# Custom GPTs Retire December 11: How to Move Them to ChatGPT Plugins

URL: https://www.synapticlabs.ai/blogs/chatgpt-plugins-migrate-custom-gpts
Author: Professor Synapse
Published: 2026-11-04

Source: Synaptic Labs, https://www.synapticlabs.ai/blogs/chatgpt-plugins-migrate-custom-gpts

---

> "Moving house is how you learn what you own. Pack the things you use, label the boxes, and test the lights in the new place before you hand out keys." _Professor Synapse_

ChatGPT retires custom GPTs on December 11, 2026. GPTs that aren't migrated by then stop running, and ChatGPT plugins are where their work goes next. OpenAI's documentation gives the date, and as of November 3, 2026, ChatGPT shows GPT owners a banner that reads "Migrate your GPTs to plugins by December 11."

Before then, four steps keep your team's work: list your GPTs, save copies of their instructions and settings, use **Migrate to plugin** on each one you keep, then rebuild any custom actions, share the new plugin again and test it. The button carries over a GPT's instructions (as a skill), its knowledge files and its connected apps. Custom actions, sharing settings and the chosen model stay behind, which is why the last step matters.

This guide is for teams with a handful of GPTs that people depend on. We sort each GPT by the job it does, migrate the ones that belong in a plugin, rebuild custom actions with the narrowest access that works, and move shared work into a ChatGPT Project. One section covers moving to Claude instead. Along the way we follow a fictional nonprofit team as it moves its three GPTs, so you can see each decision play out.

## What is changing for custom GPTs, and when

OpenAI's [migration guide](https://learn.chatgpt.com/docs/migrate-custom-gpts) says it directly: "We're transitioning custom GPTs to plugins." It also says what retirement means: "At retirement, custom GPTs stop running and leave the GPT directory."

The date appears in two places. OpenAI's [higher education roles and permissions page](https://learn.chatgpt.com/docs/enterprise/higher-education-roles-and-permissions) states: "Custom GPTs are scheduled to retire on December 11, 2026; follow the owning workspace's notice." And inside ChatGPT, the My GPTs page and the GPT directory both carry a banner. When we checked a ChatGPT Pro account on November 3, 2026, it read: "GPTs that aren't migrated won't be available after December 11. Migrate yours so people can keep using them as plugins."

![ChatGPT's My GPTs page with the banner "Migrate your GPTs to plugins by December 11" and Migrate to plugin buttons on two published GPTs; names blacked out](https://cms.synapticlabs.ai/api/media/file/4c80f01a95e17db543cbaf5e93299e4dbb632126ff552b5776b383b1cde8c5ba.png?v=4c80f01a95e17db5)

My GPTs on November 3, 2026. Each published GPT has a Migrate to plugin button. The draft at the top offers only Delete GPT in its menu.

Before you start, four details from the screens and the guide matter.

- **You can't make new GPTs.** Selecting **Create a plugin** on the My GPTs page now shows: "You can no longer create GPTs. Plugins are the new home for custom tools."
- **Only published GPTs can be migrated.** The **Migrate to plugin** button sits on each published GPT's row; drafts don't get one. The person migrating must be the GPT's creator or a workspace owner or admin, and plugins must be enabled for them. Shared access alone does not give someone permission to migrate.
- **Admins can migrate in batches.** The GPT directory shows a **Migrate GPTs** control, and the guide says workspace admins can use the ChatGPT Admin plugin to migrate selected batches.
- **The original GPT freezes.** "Once you migrate a GPT, the original becomes read-only." It stays usable until retirement, so make your last edits before you migrate, not after.

OpenAI's migration guide is written for ChatGPT Enterprise workspaces and notes that "public sharing and timelines for other plans may differ." The banner and the dated admin page both point to December 11, but confirm the notice in your own workspace. Converting a GPT costs no credits; under credit-based pricing, editing or updating the plugin afterward does.

## What are ChatGPT plugins in 2026?

A ChatGPT plugin is, in OpenAI's words, "an installable bundle that can include skills and Model Context Protocol (MCP) servers." ChatGPT and Codex use the same public plugin catalog, so a plugin is not tied to one chat window.

A **skill** "packages instructions and supporting resources for a specific task or workflow," according to OpenAI's [skills and plugins page](https://learn.chatgpt.com/docs/skills-and-plugins), and ChatGPT can choose one when your request matches its purpose. When you migrate a GPT, its instructions become a skill. An **app** connects ChatGPT to an outside service; OpenAI's [plugins page](https://learn.chatgpt.com/docs/plugins) describes MCP servers that "connect ChatGPT and Codex to tools and information in systems such as GitHub, Slack, or Google Drive," and a plugin's page lists them under **Apps**.

