Custom GPTs Are Retiring: What It Means and What to Do

Published
November 5, 2026
Reading
16 min read
Share
In This Piece
  1. The Short Version
  2. What Exactly Is Retiring
  3. Every Date in the Retirement
  4. What a Plugin Is, and How It Differs From a GPT
  5. What Moves Across, and What Stays Behind
  6. If You Built GPTs: How to Migrate, Step by Step
  7. If You Use Someone Else's GPT
  8. If You Run a Business or Enterprise Workspace
  9. If You Built on the Assistants API
  10. When the Migration Misbehaves
  11. Migrate, Rebuild, or Move
  12. Keep Each Client's Record Outside Any Chat Tool
  13. Where GoMimic Fits
  14. Questions People Ask

OpenAI is retiring custom GPTs. On December 11, 2026, every custom GPT in ChatGPT, and the page it lives on, stops working. OpenAI's replacement is called a plugin, and it works differently enough that "migrate" undersells the change.

This guide covers all of it: what is going away, every date, what moves across and what doesn't, how to migrate step by step, what to do if you only use other people's GPTs, and where to go if you'd rather not rebuild inside ChatGPT at all. Everything here comes from OpenAI's own help pages, linked as we go, and was checked on November 5, 2026. OpenAI updates these pages often, so treat the notice in your own account as the final word.

The Short Version

  • Custom GPTs retire on December 11, 2026. On that date, OpenAI says, "custom GPTs and their GPT pages become inaccessible". That covers every plan.
  • Your GPTs still work until then. Existing GPTs stay usable up to that date, subject to your permissions.
  • New ones are mostly over. New GPT creation is off on Free, Go, Plus, and Pro, and ended on October 26 for affected Enterprise workspaces.
  • The replacement is a plugin. Your instructions become a "skill", your knowledge files become "reference files", and your connected apps come along.
  • Some things don't move. Custom actions, past conversations, sharing settings, the people who used your GPT, and its selected model all stay behind. OpenAI says your past conversations stay readable after retirement.
  • After you migrate, the original GPT is frozen. It becomes read-only, and you can't delete it.
  • FedRAMP workspaces and workspaces with an approved deferral get until February 11, 2027. Nobody else does.

What Exactly Is Retiring

The whole custom GPT feature, not just the GPT Store. OpenAI's release notes for September 11, 2026 announce it will "retire custom GPTs across ChatGPT plans". In practice that means:

  • Every custom GPT, whether you built it for yourself, for your team, or published it publicly.
  • Every GPT page. That includes the pages people reach from a share link or from the GPT Store. Old links will simply stop working.
  • The GPT builder. It's gone from personal accounts, and new custom GPT creation ended on October 26, 2026 for affected Enterprise workspaces, FedRAMP workspaces excepted.

Two related changes often get mixed up with this one:

  • The Assistants API already shut down, on August 26, 2026. That was a developer product, separate from ChatGPT's GPTs. More on it below.
  • Google is making a similar move. Gemini's Gems, its version of custom GPTs, turn into "skills" for personal accounts on November 17, 2026, and are removed on December 17.

Every Date in the Retirement

Timeline of the custom GPT retirement. Already happened: the retirement is announced on September 11, Enterprise migration has a September 22 target, and Enterprise GPT creation ends October 26. Still to come: custom GPTs retire December 11, 2026, and FedRAMP workspaces and approved deferrals end February 11, 2027.
December 11 is the date that applies to almost everyone.
  • No published date: personal accounts lost the builder. OpenAI's GPTs help page says new GPT creation and publishing "are not available on personal ChatGPT accounts".
  • September 11, 2026: the retirement is announced. Enterprise admins got a notice the same day.
  • September 22, 2026: migration opens for Enterprise workspaces. That was OpenAI's target, "not a guarantee that the option is available in every workspace". Workspaces that chose delayed banners had an October 1 target instead.
  • October 26, 2026: affected Enterprise workspaces stopped creating GPTs. FedRAMP workspaces are the exception. You can still edit GPTs you haven't migrated.
  • December 11, 2026: custom GPTs retire. This is the standard date for every plan.
  • February 11, 2027: the later date. It only applies to FedRAMP workspaces and to workspaces with a qualifying, approved deferral.

OpenAI says migration "may become available at different times". There is no published date for when the migration option reaches a Plus or Pro account. Watch for the notice in your account.

What a Plugin Is, and How It Differs From a GPT

A plugin is a bundle. It can hold:

  • Skills. Reusable instructions, which is where your GPT's instructions end up.
  • Reference files. Your knowledge files.
  • Connected apps. Links to outside services like Slack or Google Drive.

You build and edit plugins by chatting with a helper called Plugin Creator (you install it from the plugin directory, then type @plugin-creator).

