Custom GPTs Are Retiring: How to Migrate to Plugins

Ellie at Willa, Editorial lead, Willa · Published September 24, 2026 · Updated September 24, 2026

If you ever built a custom GPT, even a small one that rewrites your emails or answers questions about your product docs, there is a deadline on your calendar now. OpenAI announced in September that custom GPTs are being retired and replaced by plugins, and the migration is not automatic. The good news: moving your GPT over takes less than an hour for most people, and the replacement is more capable than what you have today. This guide walks through the timeline, what carries over, what breaks, and the exact steps to migrate, written for people who built their GPT with instructions and uploaded files, not code.

When are custom GPTs going away?

According to OpenAI's retirement and migration FAQ, the key dates are:

• October 26, 2026: the ability to create new custom GPTs ends. Personal accounts have already lost the ability to create and publish new GPTs, so this date mostly formalizes what is already true.
• December 11, 2026: existing custom GPTs stop running. After this date, GPTs and their share pages become inaccessible.
• February 11, 2027: the extended deadline for Enterprise workspaces that request and receive a deferral. If you are on a personal Plus or Pro plan, this extension does not apply to you.

Enterprise workspaces got admin notices on September 11 and started seeing the migration experience the week of September 22. Free, Plus, and Pro accounts are expected to follow a similar path, with notices appearing in the product itself. The safe assumption for everyone: your GPT needs to be migrated before December 11, 2026, and waiting until December is a bad idea because you will want time to test.

What is replacing custom GPTs?

Plugins. In ChatGPT's current vocabulary, a plugin is a bundle of reusable instructions (called skills), reference files, and connected apps. If that sounds a lot like what a custom GPT already was, you are not wrong. The practical differences matter, though.

A custom GPT lived in its own chat window: you opened it from the sidebar or the GPT store and talked to it there. A plugin attaches to your regular ChatGPT experience and, per Mixed News, works across Chat, Work, and Codex rather than living in one place. Your brand voice instructions can now follow you into a document analysis session or a data question instead of being locked inside one dedicated chat.

If you have been thinking of your GPT as a coworker with one job, plugins push you toward something closer to a skill your AI carries everywhere. That is the same direction the whole industry is moving, and it is the model we described in How to Build an AI Teammate That Does Real Work.

What carries over when you migrate, and what does not?

This is the part to read twice, because the migration moves less than you might expect.

What transfers automatically:

• Your GPT's instructions become a skill inside the new plugin.
• Your uploaded knowledge files become reference files.
• Your connected apps carry over as plugin apps.

What does not transfer:

• Custom actions. If your GPT called an outside service through an action (a scheduling API, a product database, anything with an API schema you pasted in), that connection does not migrate. OpenAI says actions must be rebuilt using supported connectors or a custom MCP server.
• Your conversations. Existing chats stay in the original GPT. They do not move into the plugin.
• Sharing settings and your audience. A public GPT's link and its store page do not carry over, and people who used your GPT do not automatically get access to the replacement plugin. If coworkers or clients rely on your GPT, you have to share the new plugin with them deliberately.
• Model choice. If you pinned your GPT to a specific model, that preference is dropped.
• Drafts. Migration uses the latest published version only, so publish any unfinished edits before you migrate.

One more caveat straight from OpenAI's FAQ: a migrated plugin "may respond differently" than the original GPT. Same instructions, same files, slightly different behavior. That is why testing is a step, not an afterthought.

How do I migrate my custom GPT to a plugin?

Here is the full process, in five steps.

1. Take inventory before you touch anything. Open your GPT's configuration and copy the instructions into a document you control. Download or locate your original knowledge files. Screenshot your conversation starters. If you have actions, save the schema. This takes ten minutes and means nothing is lost no matter what the migration does.

2. Publish any pending edits. The migration only picks up the latest published version. If you have draft changes sitting in the editor, publish them first or they disappear from the copy.

3. Run the migration. Go to My GPTs and select "Migrate to plugin" on the GPT you want to keep. When it finishes, the original GPT becomes read-only but stays usable until the retirement date, so you are not cut off during the transition. If you do not see the option yet, your account tier likely has not received the rollout; check back rather than rebuilding by hand.

4. Test with real prompts. Give the new plugin five to ten prompts you actually used with the old GPT and compare the answers. Pay attention to whether it pulls from your reference files the way the GPT did and whether the tone still matches. If something drifted, edit the skill's instructions directly. Our guide on making AI write like you has a fast method for tightening voice instructions when the output sounds generic.

5. Rebuild actions and re-share. If your GPT had custom actions, recreate them with a supported connector, or with an MCP server if you are comfortable going deeper. Then share the plugin with every person who used the old GPT, because none of them were carried over. Send them the new access directly and tell them the old link dies December 11.

What if you only use someone else's GPT?

You do not migrate anything, and in fact OpenAI's FAQ warns against trying. The creator has to migrate, and having access to a public GPT does not guarantee you access to its replacement. So the one action worth taking: if a GPT someone else built is part of your daily routine, message the creator now and ask whether a plugin replacement is coming. If the GPT was abandoned, which describes a large share of the GPT store, copy out whatever you can see of its behavior and recreate the instructions yourself. Rebuilding a simple assistant from scratch is easier than it sounds; our walkthrough on building your first AI agent without code covers the same skills.

Is this worth doing, or should you just let your GPT die?

Be honest about which GPTs you still use. Most people who experimented with custom GPTs in 2024 and 2025 have two or three that mattered and a graveyard of tests. Migrate the ones with instructions you refined over months, the ones coworkers use, and the ones wired into a real workflow. Let the rest retire.

For the GPTs that made the cut, the migration is a real upgrade: the work you put into instructions and knowledge files survives, and the result travels across more of ChatGPT than the original ever did. It is also a natural moment to reconsider the tool itself. Instructions plus reference files is a portable pattern, and Claude Projects and similar setups accept the same material, so the document you created in step one is now your insurance policy in every direction.

Key takeaways

• Custom GPTs stop running on December 11, 2026, and new GPT creation ends October 26, 2026. Enterprise workspaces with an approved deferral get until February 11, 2027.
• Migration is manual. You start it yourself from My GPTs with "Migrate to plugin," and nothing happens if you do nothing except eventual shutoff.
• Instructions, knowledge files, and connected apps transfer. Custom actions, conversations, sharing settings, and model choice do not.
• Test the migrated plugin against real prompts before retiring your workflow, because OpenAI says replacements may respond differently.
• If you rely on someone else's GPT, contact the creator now or plan to rebuild it yourself.

Deadlines like this one are easier with company. Willa members compare notes on exactly these migrations at our weekly events, often live with screens shared, and someone in the room has usually already hit the snag you are about to. Come to an upcoming event, or join Willa and bring the GPT you are not sure how to save.

• ChatGPT
• custom GPTs
• plugins
• OpenAI
• AI tools

Join Willa, the community where women learn AI together.

All posts