How to Create Custom GPTs: Complete Builder Guide (2026)

A custom GPT is a saved ChatGPT configuration with its own instructions, uploaded knowledge files, and optional apps or API actions, built in the GPT builder without writing code. To create one, open Explore GPTs in ChatGPT, select Create, describe what the GPT should do or fill in the Configure view, attach reference documents, optionally connect an app or an external API through Actions, test it in Preview, and save it. As of September 16, 2026, personal accounts (Free, Go, Plus, and Pro) cannot create or publish new GPTs, so building one requires a Business, Enterprise, or Edu workspace that allows it, and OpenAI is planning to retire custom GPTs and migrate them to plugins. A custom GPT runs inside ChatGPT and cannot be embedded in your own website, although its Actions can call your own API.

By Adi Leviim, Co-Founder at Infi Developments. Last reviewed September 16, 2026 against OpenAI's Help Center.

A custom GPT is more than a saved prompt: it is an assistant with its own instructions, knowledge files, capabilities, and, optionally, the ability to call external APIs. OpenAI describes GPTs as versions of ChatGPT configured for a specific purpose, and anyone signed in to ChatGPT can use the GPTs they have access to.

This guide covers how to build one in an eligible workspace, step by step, from writing instructions to connecting an API, and what changes now that OpenAI is retiring custom GPTs in favour of plugins. Along the way, you will see how AI Toolbox helps you keep track of the conversations you have with the GPTs you build and use.

What Are Custom GPTs and Why Do They Matter?

Custom GPTs are specialized ChatGPT assistants whose instructions, uploaded knowledge, and optional integrations apply to every conversation, so they behave consistently each time you use them.

Think of a custom GPT as ChatGPT with fixed instructions plus optional tools. Here is what makes one different from starting a new chat:

• Persistent instructions: You define the GPT's role, tone, and rules once, and they apply to every conversation.
• Knowledge files: Upload documents, spreadsheets, or other reference files for the GPT to draw on when answering.
• Capabilities: Turn on built-in tools such as web search, image generation, Canvas, or Code Interpreter & Data Analysis.
• Apps or Actions: Connect services users have connected, or define your own API (a CRM, database, or any REST API) so the GPT can fetch or send data. A GPT can use apps or actions, but not both at the same time.
• Shareability: In a managed workspace, share with specific people or groups, your workspace, anyone with the link, or the GPT Store, when workspace settings allow it.

GPTs do not use saved memory, custom instructions, or previous conversations; each conversation starts fresh. Creating new custom GPTs is limited to ChatGPT Business (formerly Team), Enterprise, and Edu workspaces, subject to workspace settings. Personal accounts on Free, Go, Plus, and Pro can keep using existing GPTs but cannot create or publish new ones. OpenAI also plans to retire custom GPTs in favour of plugins: for affected Enterprise workspaces, retirement is scheduled for December 11, 2026, and other plans may follow the same timeline.

How to Create Custom GPTs in 5 Steps

In a Business, Enterprise, or Edu workspace that still allows GPT creation, building a custom GPT takes five steps. For affected Enterprise workspaces, creation of new GPTs is planned to end September 25, 2026, so these steps apply only until your workspace's cutoff, and to editing GPTs you already have.

1. Open the GPT builder: go to Explore GPTs (chatgpt.com/gpts) and select Create.
2. Write the instructions: define the role, process, output format, and constraints.
3. Upload knowledge files: add up to 20 reference files, each up to 512 MB.
4. Add capabilities and actions: enable the tools you need and, optionally, connect an API with an OpenAPI schema.
5. Test, save, and share: run real prompts in Preview, select Create, then choose who can access the GPT.

Step 1: Access the GPT Builder

Open Explore GPTs in the ChatGPT sidebar or visit chatgpt.com/gpts, then select Create to open the GPT builder, which pairs the editor with a Preview pane for testing.

GPT creation works on the web; the ChatGPT mobile apps support using GPTs but not creating them. If Create is missing, you are most likely on a personal account, or your workspace role does not include GPT creation. The builder offers two ways to work:

• Create tab (conversational builder): You describe what you want, and ChatGPT helps draft the GPT for you. Good for a first draft.
• Configure tab (configuration view): You edit each field directly: name, description, instructions, conversation starters, knowledge, recommended model, capabilities, and actions. Better for precise control.

You can switch between both. A practical approach is to draft in the Create tab, then switch to Configure to tighten each field.

Step 2: Define Instructions (The Most Important Step)

Your instructions define how the custom GPT behaves: what it should do, how it should respond, and what it should avoid, applied to every conversation.