![ChatGPT's Google Drive plugin page, by OpenAI, showing Manage and Try in chat buttons, Apps 1 (Google Drive) and Skills 5, starting with Google Docs](https://cms.synapticlabs.ai/api/media/file/f12e738288137ac15d7fcc57bf91ff389eb317015339fc97126a5d7f12fdad31.png?v=f12e738288137ac1)

OpenAI's Google Drive plugin bundles one app and five skills: Google Docs, Google Drive, Google Drive Comments, Google Sheets and Google Slides.

A ChatGPT Project is a different thing again: a place for related chats and files, covered below. And if you remember "ChatGPT Plugins" from 2023, that was an earlier, separate feature; articles from that time describe something else.

## GPTs vs ChatGPT plugins vs ChatGPT Projects: sort before you migrate

OpenAI's guide explains how to move a GPT into a plugin. It does not ask whether every GPT belongs in one, and its migration pages don't mention Projects. So we add a step before anyone clicks a button: for each GPT, ask what job it does.

The GPT is really...

Best next home

Why

A repeatable procedure, such as a house style or report format

A plugin with a skill

The official path; ChatGPT can choose the skill when a request matches

A place where ongoing work and shared files live, such as a board cycle

A ChatGPT Project

Projects "keep related chats, files, instructions, and sources together"

A tool that reads from or writes to another system

A rebuilt plugin connection with the narrowest access that works

Custom actions do not transfer

Unused, or duplicated by another GPT

Retire it

Note what it did first

To make this concrete, we will follow one team through the whole move. Riverbend Literacy Alliance is a nonprofit we made up for this guide: twelve staff running after-school reading programs, with a ChatGPT workspace they have used for two years. Over that time their development lead built three GPTs that the team now leans on every week. The grants team uses one to turn program data into funder reports in the house style. The development team uses one to check a donor's giving history before a call. And the executive director's assistant uses one to prepare for board meetings. Then the banner appeared: migrate by December 11.

Their first instinct was to click **Migrate to plugin** on all three. Sorted by the job each one does, they actually go to three different places:

GPT

What it really does

Next home

Grant Report Helper

Drafts funder reports in house style from uploaded outcome data

A plugin (skill plus knowledge files)

Donor Lookup

A custom action that queries the donor database with an API key

A rebuilt, read-only connection in its own plugin

Board Packet Assistant

Holds this year's board materials and minutes for ongoing prep

A ChatGPT Project

For each GPT on your own list, also record who owns it, who uses it and what it connects to. That list becomes your migration plan and later your test plan.

![Sorting chart: a repeatable procedure goes to a plugin with a skill, shared ongoing work to a ChatGPT Project, system access to a rebuilt connection, unused GPTs retire; Riverbend's three GPTs shown](https://cms.synapticlabs.ai/api/media/file/5e193decf14d376351bdf612860e838584a80c10a0b8277ef675ae472572978f.png?v=5e193decf14d3763)

Riverbend Literacy Alliance is the fictional nonprofit from this guide.

## What the migration button moves and what it leaves

Every row below comes from OpenAI's [migration guide](https://learn.chatgpt.com/docs/migrate-custom-gpts), checked on November 3, 2026.

GPT part

What Migrate to plugin does

What your team still does

Instructions

"turns your GPT's instructions into a skill"

Read the skill and test it on real tasks

Knowledge files

Brings them over, though "transfer and retrieval may differ"

Confirm each file is present and used

Connected apps

They "remain part of the plugin"

Each person still needs their own access; "Plugin installation does not add app or file permissions"

Custom actions

"Custom actions do not transfer"

Rebuild them as a separate plugin (see below)

Conversation starters

They "do not transfer one-to-one"

Save them as example prompts

Selected model

"The selected model does not carry over"

Retest the outputs you care about

Capabilities

Settings "do not guarantee the same tools or behavior"

Test each capability the workflow needs

Previous chats

They "may not copy"

Save any context that matters

Sharing

"Your new plugin starts private," and bulk migration "does not copy sharing settings"

Share it again, then have a teammate test it

In short, what lived inside the GPT's configuration mostly moves, and anything that depended on a connection, an audience or a model choice needs a person to set it up again.

