Custom GPTs are retiring: how to keep your instructions working in any AI tool

Custom GPTs were one of the easiest ways to get consistent results from ChatGPT. You wrote the instructions once, attached a few reference files, and everyone who used the GPT started from the same setup.

OpenAI is retiring them on December 11, 2026. Most GPTs can move to ChatGPT plugins, but some parts don’t make the trip, and the new version may not respond the same way. If your team’s best instructions live inside a GPT, this is a good moment to decide where they should live next.

What’s changing with custom GPTs, and when

OpenAI plans to retire custom GPTs across all ChatGPT plans and help creators migrate them to plugins. According to OpenAI’s custom GPT retirement and migration FAQ, custom GPTs are scheduled to retire on December 11, 2026. On that date, custom GPTs and their GPT pages become inaccessible.

Date	What happens	Who it applies to
October 26, 2026	New GPT creation ended. Existing, unmigrated GPTs can still be edited.	Affected Enterprise workspaces, except FedRAMP
Now until retirement	Existing GPTs keep working. Creators can migrate a GPT to a plugin when the option appears.	All plans, as migration becomes available
December 11, 2026	Standard retirement. Custom GPTs and their pages become inaccessible.	All plans, unless your notice says otherwise
February 11, 2027	Later retirement date	FedRAMP workspaces and workspaces with a qualifying, approved deferral

OpenAI says migration may reach accounts at different times and asks everyone to follow the notice in their own account or workspace. The workspace where a GPT was created sets its timeline, even if you use it from another account. That matters for public GPTs, too. A public GPT built in an affected Enterprise workspace follows the Enterprise timeline, whatever plan you’re on.

Your chat history is safe. Existing conversations with custom GPTs remain accessible after retirement, and you don’t need to do anything to keep them.

Dates in this post come from OpenAI’s FAQ as of November 14, 2026. Check the notice in your account for your workspace’s timeline.

What moves to a plugin, and what doesn’t

Plugins are OpenAI’s replacement for GPTs. A plugin in ChatGPT can bring reusable instructions, reference files, and connected apps together. When a creator selects Migrate to plugin, here’s what happens to each part of the GPT.

Part of your GPT	After migration
Instructions	Become a skill inside the new plugin
Knowledge files	Copied into the plugin’s reference files
Connected apps	Added to the plugin as apps
Custom actions	Don’t transfer. You’ll need an available app or a custom MCP server to rebuild them.
Selected model	Doesn’t carry over. Enterprise workspaces use Enterprise defaults.
Sharing settings	Don’t carry over. The plugin starts private, and no one gets access automatically.
Drafts and unpublished edits	Don’t transfer. Migration uses the latest published version.
Your original GPT	Becomes read-only, stays usable until retirement, and can’t be deleted by its creator

The biggest change is how instructions get used. A GPT applied its instructions to every message in the chat. A skill in a plugin works differently. ChatGPT may use an installed skill automatically when its description matches your request, but OpenAI notes the plugin “will not necessarily run on every request.” You can also call it directly with an @ mention.

That’s why OpenAI recommends testing a migrated plugin with familiar prompts and at least one harder case before relying on it. If your GPT existed to make every answer follow the same rules, check that the plugin still does.

OpenAI also suggests a ChatGPT project for people who used a GPT to keep instructions and files together for ongoing work, like research or planning. A project keeps related chats, files, and instructions in one place, though it doesn’t carry over custom actions or sharing settings either.

A four-step checklist before December 11

1. List the GPTs you rely on. Note who created each one. Only a GPT’s creator, or a workspace admin in Enterprise, can migrate it. If you only use someone else’s GPT, watch for the creator’s guidance about a replacement.
2. Copy out what makes each GPT work. Save the instructions, reference files, and a few prompts you use often, along with the kind of answer you expect. You’ll want these for testing, and a GPT becomes read-only once it’s migrated.
3. Migrate and test. Finish and publish any edits first, since drafts don’t transfer. Then check that the plugin picks the right skill, follows your instructions, uses the right reference material, and returns answers in the format you need.
4. Rebuild custom actions and reshare. Allow the most time for GPTs that connect to other services. Once the plugin works, share it with the original audience and confirm they can install and use it before asking them to switch.

Step two is the one to start today. Even if migration goes smoothly, a plain copy of your best instructions protects them from the next change, too.

Why your instructions shouldn’t live in one AI tool

A custom GPT bundled two different things. One was the setup that belongs in ChatGPT, like reference files, connected apps, and actions. The other was the instructions themselves, the rules for tone, format, and what to check, which your team wrote and refined over months.

