General-purpose AI handles general-purpose tasks. When you need an assistant that knows your company's style guide, references your product documentation, or connects to your internal APIs, you need a custom GPT. ChatGPT's GPT Builder lets anyone create one — no coding required.
Describe what your GPT should do in plain English. Upload knowledge files. Connect external services. Publish to the GPT Store and earn revenue from user engagement. The entire process takes minutes, not months.
ChatGPT Custom GPTs are specialized AI assistants built on top of GPT-4 and GPT-4o. The GPT Builder provides a conversational creation interface alongside a manual Configure panel. Builders define custom instructions (personality, rules, domain expertise), upload up to 20 knowledge files totaling approximately 2 million tokens, enable built-in tools (web browsing, DALL-E, Advanced Data Analysis), and configure external API actions via OpenAPI schemas. Publishing options include private (personal use), link-shared, and public (GPT Store listing with revenue sharing). Creating custom GPTs requires a ChatGPT Plus, Team, or Enterprise subscription. Using published GPTs is available on all plans. All custom GPTs operate within SOC 2 Type II, GDPR, CCPA, and ISO 27001 compliance boundaries.
Two paths to creation: conversational GPT Builder or manual configuration. Both produce the same result.
Click your profile icon, select "My GPTs," then "Create a GPT." The GPT Builder opens a split-screen view: a conversational interface on the left and a live preview on the right. Tell the Builder what your GPT should do. "I want a GPT that helps real estate agents write property listings. It should use a professional but warm tone, include neighborhood data, and follow MLS formatting rules."
The Builder asks clarifying questions. Should it generate images of properties? Should it access web data for recent comparable sales? How formal should the language be? Your answers become the GPT's instruction set. The live preview lets you test responses immediately, refine tone, and adjust behavior before saving.
This conversational approach eliminates the need for prompt engineering expertise. Describe outcomes, not technical configurations. The Builder translates your intent into a structured instruction set that the underlying model follows consistently across every user conversation.
Switch to the Configure tab for direct access to every setting. Write custom instructions in a text field — as detailed or as minimal as needed. Upload knowledge files that the GPT references when answering questions. Enable or disable web browsing, DALL-E image generation, and Advanced Data Analysis individually. Add API actions by providing an OpenAPI schema URL or pasting the JSON specification directly.
The conversation starters field lets you define suggested prompts that appear when users first open your GPT. These guide new users toward the GPT's strongest capabilities and reduce friction. A tax preparation GPT might suggest: "Estimate my 2024 federal tax liability" or "What deductions apply to home office workers?"
Experienced builders prefer the Configure tab because it exposes the full instruction text for revision, allows copy-paste of tested prompt patterns, and provides precise control over tool permissions. Both approaches produce identical GPTs.
Upload documents that your GPT references when answering questions. Supported formats include PDF, DOCX, TXT, CSV, JSON, PPTX, XLSX, Markdown, and source code files. The combined limit is approximately 2 million tokens across up to 20 files — that is roughly 1.5 million words or about 3,000 pages of text.
ChatGPT uses retrieval-augmented generation (RAG) to search these files. When a user asks a question, the system identifies the most relevant sections from your uploaded documents and includes them in the context window. This grounding reduces hallucinations because the GPT bases answers on your specific source material rather than its general training data.
A consulting firm uploads its methodology framework, case study archive, and client presentation templates. The resulting GPT generates proposals in the firm's voice, references real case outcomes, and follows established frameworks. An open-source project uploads its entire documentation set, creating a support GPT that answers contributor questions with accurate, version-specific guidance. According to the National Science Foundation (NSF), retrieval-augmented approaches represent a significant advancement in reducing factual errors in AI-generated content.
Actions let your custom GPT call external APIs during conversations. Define endpoints using OpenAPI 3.0 schema specification. The GPT handles authentication (API keys, OAuth 2.0), constructs requests from user input, and parses responses into natural language answers.
Practical examples: a CRM GPT that pulls customer records from Salesforce before drafting follow-up emails. A project management GPT that creates Jira tickets from conversation summaries. A finance GPT that retrieves live stock data from a market data API. A travel GPT that queries airline and hotel booking systems.
Users interact in natural language without seeing or understanding the underlying API mechanics. The GPT determines when to call which endpoint based on the conversation context. Rate limiting, error handling, and retry logic follow the API provider's specifications. For full programmatic integration, see our API Access page.
Every setting available when building a custom GPT through ChatGPT's GPT Builder.
| Configuration | Description | Limits | Required |
|---|---|---|---|
| Name | Display name shown in the GPT Store and sidebar | 50 characters max | Yes |
| Description | Short summary explaining the GPT's purpose | 300 characters max | Yes |
| Instructions | Custom system prompt defining behavior, tone, rules, and constraints | ~8,000 characters | Yes |
| Knowledge Files | Documents the GPT references via RAG retrieval | 20 files, ~2M tokens total | No |
| Conversation Starters | Suggested prompts shown on first interaction | 4 starters | No |
| Web Browsing | Enable live Bing search capability | Toggle on/off | No |
| DALL-E | Enable image generation within conversations | Toggle on/off | No |
| Code Interpreter | Enable Python execution and file processing | Toggle on/off | No |
| Actions (APIs) | External API endpoints via OpenAPI 3.0 schema | Multiple endpoints per action | No |
| Authentication | API key or OAuth 2.0 for action endpoints | Per action set | If actions used |
| Profile Image | Custom icon displayed in the GPT Store and sidebar | DALL-E generated or uploaded | Yes |
| Publishing | Private, link-shared, or public (GPT Store) | Three visibility levels | Yes |
The gap between a general AI assistant and a custom-built specialist is the gap between "helpful" and "indispensable."
