The custom GPT retirement is set for 11 December 2026, on every ChatGPT plan. After that date the GPTs and their pages stop being accessible. The replacement is something OpenAI calls plugins, and there is a migration route.
The migration is the easy part. The hard part is that in most small firms, nobody could tell you how many custom GPTs the business relies on, or whose account they sit in.
The GPT belongs to whoever built it
A custom GPT usually starts as one person’s good idea. Someone in the office builds a GPT that drafts quotes in the house style, or answers the same five customer questions, or turns site notes into a report. It works, it gets shared round, and within a few months it is part of how the job gets done.
It also lives in that one person’s ChatGPT account. OpenAI’s own guidance is plain on this: permission to use someone else’s GPT does not give you permission to migrate it. Only the creator can press the button. If that person has left, changed roles or built it on a personal login, the business is relying on something it does not control.
This is the same pattern as a Microsoft 365 account only one person can get into. The software is fine. The ownership was never decided.
What moves across and what does not
When migration is available on your account, the creator goes to My GPTs and selects Migrate to plugin. The GPT’s instructions and knowledge files come across. Several things do not:
- Conversations stay behind in the old GPT.
- The model the GPT was set to use does not carry over.
- Custom actions, the bit that connects a GPT to another system, do not transfer. They need rebuilding separately, and that can be a technical job.
- Sharing settings do not carry over. The new plugin starts private, so everyone who used the old GPT loses access until it is shared again.
Migration also takes the last published version. Drafts and unpublished edits are left behind, and once migrated the original GPT becomes read-only.
Plan for the custom GPT retirement now, not in December
The custom GPT retirement does not need a project. It needs a list.
- Ask the team which GPTs they use for actual work, not experiments.
- For each one, note who created it and on which account.
- Copy the instructions and reference files somewhere the business owns, such as a shared drive. That is your process written down, and it is worth having whatever tool runs it next.
- Get the creator to migrate it, then run the same few familiar requests through the new version and compare the results.
- Share it back to the people who need it and check they can open it.
Anything with a custom action connecting to your CRM or job system goes to the top of the list, because that is the part that will not come across on its own.
The wider lesson is the one from the CRM automations I found switched on and doing nothing. Anything the business runs on needs a named owner and a written description of what it does, whether it lives in a CRM, a spreadsheet or a chatbot.
If you want help working out which of your tools the business actually depends on, the free audit at digilyse.co takes about ten minutes.
Want this kind of analysis monthly?
The Digilyse Brief is a monthly LinkedIn newsletter from Tom Burke covering what works, what doesn’t, and the operational patterns we keep seeing across owner-managed UK firms. No pitching, no hype. Subscribe on LinkedIn.
Published by Digilyse. Practical systems for growing businesses.