The bigger difference is how you use one. A custom GPT was a place you went: you opened it, and every message in that chat went through its instructions. A plugin is a tool that joins an ordinary chat. You call it with an @ mention, or ChatGPT picks it up on its own when your request seems to match. OpenAI is plain that "the plugin will not necessarily run on every request."

A comparison of custom GPTs and plugins on where they run, how they start, how they are shared, and how they are listed.A comparison of custom GPTs and plugins. A custom GPT has its own chat, starts when you open it, applies to every message, is shared by link or the GPT Store, and has a public listing in the GPT Store. A plugin works inside any chat, starts with an @ mention or automatically, may not apply to every message, starts private, and needs its own submission to be listed publicly.
A GPT was somewhere you went. A plugin is something you call.

That's why some people in OpenAI's own forum say it isn't an equivalent swap. If your GPT's value was that everything in its chat followed one set of rules, check carefully that the plugin does the same for the work you actually do.

What Moves Across, and What Stays Behind

Two lists. Moves to the plugin: instructions become a skill, knowledge files become reference files, connected apps become plugin apps. Stays behind: custom actions, past conversations, sharing settings and users, the model it was set to use, and drafts and unpublished edits.
Plan for everything in the right-hand list before you migrate.

What moves, according to OpenAI's migration FAQ:

  • Instructions become a skill inside the new plugin.
  • Knowledge files are copied into the plugin's reference files.
  • Connected apps are added to the plugin as apps.

What doesn't:

  • Custom actions. If your GPT called another service through an action, that stops working in the plugin until you rebuild it. OpenAI says a rebuild "may require a custom MCP server and technical setup", and that a rebuilt version shouldn't be assumed to do everything the original did.
  • Your conversations. Past chats with the GPT don't move to the plugin. You can keep using the original GPT until it retires, and OpenAI now says your conversations stay accessible after that, with nothing for you to do.
  • Sharing and users. The plugin starts private. Nobody who used your GPT gets access to the plugin automatically, though in a managed workspace Share Plugin can fill in the original audience for you to review. A public plugin needs its own submission.
  • The "selected model". Whichever one you picked for the GPT doesn't carry over.
  • Drafts. Only the latest published version migrates. Unpublished edits are lost. Publishing doesn't mean making it public, though: you can publish it privately.

Two things OpenAI hasn't addressed yet: conversation starters and capability toggles like web search or image creation. Check them by hand after you migrate.

Can you export a GPT?

Not directly. OpenAI doesn't offer a way to download a GPT's setup as a file. The in-product migration is the only official route. That's why the first step below is making your own copy.

You don't need an export to keep your GPT conversations, by OpenAI's account, but you can still make one, on Free, Go, Plus, Pro, and eligible Edu accounts: go to Settings, then Data controls, then Export data. The export can take up to 7 days, and the download link expires 24 hours after it arrives. Self-service export isn't available on Business, Enterprise, or Healthcare workspaces; ask your workspace owner instead.

If You Built GPTs: How to Migrate, Step by Step

Six migration steps: back it up, finish and publish edits, migrate to plugin, test with familiar requests and one hard case, rebuild any custom actions, then share and move people over.
Step four is the one people skip, and the one that tells you whether it worked.
  1. Make your own copy first. This is our advice, not OpenAI's, and it doesn't take long. Copy the instructions into a document. Download every knowledge file. Write down any custom actions and what they connect to. Save five to ten requests you use often, with the answers you were happy with. Once you migrate, the original is read-only; once it retires, it's gone.
  2. Finish and publish any edits. Migration takes the latest published version only, and you can't edit the original afterwards. You can publish privately.
  3. Migrate. When the option reaches your account, go to My GPTs, choose Migrate to plugin on the GPT, review the details, and follow the steps. Only the person who created the GPT can do this (or, in an Enterprise workspace, an admin).
  4. Test it against your saved requests. Run the ones you saved in step one, plus at least one harder case. OpenAI suggests checking that the plugin picks the right skill, uses the right reference files, answers in the format you need, and has the tools it needs. A migrated plugin "may respond differently", so compare the answers side by side.
  5. Rebuild custom actions. Check first whether an existing app in the plugin directory already does the job. If not, a rebuild usually means a developer and a custom MCP server.
  6. Share it, then move people over. The plugin starts private. Share it, ask one person to install and test it, and only then point everyone else at it. An installed plugin doesn't run on every request, so tell people how to call it.

Leave more time than you think for any GPT with actions. Everything else is mostly copy and check. Actions are real rebuild work.