Upload your product documentation, FAQ database, troubleshooting guides, and return policy. The resulting GPT answers customer questions with accurate, brand-consistent responses 24 hours a day. It cites specific documentation sections, follows your escalation rules (redirecting billing disputes to human agents), and maintains your company's tone of voice. A mid-sized SaaS company reduced their support ticket volume by 38% within two months of deploying a custom GPT trained on their knowledge base.
A law firm uploads contract templates, precedent documents, jurisdiction-specific statutes, and internal style guides. The custom GPT drafts NDAs, employment agreements, and vendor contracts that follow firm standards. It flags missing clauses, suggests alternative language for ambiguous terms, and cross-references uploaded case law. Attorneys review and refine rather than drafting from scratch, cutting document preparation time by roughly 60%.
A physics professor creates a GPT loaded with course textbooks, problem sets, and grading rubrics. Students ask questions and receive explanations calibrated to their course level — not generic internet explanations that may be too advanced or too basic. The GPT walks through derivations step by step, references specific textbook sections, and generates practice problems tailored to upcoming exams. The U.S. Department of Education (ED.gov) has published research on how AI tutoring tools improve learning outcomes when aligned with specific curricula.
Enterprise teams build GPTs loaded with HR policies, engineering documentation, onboarding materials, and process guides. New employees ask "How do I request PTO?" or "What is our deployment checklist?" and receive instant, accurate answers sourced from official company documents. This eliminates the Slack-channel-search-and-hope pattern and keeps institutional knowledge accessible even as team members change roles. Learn more about ChatGPT plugins and tools that enhance these workflows, or compare what GPT models power your custom GPT behind the scenes.
From private tool to public product — the GPT Store creates a marketplace for specialized AI assistants.
Publish your custom GPT to the GPT Store and earn revenue based on user engagement. The program launched in 2024, with payments calculated from a combination of unique user counts and conversation volume. Builders access analytics dashboards showing total users, returning users, average session length, and conversation counts. Top-performing GPTs serve specific professional niches — tax preparation, academic writing, coding interview prep — where domain expertise creates defensible value.
Verify your identity through a domain or social media account to display a verified badge on your GPT Store listing. Verified builders receive higher visibility in search results and category rankings. The verification process takes minutes and significantly increases user trust and adoption rates for public GPTs.
Team and Enterprise admins control which custom GPTs are available in their workspace. Admin-approved GPTs appear in the team sidebar while unapproved store GPTs remain inaccessible. This governance layer lets organizations deploy specialized internal GPTs while maintaining security policies. Admins set data retention rules, audit conversation logs, and manage GPT access permissions at the user and group level. Review Enterprise plan details for full admin capabilities.
ChatGPT Plus subscribers can create, configure, and publish custom GPTs in minutes. No coding required.
Get Started FreeDetailed answers about building, publishing, and using custom GPTs.
Click your profile icon in ChatGPT, select "My GPTs," then click "Create a GPT." The GPT Builder opens with two tabs: a conversational Create tab where you describe your GPT in natural language, and a Configure tab for manual control over instructions, knowledge files, tools, and API actions. Test your GPT in the live preview panel, refine as needed, then save with your preferred visibility setting (private, link-shared, or public GPT Store listing). Creating GPTs requires a ChatGPT Plus, Team, or Enterprise subscription. Learn about the GPT models that power your custom assistants.
Custom GPTs maintain persistent instructions that apply to every conversation without re-prompting. They reference uploaded knowledge files for source-grounded answers. They connect to external APIs for live data retrieval and actions. They enforce domain-specific rules, output formats, and behavioral constraints automatically. Regular ChatGPT starts each conversation with only its general training data and your initial prompt. A custom GPT begins with deep context about your specific domain, documents, and workflows already loaded.
Yes. The GPT Store revenue-sharing program pays creators based on user engagement metrics including unique user counts and conversation volume. Publish your GPT to the store with a verified builder profile to participate. Top earners build GPTs that serve specific professional niches with deep domain expertise. Analytics dashboards track performance metrics. Payments process according to the program terms. Visit the Plugins page for details on tools that enhance custom GPT capabilities.
Custom GPTs accept up to 20 knowledge files with a combined maximum of approximately 2 million tokens (roughly 1.5 million words or 3,000 pages). Supported file types include PDF, DOCX, TXT, CSV, JSON, PPTX, XLSX, Markdown, and source code files (.py, .js, .ts, .html, .css). ChatGPT uses retrieval-augmented generation (RAG) to search these files and include relevant sections in the context window when answering user queries.
Custom GPTs inherit ChatGPT's full security infrastructure. Knowledge files are encrypted at rest with AES-256 and in transit with TLS 1.3. The platform maintains SOC 2 Type II certification, GDPR compliance, and CCPA compliance. Enterprise and Team plans add admin-approved GPT whitelisting, data isolation, conversation audit logging, and configurable data retention policies. Instructions and knowledge file contents are not visible to end users unless the builder explicitly chooses to expose them. Review our security page and plan comparison for full details.
Custom GPTs work alongside every other ChatGPT capability.
Understand which GPT models power your custom GPTs and how they affect response quality.
Interact with your custom GPTs through natural voice conversation on mobile and desktop.
Enable image analysis within custom GPTs for visual document processing and photo understanding.
Add web browsing, DALL-E, and Code Interpreter capabilities to your custom GPTs.
Build custom GPT-like experiences programmatically with the Assistants API and function calling.
See which plans include GPT creation, workspace GPTs, and enterprise admin controls.
ChatGPT