![Two columns: Migrate to plugin moves instructions as a skill, knowledge files and connected apps; you rebuild custom actions, sharing, model, conversation starters, capabilities and previous chats](https://cms.synapticlabs.ai/api/media/file/f8723c7047f72c252acdb3ff64629b4eedb7a2c09301b5478f43348cb3fe0ff0.png?v=f8723c7047f72c25)

## Walkthrough: migrating the Grant Report Helper

The Grant Report Helper is a good first candidate: a procedure with files and no custom action. We would migrate it in this order.

1. **Make final edits and save copies.** The original becomes read-only, so finish changes first. Copy the instructions, conversation starters and list of knowledge files into a document, and note the **Recommended Model** and which **Capabilities** boxes are checked (Web Search, Canvas, Image Generation, Code Interpreter & Data Analysis).
2. **Save three real prompts and their answers.** Use prompts the grants team relies on and keep the GPT's answers for comparison.
3. **Migrate.** On the My GPTs page, select **Migrate to plugin** on the GPT's row.
4. **Read the new skill.** Check that the house style and report structure came through intact.
5. **Check the files.** Ask something only one knowledge file can answer, such as a figure from last year's outcome data. If the answer ignores the file, fix that before going further.
6. **Rerun the three prompts.** The model may differ, so compare for missing sections, changed tone or lost formatting rather than identical wording.
7. **Share and hand it over.** The plugin starts private. Share it with the grants team and ask one colleague to install it and run a real task with their own account.

That matches OpenAI's own advice: "Review migrated files and test required tools before sharing."

## Rebuilding a custom action without widening access

Donor Lookup is the hard one. It calls Riverbend's donor database through a custom action, and "Custom actions do not transfer." The connection has to be rebuilt, and the rebuild is the moment to decide how much access it really needs.

### OpenAI's route for replacing a custom action

OpenAI's MCP page has a section on [replacing a GPT custom action](https://learn.chatgpt.com/docs/extend/mcp#replacing-a-gpt-custom-action). In short:

1. Open the original GPT and ask it to write a prompt for rebuilding the workflow as MCP tools. The prompt should separate read operations from operations that change data, cover inputs, returned fields and answer format, and leave out credentials and hidden instructions.
2. Before migrating, check that prompt against the GPT editor's **Actions** section. Save the action schema, the authentication method and the non-secret connection settings.
3. Start a new task, select **@Plugin Creator**, paste the reviewed prompt, attach the schema or API documentation, and ask it to host the server on Sites (listed in ChatGPT's sidebar).
4. Connect your system, publish and install the plugin, then ask it to do the original task.

The result is a separate plugin. OpenAI notes that Sites "doesn't add the connection to your migrated plugin," and that you can use both plugins in one conversation. Sites plugins don't currently support editing skills or reference files, so keep instructions in the migrated plugin and the connection in the new one.

### Decide what the connection may do

Before the rebuild, list every operation the old action could perform: read, create, update, delete or archive, and send. Then grant only what the job needs. Donor Lookup only looked up gift history, so its replacement should only read. A GPT that looked up records should not come back able to send email or edit donor profiles. Two details help:

- **Read-only tools must be marked.** OpenAI's [custom MCP guide](https://developers.openai.com/api/docs/guides/custom-mcp-server) says ChatGPT respects the `readOnlyHint` annotation, and "Tools without this hint are treated as write actions," which require confirmation by default. Ask whoever builds the connection to mark the lookup tools read-only.
- **Plan the sign-in again.** The old action used an API key. The custom MCP guide lists OAuth and no authentication as the options for these connections, and for Sites-hosted plugins, OpenAI says Sites "handles sign-in between ChatGPT and your MCP server." Don't paste the old key into a chat; work out the new sign-in with whoever owns the donor database.

### Confirmation is not a restriction

Each installed plugin has a **Permissions** page in ChatGPT's settings (Settings, then **Plugins**, then the plugin, then **Permission**): "Choose when ChatGPT needs your approval to use tools." When we checked on November 3, 2026, it offered four choices:

- **Always ask**: "Ask before reading or making changes"
- **Allow read-only tools**: "Read without asking, but ask before making changes"
- **Allow low-risk tools**, marked **Default**: "Automatically approve low-risk tools, but sensitive tools may be denied"
- **Allow all tools**, marked **Elevated risk**: "Use tools without asking. This comes with elevated risk"

For a rebuilt lookup like Donor Lookup, **Allow read-only tools** fits: lookups run, and anything that would change a record still asks first.

![ChatGPT's plugin Permissions page with four choices: Always ask, Allow read-only tools, Allow low-risk tools (Default) and Allow all tools (Elevated risk)](https://cms.synapticlabs.ai/api/media/file/75fbe2a85db9e756c548baf10f937bef34f14eb10364c1b0136834cb30237b67.png?v=75fbe2a85db9e756)

The Permissions page for the Google Drive plugin, November 3, 2026.