If You Use Someone Else's GPT

You can't migrate a GPT you didn't create, and you don't need to. What you do need to know:

  • Check the notice on the GPT. Creators have been asked to point users to a replacement.
  • Your access doesn't carry over. Being able to use a GPT today "does not guarantee access to its replacement plugin". You may need the creator to share it with you.
  • Where it was made matters, not your plan. Public GPTs made in Enterprise workspaces follow the Enterprise timeline, even if you use them from a personal account.
  • Save what you need from it. If you rely on a public GPT for something specific, write down what you ask it and what good answers look like, in case no replacement appears.

If You Run a Business or Enterprise Workspace

Admins have the most to organize, and a few details only apply to you:

  • List what your team relies on. Which GPTs, who built them, who maintains them, and who needs access to each replacement.
  • Check plugins are switched on. Migration won't appear if plugins are disabled, if the GPT isn't published, or if the person trying isn't its creator or an admin.
  • Migrate in bulk if you have many. OpenAI says bulk migration is available through its ChatGPT Admin plugin. It runs in the background, and starting it doesn't mean every GPT moved, so review the results.
  • Sharing and publishing are separate permissions. Sharing a plugin with people needs "Share plugins". Putting it in the workspace directory needs "Publish plugins to workspace".
  • Watch the cost. On Enterprise, using a migrated GPT without apps in Instant Chat has no extra cost. Creating and editing plugins happens in Work or Codex, and actions in Work use credits.
  • Bring security in early for actions. Rebuilt integrations need the same review as any new connection to your systems.
  • Standard admin emails use the standard dates. If your workspace has a deferral or is on FedRAMP, confirm your real timeline with your OpenAI account team.

If You Built on the Assistants API

This one is already over. The Assistants API shut down on August 26, 2026, a year after OpenAI announced it. The replacement is the Responses API with the Conversations API, and OpenAI's migration guide maps the old pieces to the new ones. If a tool you pay for was built on it and still works, it has already moved. If it stopped working in late August, that's why.

When the Migration Misbehaves

  • "I don't see a migration option." It may not have reached your account yet, plugins may be switched off, or you may be signed in to a different account from the one that built the GPT.
  • "The plugin doesn't kick in." Plugins aren't guaranteed to run on every request. Call it with an @ mention when it matters.
  • "The answers changed." The selected model didn't carry over, and the instructions now sit inside a skill rather than running the whole chat. Test before you rely on it.
  • "My integration broke." Custom actions don't migrate. See step five above.
  • "My users can't find it." They lost access at migration and need it shared again.
  • "I can't delete the old GPT." After migration, creators can't. It disappears at retirement.
  • "What about GPT Store earnings?" OpenAI hasn't said anything about builder payouts or usage stats. If you depended on either, ask OpenAI directly.

Migrate, Rebuild, or Move

Three options. Migrate: for GPTs that are instructions and files, using the built-in migration. Rebuild: for GPTs with custom actions, which need an app or developer work. Move: for work that was never really a chat job, such as repeated tasks across many clients, which may belong in a purpose-built tool.
Most GPTs are the first kind. The expensive ones are the other two.

Migrate if your GPT is instructions plus a few files. The built-in migration covers it, and your main job is testing.

Rebuild if it depends on custom actions. Budget for developer time, and check the plugin directory first for an app that already connects to the service.

Move if the GPT was a workaround for a job that isn't really a chat job. That's usually true when the same task repeats many times, for different people or clients, and the result needs checking or sign-off by someone else. Rebuilding that as a plugin keeps the workaround going.

If you do decide to leave ChatGPT for this, these are the closest equivalents elsewhere:

  • Claude has Projects, workspaces with their own instructions, files, and chat history, and Skills, reusable instructions it loads when relevant. Projects are on every plan, with five on the free plan.
  • Gemini is moving from Gems to skills: Gems for adult personal accounts turn read-only on November 17, 2026, and go on December 17. Workspace accounts follow in March 2027, and education accounts in June 2027. Google says it will recreate your Gems as skills automatically, though some files may not carry over.
  • Microsoft 365 Copilot has what it calls "agents", which you build with instructions and sources like SharePoint items.
  • Perplexity has Projects: shareable workspaces with instructions, files, and connected tools.

Look at what these have in common. Microsoft shut down custom GPTs in Copilot Pro in July 2024 and deleted the ones people had made. OpenAI is retiring its version now. Google is reshaping its own. Every one of these is a feature inside someone else's product, and features change. Whatever you move to, keep the thing that matters in a form you own.

Keep Each Client's Record Outside Any Chat Tool

If you built one custom GPT per client, you know the setup: paste the client's brand guide in, add a few of their posts, and you had a "client voice" GPT the whole team could use. For a lot of teams it was the best option available, and it worked well enough.

The retirement shows the weak spot. That client's voice lived in a chat feature. When the feature changes, the voice has to be rebuilt, retested, and re-shared, and in Microsoft's case, it was deleted. As Ragan put it, getting a GPT to sound like your CEO "was not a quick afternoon project."

