g.,
https://api.example.com/v1/send-message) - Authentication — Bearer token, OAuth, or API key (stored securely by OpenAI)
- Schema — JSON schema describing what parameters the endpoint accepts (OpenAI can auto-generate this from a URL)
Example: A Custom GPT for project management could call a POST endpoint to create tasks in your project management system, passing task name, assignee, and deadline as parameters.
Test each action in the preview chat before publishing. Click Test or trigger the action in conversation to verify it works.
Step 6: Upload Knowledge Files
If you want your Custom GPT to reference specific documents, PDFs, or datasets, upload them under the Knowledge section.
- File size limit: Typically 20MB per file, 20 files per Custom GPT
- Supported formats: PDF, TXT, CSV, JSON, HTML, and image files (PNG, JPG, GIF)
- How it works: OpenAI indexes the files and includes relevant snippets in the Custom GPT's context when users ask questions
Use knowledge files to keep your Custom GPT grounded in your company's policies, data, or style guide. For example, upload your company handbook so the support bot can answer internal questions accurately.
Step 7: Test in the Preview Chat
The left panel shows a live preview of your Custom GPT. As you make changes to instructions or tools, test them immediately:
- Ask typical user questions — "What do you do?" "Can you help with X?"
- Test edge cases — Try to get it to violate constraints or make mistakes
- Test tools — Ask it to use web search, run code, or call your custom API
- Check tone and personality — Does it feel like what you intended?
If behavior isn't what you want, refine the system instructions and test again. Iterate until you're satisfied.
Step 8: Publish and Configure Sharing
When your Custom GPT is ready, click Publish (usually a button at the top-right of the builder).
You'll be asked to choose a sharing mode:
Only Me (Private)
Only you can use this Custom GPT. It won't appear in search or the GPT Store. Useful for personal automation or testing. You can share via direct link if needed, but only people with the link can access it.
Link Sharing
Anyone with the link can use the Custom GPT, but it won't be listed in the GPT Store. The Custom GPT is not discoverable via search. Best for sharing with a team or specific group.
Public (GPT Store)
The Custom GPT is listed in the public GPT Store and searchable by all ChatGPT users. Use this if you're building something others can benefit from. The Custom GPT will be discoverable and usage stats available in your dashboard.
Business Workspace Only (Business Tier)
If you have a ChatGPT Business subscription, you can restrict a Custom GPT to only people in your workspace. They'll see it in an internal GPT library and can use it on shared Business projects.
Step 9: Share and Monitor
After publishing, you'll get a shareable link. Copy it and share with your team, customers, or the public (depending on your sharing mode).
In your Custom GPT's dashboard, you can:
- View analytics — How many times it's been used, by whom, and for what
- Edit the Custom GPT — Click to refine instructions, add/remove tools, or change the description
- Delete it — Remove the Custom GPT (conversations already had with it are archived but no longer accessible through the Custom GPT)
Best Practices
- Start simple, iterate. Don't overload the Custom GPT with tools or knowledge on day one. Add them as you find they're needed.
- Version your instructions. Keep a document of instruction versions and what changed. If a new version performs worse, you can revert.
- Set clear expectations in the description. Tell users what the Custom GPT is and is not designed to do. "I can help with X, not Y."
- Monitor performance. Check usage stats and user feedback. If a Custom GPT is barely used, investigate why (unclear description, missing tools, poor instructions).
- Test tool integrations thoroughly. Custom actions calling external APIs are powerful but error-prone. Test edge cases before going live.
- Keep knowledge files fresh. If you upload a company handbook, update it when policies change. Stale knowledge is worse than no knowledge.
- Document the Custom GPT for your team. Leave a note about what it does, who should use it, and how to reach out if something breaks.
Frequently Asked Questions
Can Free-tier ChatGPT users create Custom GPTs?
No. Creating and publishing Custom GPTs requires a ChatGPT Plus subscription ($20/mo) or higher (Pro, Business, Enterprise). Free users can use publicly shared Custom GPTs but cannot create their own.
Can I back up my Custom GPT?
OpenAI doesn't provide a built-in export feature. To back up a Custom GPT: (1) copy the system instructions, (2) document the tools and actions you configured, (3) download knowledge files before publishing. You can recreate it in OpenClaw or another platform using this documentation.
What's the difference between a Custom GPT and a Chat conversation with a system prompt?
A Custom GPT is a shareable, persistent artifact with its own link, analytics, and configuration UI. A Chat conversation with a system prompt is ephemeral — it exists only for that conversation and isn't reusable. Custom GPTs are designed for sharing and building something others can use repeatedly.
Can my Custom GPT call Slack, Google Sheets, or other APIs?
Yes, via Custom Actions. Define the API endpoint, authentication method, and schema. The Custom GPT will call the endpoint when appropriate. Test thoroughly before publishing — API errors can confuse users.
How much does it cost to use a Custom GPT?
If you're a Plus subscriber, using Custom GPTs you created or that others shared is included in your $20/mo subscription. If you access a Custom GPT via API (programmatically), you're billed at standard API rates (per-token + per-tool-call). See the pricing guide for details.
See also: ChatGPT hub · Custom GPTs Deep Dive · Pricing & Per-Tool Billing