How to Create a Custom GPT in ChatGPT (2026): Steps, Limits and the Plugin Switch

Custom GPTs let you build your own version of ChatGPT with fixed instructions, files and tools. But in 2026 the story changed: OpenAI is retiring custom GPTs and replacing them with plugins.

This guide covers both sides. You get the full GPT builder steps, if your workspace still allows it, plus how to migrate and how to build the plugin that replaces a GPT.

Are custom GPTs going away?

Yes. OpenAI told users and admins on September 11, 2026 that it plans to retire custom GPTs and offer a migration path to plugins.

Note: OpenAI’s help center blocks automated readers, so these dates come from OpenAI’s own Enterprise docs plus Harvard IT and other reports. OpenAI says dates can change and other plans may differ, so check the banner in your account.

Your old chats with a GPT are reported to stay readable after retirement. But the GPT itself stops running.

Who can still create a custom GPT

This is the most common question in 2026, and the answer depends on your plan.

• Free, Go, Plus and Pro: creating and publishing new GPTs is no longer offered. Existing GPTs keep working until retirement.
• Business, Enterprise and Edu: creation may still be possible if your workspace settings and role allow it. Business needs at least two seats.
• Affected Enterprise workspaces: creation stopped on October 26, 2026. You can still edit GPTs you have not migrated.
• Everyone: creation only ever worked on the web, not in the mobile apps.

On a personal plan, ChatGPT Projects are the closest free alternative. They hold instructions and files for a topic.

How to create a custom GPT, step by step

If your account still shows the GPT builder, these are the steps.

1. Open the builder. On the web, go to Explore GPTs in the sidebar or chatgpt.com/gpts, then select Create.
2. Use the Create tab first. Describe your GPT in plain words, such as “a helper that turns meeting notes into action lists”. ChatGPT drafts a name, instructions and starters.
3. Switch to Configure. Edit each field directly: name, description, instructions and conversation starters.
4. Add knowledge. Upload reference files the GPT should use.
5. Pick capabilities. Turn on only what the job needs, such as web search or Canvas.
6. Set the recommended model. Choose it in Configure.
7. Add actions if needed. Connect an outside API, covered below.
8. Test in Preview. Try normal requests and one tricky one in the right-hand pane.
9. Save. Select Create for a new GPT or Update for an existing one, then choose who can see it.

Tip: Changes save to a draft automatically. Publish your final version even if you keep it private, because only published GPTs can be migrated to plugins.

Writing good instructions

Instructions are the heart of a GPT. Clear, short sections work better than one long paragraph.

• Role: who the GPT is and who it helps.
• Process: the steps it should follow every time.
• Output format: length, headings, tables or bullet points.
• Rules: what it must never do, and what to ask when information is missing.
• Knowledge use: when to check the uploaded files first.

Role: You help small shop owners write product descriptions.
Process: Ask for the product, price and audience. Then write 3 versions.
Format: Each version under 80 words, with a bold headline.
Rules: Never invent specs. If size or material is missing, ask.
Knowledge: Follow the tone in brand-voice.pdf.

This same structure carries over to plugins, where instructions become a skill. Writing it well now saves work later.

Knowledge files and capabilities

Knowledge limits

• Files per GPT: up to 20.
• Size per file: up to 512 MB.
• Types: most document, spreadsheet, text, image and code formats. Some types need Code Interpreter turned on.

Capabilities you can turn on

• Web search: lets the GPT look up current information.
• Canvas: opens a side editor for longer writing or code.
• Image generation: creates images in the chat.
• Code Interpreter & Data Analysis: runs Python for files, charts and math.

Leave off what you do not need. Fewer tools keep answers focused.

Adding actions

Actions let a GPT call an outside API, like a booking system or your own backend.

1. Open Actions. In Configure, select Create new action.
2. Pick authentication. Choose None, API key (Basic, Bearer or a custom header) or OAuth.
3. Add the schema. Paste an OpenAPI schema in JSON or YAML, import it from a URL, or start from an example.
4. Check detected actions. If the schema is valid, each endpoint shows up in the editor.
5. Add a privacy policy URL. Public GPTs with actions need one.
6. Test in Preview. Run each action once before you publish.