A checklist of what an agency should keep for each client outside any chat tool, beside three custom-assistant features that were retired or reshaped.A checklist of what an agency should keep for each client outside any chat tool: their real published posts, the words they use and avoid, every change they've asked for, approved examples for each format, and who signs off. Beside it, three custom-assistant features that have been retired or reshaped: Copilot GPTs in 2024, ChatGPT custom GPTs in 2026, and Gemini Gems moving to skills in 2026.
If it only exists inside one tool's feature, it's rented.

So whatever you do with the GPTs themselves, keep a record for each client that doesn't depend on any one tool:

  • Their real published posts. Five to ten is plenty. These show how they write far better than a style guide describing it. We explain why in how AI content repurposing actually works.
  • Words they use, and words they'd never use. Including the ones they've corrected you on.
  • Every change they've asked for. "Don't call us a firm." "Never mention pricing." These are gold, and they usually live in one person's chat history.
  • Approved examples for each format. A LinkedIn post they signed off. A newsletter section they liked.
  • Who signs off. And how quickly they usually reply.

That record is what any new tool needs first. It's also hard to keep in a chat window for a long client list, which is the subject of why ChatGPT breaks down when you're managing 20 clients.

the part about us

Where GoMimic Fits

Create client content. Get it approved. Deliver it.

If your custom GPTs were one per client, used to turn each client's published posts into LinkedIn posts, newsletters, threads, and the like, that job doesn't have to be rebuilt as a plugin. That's the job GoMimic is built for, and it goes past the writing.

  • You can bring the GPT with you. Paste a custom GPT's instructions and upload its files. You get back suggestions to add or skip: Business Facts, rules, and Sounds Like Us pieces. The same goes for a Claude Project, a Claude Skill, or a Gemini Gem.
  • The client is the unit, not the chat. You add a client once. Every post you run for them starts from their record, whoever on the team is doing it.
  • The voice is what you set plus what it learns. Their rules, brand guidelines, and Business Facts sit on the client, and GoMimic learns each client's habits from their own writing and calls. GoMimic learns from the client's own writing, so one unusual post can't bend that client's voice. Nobody pastes a brand guide into a chat again.
  • One post in, the formats you pick out. Start from a published post or five other starting points, and pick the formats that client needs, from LinkedIn posts to newsletters, X threads, Google Business Profile posts, and more. The choice is remembered for next time.
  • The client's answer is built in. The client opens a link with no login, types their name, and answers post by post, and every decision is recorded. We covered the process in how to get client sign-off without chasing.
  • Delivery comes next. The approved version can be copied, saved to Google Drive or Notion, sent to Slack, sent as a draft to Buffer, Kit, or Mailchimp, or sent by webhook to Zapier and on to the tool you use. GoMimic never posts.

It won't replace every GPT you have. A GPT that answers questions about your own contracts, or checks a spreadsheet, belongs in a plugin or one of the tools above. It creates from what you give it, and a blog post or bylined article only from your client's own take. It won't pick a topic or plan their content strategy for you. But for the client content work, you can try it on one of your own client posts without creating an account, and see whether the result sounds like them.

Questions People Ask

Are custom GPTs retiring for everyone on December 11?

Nearly everyone. OpenAI says existing GPTs "remain usable until the retirement date", which is December 11, 2026. The exceptions are FedRAMP workspaces and workspaces with an approved deferral, which have until February 11, 2027.

Do I have to migrate?

No. If you don't, the GPT simply stops working at retirement. Migrate only the ones worth keeping.

Will my chat history with a GPT disappear?

OpenAI says no: your existing conversations with custom GPTs stay accessible after retirement, and you don't need to do anything. They don't move into the plugin, though. If you'd like your own copy, request an export from Settings, then Data controls, and check the file.

Is the GPT Store closing?

GPT pages, including store listings, become inaccessible when GPTs retire. Public discovery moves to the plugin directory, where each public plugin needs its own submission.

Can Plus and Pro users migrate their GPTs?

OpenAI says the change "affects all ChatGPT plans" and that migration access "may differ by account or workspace". It hasn't published a date for personal accounts. Follow the notice in your account.

What happens to the GPTs I use but didn't build?

They retire on the same date. Look for the creator's replacement, and expect to need access to it shared with you.

I have a GPT per client. What's the safest first move?

Copy each GPT's instructions and files out this week, and add the client's real published posts and change requests to the same folder. Then decide per client: migrate to a plugin, or move the content work to a tool built around clients rather than chats.

GoMimic

Written by the GoMimic team

GoMimic helps you create client content, get it approved, and deliver it, in each client’s voice.

Why It Works

Your Client's Yes Should Be This Easy to Find.

Start Free TrialFree for 14 daysNo credit card required

See Why It Works

Approved