OpenAI's guidance is to use explicit step structure for multi-step workflows ("When X happens, do Y"), prefer concrete "Do X" instructions over long lists of prohibitions, include brief examples of acceptable and unacceptable output when the GPT must apply definitions, and use headings and lists so priorities stand out. Here is a framework that follows that structure:

"You are [Name], a [Role] that helps users with [Task]. ## Personality - Tone: [e.g., professional, friendly, concise] - Always [e.g., provide examples, cite sources] - Never [e.g., give medical advice, generate code without explanation] ## Process 1. First, ask the user for [required context]. 2. Then, [perform task]. 3. Always end with [action item, summary, or follow-up question]. ## Output Format - Use bullet points for lists - Keep responses under [X] words unless the user asks for more - Include a 'Next Steps' section at the end ## Constraints - If you don't know something, say so instead of guessing. - Do not access or reference any data not provided in the knowledge files. - Stay within the scope of [Topic]. Redirect off-topic questions politely."

Specific, structured instructions produce more consistent results than a one-line brief such as "Be a helpful assistant." Keep a copy of the final instructions outside the builder: if you migrate the GPT, OpenAI's planned workflow turns the instructions into a skill inside a plugin, and you will want the original to test against.

Step 3: Upload Knowledge Files

Knowledge files give your custom GPT reference material, such as documentation, guides, handbooks, or internal content, to draw on during a conversation.

You can attach up to 20 files to a GPT, and each file can be up to 512 MB. OpenAI says GPTs support most common document, spreadsheet, image, text, and code file types; accepted types can vary by model, and some are only available when Code Interpreter & Data Analysis is enabled for the GPT.

Best practices for knowledge files, based on OpenAI's builder and troubleshooting articles:

• Use files for reference, not rules: Put behavior, tone, and workflow guidance in the instructions, and keep files for source material.
• Prefer clean, text-forward files: Complex layouts can make content harder for the GPT to use, and images inside files may not be reliably interpreted, so add text descriptions of important visuals.
• Say how to cite: If the GPT should quote or cite your files, say so in the instructions and specify the format.
• Test retrieval: After uploading, ask questions in Preview that require the files. If the GPT cannot answer, confirm the information is in the files and simplify their layout.
• Keep your originals: OpenAI asks you to review migrated reference files before relying on a replacement plugin, so keep the source files you uploaded.

Building or maintaining several custom GPTs and losing track of conversations?

AI Toolbox adds folders, search, export, and a prompt library to ChatGPT, used by 40,000+ people and rated 4.7/5 from 800+ ratings on the Chrome Web Store. Install free.

Step 4: Add Actions (Connect APIs)

Actions connect your custom GPT to external APIs you define, so it can retrieve data or take actions in external systems.

Actions are optional. Before building one, note that custom actions do not transfer when a GPT is migrated to a plugin: they have to be rebuilt with a supported connector or a custom MCP server. Common uses include:

• Fetching live data: Weather, stock prices, database records, CRM contacts.
• Creating records: Adding tasks to project management tools, logging support tickets.
• Sending notifications: Triggering Slack messages, emails, or webhooks.

In the builder, go to Actions and select Create new action. Each action needs an authentication method (None, API key as Basic, Bearer, or a custom header, or OAuth) and an OpenAPI schema in JSON or YAML that describes your API's endpoints. You can paste the schema, import it from a URL, or start from a built-in example, and OpenAI offers ActionsGPT for help writing or debugging actions.

{ "openapi": "3.1.0", "info": { "title": "Weather API", "version": "1.0.0" }, "servers": [ { "url": "https://api.example.com" } ], "paths": { "/weather/{city}": { "get": { "operationId": "getWeather", "summary": "Get current weather for a city", "parameters": [ { "name": "city", "in": "path", "required": true, "schema": { "type": "string" } } ] } } } }

If the schema is valid, the editor shows the detected actions; test each one in Preview. Users may be asked to approve actions before they run. A few rules to know: a GPT with actions cannot also use apps; actions are not available for Pro mode, so the model selector shows only non-Pro models that support actions; public GPTs with actions need a valid Privacy Policy URL; and in Enterprise and Edu workspaces, admins can restrict which domains actions may call.

Step 5: Test, Iterate, and Publish

Testing in the Preview pane exposes weaknesses in your instructions, such as edge cases and ambiguous queries, that you should fix before sharing.

Before saving and sharing, test your GPT with these scenarios:

1. Happy path: Ask a standard question your GPT is designed for. Does it respond correctly?
2. Edge cases: Ask something partially related. Does it stay in scope or go off-topic?
3. Out of scope: Ask something completely unrelated. Does it redirect politely?
4. Knowledge retrieval: Ask a question that requires information from your uploaded files. Does it find the right answer?
5. Action calls: If you have actions, test each API endpoint with real and invalid inputs.

