Custom GPTs Retire on 11 December 2026: What Businesses Should Do Now

If your team has built a custom GPT, or quietly relies on one someone else built, there is now a date to plan around: 11 December 2026. OpenAI is retiring custom GPTs and moving reusable workflows towards plugins. Some dates vary by plan and workspace, so check your own ChatGPT admin settings for the exact timing.

This is not a crisis. It is a prompt to do something most businesses never get round to: write down which AI workflows they depend on, and make sure they are properly looked after.

What is changing

OpenAI describes a plugin as a reusable package that can hold skills and connections to other apps. When a custom GPT is migrated:

• The GPT's instructions become a skill.
• Its knowledge files become reference files.
• The original GPT becomes read-only once it has been migrated.

What does not transfer

Check these carefully before you assume a migrated workflow behaves the same way:

• Custom actions do not transfer, so anything your GPT did by calling an external service needs to be rebuilt or reconnected.
• The model you originally chose does not carry over.
• Behaviour can differ even with identical instructions, so a migrated workflow needs testing.

A practical checklist before December

1. 1.**List every custom GPT you use.** Include ones built by colleagues, contractors or former staff. Ask the team directly; many are shared by link and never recorded anywhere.
2. 2.**Decide which ones matter.** Rank them by how often they are used and what would break if they stopped working.
3. 3.**Save a copy of the instructions and files.** Store them somewhere your business controls, not only inside ChatGPT.
4. 4.**Note any custom actions.** List what each action connects to and who owns that connection.
5. 5.**Migrate the ones you use.** Do this deliberately rather than in a rush in December.
6. 6.**Test against real examples.** Run five to ten genuine past tasks through each migrated workflow and compare the results with what you expect.
7. 7.**Decide who owns each workflow.** Name a person responsible for keeping instructions, files and connections up to date.
8. 8.**Retire what you no longer use.** Fewer, better-maintained workflows are safer than a long list nobody checks.

Where plugins fit, and where they do not

A plugin is a good home for a reusable workflow that depends on your instructions, reference documents and a few approved apps. It is not automatically the right answer for everything.

If your real need is to move data between systems when something happens, such as a new enquiry, a signed form or a paid invoice, that is usually process automation rather than a chat-based workflow. If you want AI to work with live systems in a standardised way, MCP and APIs may be part of the picture. The right choice depends on the job, not on whichever tool is newest.

Keep a human in the loop

Whatever format your workflow moves into, decide where a person approves the output. Customer-facing messages, financial figures, compliance wording and anything involving personal data should keep a human approval step. Migrating a workflow is a good moment to write those approval points down.

Do not depend on one platform forever

The wider lesson is that AI platforms change formats, features and pricing regularly. Keep your instructions, reference material and test examples in a form you own, so that moving a workflow is a short job rather than a rebuild. That is the approach we use at GrowthZone AI, and the one we recommend to clients. You can see how it applies to our own AI roles on the Meet the Assistants page.

Frequently asked questions

When do custom GPTs stop working?

OpenAI has set 11 December 2026 as the end date for custom GPTs. Some dates vary by plan and workspace, so check your own account or admin settings.

What happens to my GPT's instructions and files?

When migrated, the instructions become a skill and the knowledge files become reference files inside a plugin. The original GPT becomes read-only once migrated.

Do custom actions move across?

No. Custom actions do not transfer, so any external connections need to be rebuilt or reconnected in the new format.

Will a migrated workflow behave the same way?

Not necessarily. The model you picked does not carry over, so you should test each migrated workflow against real examples before relying on it.

Is a plugin the same as an automation?

No. A plugin packages reusable skills, reference files and app connections for an AI workflow. An automation moves data between systems and runs when an event happens. Many businesses need one, the other or both.

Where should I start?

Make a list of every custom GPT your team uses, save a copy of the instructions and files, and migrate the most important ones first. If you want a second pair of eyes, talk to Kaye.