These choices decide when ChatGPT asks you, not what the plugin is able to do. The same Google Drive plugin lists its tools in two groups, **Write 16** and **Read 30**, and the write group includes tools that create, copy and delete files. An approval prompt sits in front of those tools; it does not remove them.

![The Google Drive app panel in ChatGPT listing its Write tools, 16 in total, such as batch_update_document and bulk_update_file_comments, with a Read group of 30 below](https://cms.synapticlabs.ai/api/media/file/25767ec5eecca2605ad29784816d33c0bd0a21f8fa1f6c453d717e583fac8d9a.png?v=25767ec5eecca260)

Opening a plugin's app shows which of its tools read and which write.

The real limits live elsewhere. Workspace admins can, where supported, use the **Tools** (or **Actions**) setting to allow actions that read or change information, and OpenAI's [apps and connectors page](https://learn.chatgpt.com/docs/enterprise/apps-and-connectors) warns that "Asking for confirmation doesn't replace restrictions on what the app or connected account can access." Limit the account the connection uses, too.

## When a ChatGPT Project is the better home

The Board Packet Assistant was never a tool. It was a shared folder with instructions attached: this year's board materials, last meeting's minutes and a note on how the board likes its summaries. A plugin would package the instructions and miss the point, which is a place for ongoing board work.

That is what ChatGPT Projects are for. OpenAI's [Projects page](https://learn.chatgpt.com/docs/projects) describes them as a way to "keep related chats, files, instructions, and sources together," and notes that "Project instructions apply across its chats."

To move the Board Packet Assistant into a Project:

1. **Create the Project.** Select the add button beside **Projects** in the sidebar, name it (Riverbend would use "Riverbend Board Packet"), and check the memory choice before you select **Create project**.
2. **Add the board materials as sources.** On the **Sources** tab, **Add sources** offered **Upload**, **Space**, **Paste text**, **Google Drive** and **Slack** when we checked.
3. **Paste the instructions.** Open **Project settings** and paste the GPT's instructions into **Instructions**, which the screen describes as "Set context and customize how ChatGPT responds in this project."
4. **Read the Memory setting.** With the default memory setting, "This project can access memory from outside chats, and vice versa," so decide whether that suits confidential board work.
5. **Read Space access before you share.** The setting says the project "can access your file Space while it remains private. Sharing this project disables Space access." (Earlier on November 3 the same setting was labelled **Library access**; ChatGPT's labels are changing quickly, so look for the current name.)
6. **Share it.** Share the Project with the people who prepare the packet, and start each new cycle's chats inside it.

![ChatGPT's Create project dialog with the name Riverbend Board Packet, the note that projects keep chats, files and custom instructions in one place, a Default memory menu and a Create project button](https://cms.synapticlabs.ai/api/media/file/26e506d7f1302a098336338c6cf837acd3bd3c4eb93c0bb011e8233b43e483b9.png?v=26e506d7f1302a09)

Creating the fictional Riverbend Board Packet project. We closed the dialog without creating it.

![ChatGPT Project settings showing Project name, an empty Instructions box, Memory set to Default and Space access set to Enabled](https://cms.synapticlabs.ai/api/media/file/25be1832820900ecb0707ee42ef7d00113548c27dd8b07a7f0fb75c9b390962f.png?v=25be1832820900ec)

Project settings in ChatGPT, November 3, 2026: Instructions, Memory and Space access.

Projects and plugins also work together: OpenAI's Projects page suggests you "Install plugins to bring in context and actions from other services," so the Grant Report Helper plugin could run inside a grants Project.

## Moving custom GPTs to Claude instead

Some teams will use the deadline to move to Claude. We found no official Anthropic tool that imports a custom GPT or a ChatGPT Project (searched November 3, 2026). Claude's [memory import](https://support.claude.com/en/articles/12123587-import-and-export-your-memory-from-claude) is "experimental and still in active development," and it brings over memory text, not GPTs. So a move to Claude is a manual rebuild, and the same sort applies.

- **A procedure becomes a skill.** Claude has skills too. If the team also needs commands or connectors with it, package the skill in a [Claude plugin](https://claude.com/docs/plugins/overview).
- **A body of work becomes a Claude Project.** [Claude Projects](https://support.claude.com/en/articles/9517075-what-are-projects) are "self-contained workspaces with their own chat histories and knowledge bases," with their own instructions. On Team and Enterprise plans, members can be given **Can view** or **Can edit** access.
- **System access becomes a connector.** On Team and Enterprise plans, Owners can group a [connector's](https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities) tools into read-only and write/delete and set each to **Always allow**, **Needs approval** or **Blocked** for the whole organization. Each person still authenticates individually.