Changes save to a draft automatically. When you are satisfied, select Create (or Update for an existing GPT), then select Share, choose a sharing level and a permission level, and select Save. Sharing options depend on your workspace settings:

• Invite-only: Only you, or specific users and groups you add in a managed workspace (up to 100 recipients).
• Anyone at your workspace with the link or Your workspace: Access for workspace members.
• Anyone with the link: Shareable but not listed, if public link sharing is available to you.
• GPT Store: Public listing when account eligibility, workspace settings, and permissions allow it. You may need a category, a complete builder profile, and to pass automated policy checks.

If you plan to migrate the GPT, publishing is what matters: drafts cannot be migrated, and publishing does not require sharing publicly. Keep in mind that at retirement, custom GPTs are scheduled to leave the GPT directory.

Custom GPT Retirement: What Changes and When

OpenAI is planning to retire custom GPTs across ChatGPT plans and recommends moving GPT workflows to plugins, which package reusable instructions (skills) with connected apps. Existing GPTs remain usable until retirement. OpenAI's retirement FAQ lists these planned milestones for affected Enterprise workspaces:

• September 11, 2026: Admin notice to help organizations prepare.
• September 17, 2026: Target for the migration experience and user banner (a target, not a guarantee for every workspace).
• September 25, 2026 (planned): Creation of new custom GPTs ends.
• December 11, 2026: Scheduled retirement. Custom GPTs stop running and leave the GPT directory.

OpenAI says the transition affects all ChatGPT plans and that other plans may follow the same timeline, with in-product announcements; the dates are subject to change. Public GPTs created in affected Enterprise workspaces are included even when you use them from another plan. In the planned migration, a GPT's instructions become a skill and its connected apps are added as apps, while custom actions and the selected model do not carry over, and conversation starters and previous chats may not copy. After migration, the original GPT stays usable until retirement but becomes read-only.

How to Migrate a Custom GPT to a Plugin

Migration is a built-in option on a published GPT: select its migration button when your account or workspace offers it, then review, test, and share the plugin it creates. For Enterprise workspaces the migration experience is targeted for September 17, 2026; personal accounts should follow the instructions in their in-product notice.

1. Inventory your GPTs: List the GPTs you rely on, who created each one, and who needs access. Record the instructions, reference material, integrations, and a few familiar prompts for testing.
2. Finish edits and publish: The original becomes read-only after migration, and drafts cannot be migrated directly, so make essential edits and publish first. Publishing does not require public sharing.
3. Confirm you can migrate: In the planned Enterprise workflow, only the GPT's creator or a workspace admin can migrate it, plugins must be enabled, and the GPT must be published. Using someone else's GPT does not give you permission to migrate it.
4. Select the migration button: Open your published GPT and select its migration button, or ask a workspace admin for help.
5. Review the plugin: Check that your instructions arrived as a skill and your connected apps as apps, and review the migrated reference files, templates, examples, and tools.
6. Rebuild custom actions: Rebuild required integrations with a supported connector or custom MCP server, review them with your technical and security teams, and test before switching.
7. Test against the original: Run familiar prompts and at least one harder case, and check that the plugin selects the right skill, uses the expected reference material, produces complete answers or files in the required format, and has the tools it needs.
8. Share it: The replacement plugin starts private. Review its sharing settings, share it, and ask an intended user to install and test it. Make future changes in the plugin.

Organizing Custom GPT Conversations

Building, testing, and using several custom GPTs creates a lot of conversations, and folders and search keep them findable, which matters more now that previous chats may not copy into a migrated plugin.

Here is the problem: if you maintain 5 GPTs and have 10 conversations with each, that is 50 chats mixed in with your regular ChatGPT usage, and they are the record of how each GPT was tested and used.

AI Toolbox solves this with folders. Create a folder for each custom GPT, such as "Product FAQ Bot," "Code Review Assistant," or "Content Writer," then open the folder and use the in-folder search box to add the GPT itself and its related conversations by name. The free plan includes 2 root folders with unlimited subfolders, and Premium ($9.99/month or $99 lifetime) raises that to up to 500 folders.

Search is equally useful. Full-text search finds the exact session where you tested a feature or fixed an issue; the free plan shows the first 5 matches, and Premium shows all matches and adds Advanced Search filters. Premium can also export chats as Markdown, JSON, or PDF, and in bulk, which is a practical way to keep a record of GPT conversations before retirement.

Use cases that rely only on instructions and knowledge files are the simplest to migrate. Use cases that need actions will require rebuilding the integration as a connector or custom MCP server when you move to a plugin.

Frequently Asked Questions

Do I need to know how to code to create a custom GPT?

No. The GPT builder has a conversational Create view where you describe what you want in plain English. The only part that requires technical knowledge is adding Actions (API connections), which use OpenAPI schemas. GPTs that only need instructions and knowledge files require no coding.

Can I create custom GPTs on the free ChatGPT plan?