Those instructions are the valuable part, and they aren’t specific to ChatGPT. Many teams now use more than one assistant, such as ChatGPT for one task, Claude or Gemini for another, and Copilot inside Microsoft 365. Each tool has its own place to store instructions, and each of those features can change. Every move means copying, pasting, and hoping nobody keeps an old version.

Keeping the instructions as text you control solves that. One person who manages prompts this way described the payoff.

The best prompts are the ones you actually use. So it’s all about minimising friction to finding and using them. I store prompts and context in a text snippet tool, TextExpander. This means me and my team can use any prompt in any text box through a quick search or text replacement code.

Will Mulholland

How to save a custom GPT’s instructions as a Snippet

A TextExpander Snippet is saved text you insert by typing a short abbreviation. It works in the ChatGPT, Claude, and Gemini web apps through the Chrome extension, in desktop apps on Mac and Windows, and anywhere else you type. Five steps move a GPT’s instructions into one.

1. Copy the GPT’s instructions. Open the GPT’s configuration and copy the full instructions into a new Snippet. Do this before you migrate, while you can still edit the GPT.
2. Turn the parts that change into Fill-ins. If the instructions mention a client, a product, or a word limit, replace it with a Fill-in so you can type the detail each time.
3. Pick a clear abbreviation. A short prefix keeps AI Snippets together, like ;review for a reviewer or ;brief for a research brief.
4. Put it in a shared group. Save it to a Snippet Group your team can use, so everyone starts from the same version.
5. Use it to start a chat. In any assistant, type the abbreviation, fill in the details, and add your request.

Here’s what a support team’s “draft reply reviewer” GPT might look like as a Snippet. A word in {braces} is a Fill-in you type into, and options separated by a slash are a pop-up menu you pick from.

Draft reply reviewer (abbreviation: ;review)
Act as a support editor for {Company name}. Review the draft reply I paste at the end of this message.

Check it against these rules:
– Answer the customer’s question in the first two sentences.
– Use plain words and a {friendly / neutral / formal} tone.
– Keep it under {Word limit} words.
– Don’t promise refunds, dates, or features that aren’t in the reference text.

Return the revised reply first, then a short list of what you changed and why.

Reference text: {Paste policy or product details}
Draft reply: {Paste the draft}

Pop-up menus are a Pro feature. New TextExpander accounts get a 30-day preview of paid features, so you can build and test a Snippet like this before you pick a plan.

Snippets hold text, so they don’t replace a GPT’s knowledge files or actions. Long reference documents still belong in a plugin or project. For short facts the model needs every time, like a return window or a list of approved product names, paste them into the Snippet so they travel with the instructions.

People who rely on prompts all day describe the same relief.

TextExpander solves the frustrating problem of repetitive AI prompt input. Instead of constantly retyping or copying and pasting the same prompts throughout the day, TextExpander makes them instantly accessible through shortcuts.

Koji Noguchi, Attorney at Law, OFS Fitel, LLC

Snippet, plugin, or project: which fits each job

You don’t have to pick one. Most teams will use a mix, and the right home depends on what the GPT did.

What you need	Best home
The same instructions in several AI tools	A Snippet
Reference files the model should read	A ChatGPT plugin or project
Connected apps or actions inside ChatGPT	A ChatGPT plugin
Ongoing work with its chats in one place	A ChatGPT project
Approved wording the whole team shares, with an owner	A shared Snippet Group

The two approaches also work together. Keep the master version of your instructions in a Snippet, then paste it into a plugin’s skill or a project’s settings when you need ChatGPT’s files and apps. When the instructions change, you update the Snippet once, and everyone who types it gets the new version.

Teams that work across assistants already do this.

I have drafted these 6 Copilot prompts… Each has a TextExpander shortcut.

David Lindner Akdeniz, Renewals Specialist

Keep your team’s AI instructions in one shared place

A custom GPT gave a team one shared version of its instructions. A shared Snippet Group does the same job outside any single AI tool. Admins control who can edit each group, so the person who owns a prompt can update it once and everyone gets the change. Our guide to building an AI prompt library covers how to organize, name, and review a shared set.

If you use Claude or ChatGPT with the TextExpander MCP Server, the assistant can work with the library itself. TextExpander’s support team used it to review hundreds of Snippets, and the TextExpander MCP prompts post shows their workflow.

An incredible utility now that I’m using more AI browsers and prompts; TextExpander is invaluable.

Stephen Robles, YouTuber

For more ways to keep prompts close at hand, see how to save prompts in ChatGPT, these ChatGPT prompt templates for teams, and our roundup of AI prompt management tools.