A Claude plugin also behaves differently depending on where it runs. Anthropic's [platform support page](https://claude.com/docs/plugins/platform-support) shows that regular Claude chat loads skills but ignores agents, hooks and local MCP servers, all of which Claude Code loads. Test in the app your team will use, and don't assume a skill written for one platform runs unchanged on the other; try it with the same three prompts.

For organization-wide rollout, Owners of Team and Enterprise plans can [manage plugins](https://support.claude.com/en/articles/13837433-manage-plugins-for-your-organization) by uploading them or syncing them from a private GitHub or GitLab repository, then setting each one to **Not available**, **Available to install**, **Installed by default** or **Required**. Cowork and Skills must both be enabled first.

## Test before you share

Whatever the destination, a migrated GPT is not done until someone other than its builder has used it. Run this check on each one:

1. Run the three saved prompts through the new plugin or Project and compare them with the old answers.
2. Ask a question that only a knowledge file can answer, and confirm the file was used.
3. Run each rebuilt read action. Then run each write action against a test record, and read the result back in the source system.
4. Have a second person install or open it and use it with their own account.
5. Record who owns it from now on.

![Five-step test card: run the three saved prompts, ask a question only a knowledge file can answer, run each rebuilt action, have a second person use it, record who owns it](https://cms.synapticlabs.ai/api/media/file/bcee505454d529d2b0fa497209ac89e197c60368b34d5f9be83c0b096b2f9456.png?v=bcee505454d529d2)

To keep track, copy this inventory into a document or spreadsheet, one block per GPT:

```
GPT name:
Owner:
Who uses it:
What job it does:
Next home (plugin / Project / rebuilt connection / retire):
Knowledge files:
Custom actions (list each operation: read, create, update, archive or delete, send):
Access the rebuild needs:
Three test prompts saved (yes/no):
Migrated on:
Tested by a second person (name, date):
New owner:
```

Sort first, then migrate the procedures, rebuild connections with no more access than they need, and move shared work into Projects. If you'd like a second pair of hands with your inventory, [book a call with us](https://www.synapticlabs.ai/book-a-call).

## Frequently Asked Questions

### Are ChatGPT plugins still available in 2026?

Yes. As of November 2026, ChatGPT has a plugin directory, and plugins are where custom GPT work is moving. A plugin is an installable bundle that can include skills and MCP servers, and ChatGPT and Codex use the same public plugin catalog.

### Did ChatGPT remove plugins?

No. What is being retired is custom GPTs, which OpenAI is transitioning to plugins, with December 11, 2026 given as the retirement date. The name "ChatGPT Plugins" was also used for an earlier, different feature in 2023, so older articles about ChatGPT plugins describe something else.

### What are ChatGPT plugins?

A ChatGPT plugin is an installable bundle of skills (packaged instructions for a task) and MCP servers (connections to outside services such as Google Drive). OpenAI's Google Drive plugin, for example, contains one app and five skills.

### How do I add plugins to ChatGPT?

Open **Plugins** in ChatGPT, browse or search the directory, and add the plugin you need. Some plugins ask you to sign in to a partner service at install or first use. Signing in does not approve the plugin's actions automatically; you review and approve its permissions as a separate step.

### What is the difference between GPTs and ChatGPT plugins?

A custom GPT is a configured version of ChatGPT with its own instructions, files and actions, and it is being retired. A plugin is an installable bundle of skills and connections that works across ChatGPT and Codex. When you migrate a GPT, its instructions become a skill inside a new plugin.

### Do custom GPT actions transfer to plugins?

No. OpenAI's guide says "Custom actions do not transfer." Its recommended route is to rebuild the connection as a new plugin with Plugin Creator, which becomes a separate plugin from the migrated one. Save the action's schema and authentication method before you migrate.

### Will my teammates still have access after I migrate a GPT?

Not automatically. "Your new plugin starts private," and bulk migration does not copy sharing settings. Share the plugin again, and remember that each person still needs their own access to any connected app.

### Can I move a custom GPT to Claude?

Yes, but by hand. We found no official tool that imports a custom GPT into Claude. Rebuild procedures as Claude skills, shared work as Claude Projects, and system access as connectors with per-tool permissions, then test with the same prompts you used in ChatGPT.