Warning: Custom actions do not transfer when you migrate to a plugin. You will need to rebuild them with a supported app or your own MCP server, so keep a copy of your schema.

A GPT can use apps or actions, but not both. Actions are also not available in Pro mode.

Sharing and the GPT Store

Select Share in the builder, choose a level, then save. The options depend on your workspace.

Store listings need a category, a complete builder profile and policy checks. But at retirement GPT pages, including Store listings, go away. Public discovery moves to the plugin directory.

How to migrate a GPT to a plugin

OpenAI’s migration guide gives this path for creators.

1. Back up first. Copy your instructions, knowledge files, action schema and a few sample prompts with good answers.
2. Publish the latest version. Drafts cannot migrate.
3. Open My GPTs. Find the GPT under Created by me and select Migrate to plugin.
4. Test the plugin. Start a new chat, type @, pick the plugin and run your saved prompts.
5. Rebuild actions. Check the plugin directory for an existing app before building your own.
6. Share it again. Sharing settings do not carry over, so share with people or groups manually.

What moves and what does not

• Moves: instructions (as a skill), knowledge files and connected apps.
• Does not move: custom actions, the selected model, conversation starters, past chats and sharing settings.
• Afterwards: the original GPT becomes read-only.

How to build a plugin instead

If you are starting fresh, build a plugin. A plugin can bundle skills, reference files and connected apps, and it works in Chat, Work and Codex.

1. Open a chat. Use Chat or ChatGPT Work on the web.
2. Call Plugin Creator. Type @, search for Plugin Creator and select it.
3. Describe the workflow. Cover purpose, inputs, output format and rules.
4. Attach reference files. Add templates or examples it should follow.
5. Refine. Answer its questions until the instructions match the job.
6. Finish and test. Install it, start a new chat, type @ and pick your plugin.
7. Edit later. Open Plugins, select it and choose Edit Plugin.

Developers can also package plugins by hand with an MCP server. See my guides on the best MCP servers and building an MCP server in Python.

Troubleshooting: Common Problems and Fixes

• No Create button: your plan or workspace no longer allows new GPTs. Build a plugin or use a Project.
• Migrate to plugin is missing: the GPT must be published, you must be its creator or an admin, and plugins must be enabled.
• Action schema will not load: check the OpenAPI JSON or YAML is valid and the server URL uses HTTPS.
• Cannot share publicly: your workspace may block it, or an action is missing a privacy policy URL.
• Plugin does not trigger: plugins do not run on every request. Type @ and select it.
• Plugin answers differently: the model setting does not carry over. Tighten the skill instructions and retest.
• Plugin Creator not found: it depends on your workspace. Ask your admin to enable plugins.

Frequently Asked Questions

Can I still create a custom GPT in 2026?

Only in some Business, Enterprise or Edu workspaces where admins allow it. Personal plans can no longer create new GPTs.

When are custom GPTs retired?

Reports based on OpenAI’s notices give December 11, 2026, with February 11, 2027 for FedRAMP workspaces and approved deferrals. OpenAI says dates may change.

What replaces custom GPTs?

Plugins. They bundle skills, reference files and connected apps, and you can build one by chatting with Plugin Creator.

How many files can a custom GPT hold?

Up to 20 knowledge files, each up to 512 MB.

Do my GPT actions move to the plugin?

No. Custom actions do not transfer. Rebuild them with a supported app or your own MCP server.

Will the GPT Store still exist?

GPT pages, including Store listings, become inaccessible at retirement. Public discovery moves to the plugin directory.

Can I create a GPT on my phone?

No. GPT creation has always been web only. The mobile apps let you use GPTs, not build them.

Sources

• OpenAI: Moving your custom GPT workflows to plugins
• OpenAI: Build plugins
• OpenAI: Plugins
• Harvard IT: OpenAI will retire Custom GPT functionality
• Mimic: Custom GPTs are retiring, dates and migration
• AI Toolbox: Custom GPT builder guide 2026