Quick Answer
Custom GPTs retire on December 11, 2026, or February 11, 2027 for FedRAMP workspaces and approved deferrals. OpenAI's built-in migration turns a GPT's instructions into a skill inside a ChatGPT plugin, copies its knowledge files and adds its apps. Custom actions don't transfer. If your team only uses ChatGPT, that migration may be all you need. If it also uses Claude, Copilot, Cursor or Grok, turn each GPT's instructions into a skill every AI can use, and keep its files in your own SharePoint or Google Drive.
What's Changing and When
These are the dates in OpenAI's retirement FAQ. Follow the notice in your own account or workspace, since some workspaces have a different timeline.
| Date | What happens |
|---|---|
| September 11, 2026 | OpenAI sends Enterprise admins a notice to help them prepare |
| October 26, 2026 | New GPT creation ends in affected Enterprise workspaces, except FedRAMP workspaces |
| December 11, 2026 | Standard retirement: custom GPTs and their GPT pages become inaccessible |
| February 11, 2027 | Retirement for FedRAMP workspaces and workspaces with an approved deferral |
Until then, existing GPTs keep working. Once you migrate a GPT, the original becomes read-only, so finish any edits first. Your past conversations with GPTs stay accessible after retirement.
What OpenAI's Migration Moves, and What It Doesn't
When migration is available in your account, open My GPTs and select Migrate to plugin. Here's what happens to each part of the GPT:
| Part of the GPT | After migration |
|---|---|
| Instructions | Become a skill inside a new ChatGPT plugin |
| Knowledge files | Copied into the plugin's reference files |
| Connected apps | Added to the plugin as apps |
| Custom actions | Don't transfer. Rebuilding one may need a custom MCP server |
| Selected model | Doesn't carry over |
| Sharing | The plugin starts private, so you share it again |
| Drafts | Only the latest published version migrates |
The result is a ChatGPT plugin. It works in ChatGPT, and nowhere else.
When ChatGPT's Own Migration Is Enough
If everyone who uses the GPT works only in ChatGPT, OpenAI's migration is the quickest route. For a GPT that was really a personal workspace, such as notes and files for ongoing research, OpenAI suggests a ChatGPT project instead.
It's worth doing more when the GPT holds a way of working the whole business relies on, and the team doesn't all use ChatGPT. Some people may be in Claude or Copilot, developers in Cursor, someone else in Grok. A ChatGPT plugin reaches none of them, so the same instructions end up copied into every tool and drift apart.
How to Move a GPT to a Skill for Every AI
- List your GPTs. For each one, note who created it, who uses it, whether it has knowledge files and whether it has custom actions. Only a GPT's creator or a workspace admin can migrate it in ChatGPT.
- Copy the instructions. Open the GPT's settings and copy its instructions in full, before you migrate it and the original becomes read-only.
- Move the knowledge files. Put them in SharePoint or Google Drive, in the folder for the team that uses them, rather than inside any one AI tool.
- Turn the instructions into a skill. Paste them into GPT Rescue, or the Skill Builder's Improve a skill tab. It shapes them into a skill your AI can follow: when to use it, what to collect, the steps, the output and the checks.
- Point the skill at the files. Name the documents the skill needs rather than pasting their contents in, so updates to the files reach the skill without editing it.
- Publish it for the team. In Contexta, every connected AI fetches the published version the next time it's asked, so ChatGPT, Claude, Copilot, Cursor and Grok all run the same skill.
- Rebuild custom actions separately. If the GPT called another system through a custom action, check whether your AI already has a connector for it. If not, it needs an MCP server. That's the same work whichever route you take.
Worked Example: The Proposal Writer GPT
A 40-person consultancy built a Proposal Writer GPT in 2025. Its instructions set the structure of a proposal and the house style, and its knowledge files held the rate card, two winning proposals and the brand guide. The partners use Copilot, the consultants use ChatGPT and Claude.
Migrating it in ChatGPT would have kept it working for the consultants in ChatGPT only. Instead, the rate card and sample proposals went into the SharePoint folder for the sales team. The instructions went into the Skill Builder and came back as a proposal skill that names those files. The house style moved into the company context, so every skill writes in it, not just this one.
Now the partners in Copilot and the consultants in ChatGPT and Claude all use the same skill. When the rate card changes, someone updates one file in SharePoint.
Test It: Three Cases
Before December 11, run the new skill against three prompts you used with the GPT, and compare the results with what the GPT produced:
- A normal request. The job the GPT did most often. The output should match in structure, tone and length.
- A request with something missing. Leave out a detail the GPT needed. A good skill asks for it rather than guessing.
- An edge case. An unusual scope, a discount or a client the GPT handled differently. Check it follows the same rule.
Run the same three cases in a second AI tool your team uses. If the results differ, the fix usually belongs in the skill's steps or checks, not in the AI.
Where Contexta Fits
Contexta holds your company's skills and context in one place and delivers them to Claude, ChatGPT, Copilot, Cursor and Grok through one connection. Your files stay in your own SharePoint or Google Drive. Every change to a skill is versioned, so you can see who changed what and restore an earlier version. It's free for one person. See How It Works and Integrations.