No. Personal accounts on Free, Go, Plus, and Pro cannot create or publish new custom GPTs. Creating them is limited to Business (a Standard seat costs $25 per user per month billed monthly, or $20 billed annually), Enterprise, and Edu workspaces, subject to workspace settings, and creation is planned to end September 25, 2026 for affected Enterprise workspaces. Free users can still use existing GPTs they have access to while signed in.

How many knowledge files can I upload?

You can attach up to 20 files to a GPT, and each file can be up to 512 MB. GPTs support most common document, spreadsheet, image, text, and code file types, and some types require Code Interpreter and Data Analysis to be enabled. Clean, text-forward files tend to work best.

Can my custom GPT access the internet?

Yes, if you enable the Web search capability in the Configure tab. A GPT can also reach outside services through apps or through Actions (API calls), but not both at once. Without these, the GPT relies on the model and its uploaded knowledge files.

Are custom GPTs being discontinued?

Yes. OpenAI announced on September 11, 2026 that it is planning to retire custom GPTs across ChatGPT plans and recommends moving workflows to plugins, with a migration path planned. Existing GPTs remain usable until retirement. OpenAI says other plans may follow the Enterprise timeline and that the dates are subject to change.

When do custom GPTs stop working?

For affected Enterprise workspaces, retirement is scheduled for December 11, 2026, when custom GPTs stop running and leave the GPT directory. The migration experience is targeted for September 17, 2026, and creation of new custom GPTs is planned to end September 25, 2026. Other plans may follow the same timeline, so watch the in-product notice for your account or workspace.

How do I migrate my GPT to a plugin?

When migration is available to you, open your published GPT and select its migration button, or ask a workspace admin; personal accounts should follow their in-product notice. In the planned Enterprise workflow, only published GPTs can be migrated, by their creator or a workspace admin, with plugins enabled. The instructions become a skill and connected apps become apps. Review and test the plugin, then share it, because it starts private and the original GPT becomes read-only.

What happens to GPT actions?

Custom actions do not transfer through the migration workflow. The person maintaining the workflow needs to rebuild the required integrations with a supported connector or a custom MCP server, review them with technical and security teams, and test before switching. A rebuilt integration may not provide every capability of the original action.

How do I organize conversations with different custom GPTs?

Install AI Toolbox and create a folder for each custom GPT, then add the GPT and its chats by searching inside the folder. The free plan includes 2 root folders (unlimited subfolders) and 2 pinned chats. The Premium plan ($9.99/month or $99 lifetime) raises the caps to up to 500 folders and 500 pinned chats and shows all search matches.

Conclusion

Custom GPTs remain a useful way to package instructions, knowledge, and tools inside ChatGPT, but they are now a transitional feature. If your workspace still allows creation, the builder is quick to use, and the time you invest in clear instructions and clean knowledge files carries over, because instructions become a skill when you migrate. Plan every new GPT with its move to a plugin in mind, and rebuild any actions early.

As you build, test, and migrate custom GPTs, your ChatGPT history fills up with testing conversations and feedback sessions. Stay organized with AI Toolbox: create folders for each GPT project, pin important conversations, search across your chats, and export the ones you need to keep. Download it free from the Chrome Web Store.

Last updated: September 16, 2026

Key Terms

Plugin

OpenAI's recommended replacement for custom GPTs: a package of reusable instructions (skills), connected apps, and other capabilities for ChatGPT and Codex.

AI Toolbox

Chrome extension with 40,000+ users that adds folders, search, export, and prompt management to ChatGPT. Runs on Chromium browsers such as Chrome, Edge, Brave, Opera, and Arc.

Free Plan

2 root folders (unlimited subfolders), 2 pinned chats, 2 saved prompts, the first 5 search matches, single-chat TXT export, and RTL support.

Premium

$9.99/month or $99 one-time lifetime: up to 500 folders, pinned chats, prompts, and prompt chains, all search matches with Advanced Search, and Markdown, JSON, PDF, and bulk export.

References

Sources, tool names, and authoritative documentation referenced in this article:

• OpenAI Help Center, Custom GPT retirement and migration FAQ
• OpenAI Help Center, ChatGPT release notes (September 11, 2026: Planned custom GPT retirement and migration to plugins)
• OpenAI Help Center, Plugins in ChatGPT and Codex
• OpenAI Help Center, Creating and editing GPTs
• OpenAI Help Center, GPTs in ChatGPT
• OpenAI Help Center, Sharing and publishing GPTs
• OpenAI Help Center, Configuring actions in GPTs
• OpenAI Help Center, Troubleshooting GPTs
• OpenAI Help Center, ChatGPT Business overview
• OpenAPI Specification
• AI Toolbox (formerly ChatGPT Toolbox), Chrome Web Store listing
• AI Toolbox, Official site

Retrieved September 16, 2026.