The short answer
A custom GPT is a version of ChatGPT that someone set up for one job, with saved instructions, uploaded knowledge files, chosen tools and optional custom actions that call outside services. OpenAI is retiring custom GPTs on 11 December 2026 (11 February 2027 for Enterprise workspaces with an approved deferral). Creators can move a GPT to a ChatGPT plugin from My GPTs with Migrate to plugin. Instructions and files become a skill, but custom actions do not transfer and must be rebuilt.
| Made by | OpenAI, inside ChatGPT |
|---|---|
| Status | Retiring on 11 December 2026 |
| Later date | 11 February 2027 for Enterprise workspaces with an approved deferral |
| Replaced by | ChatGPT plugins, which hold skills, apps or both |
| How to migrate | My GPTs, then Migrate to plugin on the GPT |
| Does not transfer | Custom actions |
| Google's equivalent | Gems, which become Gemini skills from 17 November 2026 |
What a custom GPT is
A custom GPT is ChatGPT with a job description. Instead of explaining your brand voice, your process or your reference material in every chat, you saved it once: a name, instructions, knowledge files such as a style guide or a product catalogue, the tools it could use (web search, image generation, data analysis) and, for more advanced builds, custom actions that let it call another service. You opened the GPT, or mentioned it with @, and it behaved the same way every time. Teams used them for things like a proposal writer, an HR policy helper or a first-line support assistant.
Why custom GPTs are being retired
OpenAI is moving custom GPTs to plugins. In OpenAI's words, a plugin is a reusable package for a workflow that can include skills, apps or both. A skill explains how to do a task, such as writing in your team's style. An app connects ChatGPT to another service for information or actions. Plugins can also include MCP servers, the open standard for connecting AI to tools, which is explained in what is MCP.
The key dates, from OpenAI's custom GPT retirement FAQ:
- 11 December 2026: custom GPTs retire.
- 11 February 2027: the later date for Enterprise workspaces with an approved deferral.
The biggest practical difference is how the work starts. With a GPT, you chose it explicitly. With plugins, you can still pick one with an @ mention or from the + menu under More, but ChatGPT can also use a relevant installed skill automatically when its description matches your task. OpenAI warns that an installed plugin will not run on every request, so you need to test both ways.
What transfers and what does not
| Part of your GPT | After migration |
|---|---|
| Instructions | Become a skill |
| Knowledge files | Become reference files in the skill; review and test how they are used |
| Connected apps | Remain part of the plugin |
| Custom actions | Do not transfer. Rebuild with a supported app's actions or a custom MCP server |
| Sharing | Each replacement plugin starts private; it is shared or published under workspace permissions, so review access before rollout |
Only the GPT's creator or a workspace owner or admin can migrate it, and OpenAI's guide is about published GPTs, telling creators to publish any drafts they need while GPT creation is still available. If you use someone else's GPT, ask its creator to migrate it and share the plugin with you.
How to migrate a custom GPT, step by step
- Copy the GPT's full instructions into a document you own and list its knowledge files and any actions.
- Save three or four prompts you use often, with good past answers, including one hard case. You will use these to test the replacement.
- Publish the latest version if you have unpublished edits.
- In ChatGPT, open My GPTs, choose Migrate to plugin on the GPT and follow the prompts.
- Test the plugin with your saved prompts, both by selecting it with @ and by just describing the task.
- If it does not pick up the skill when you expect, make the skill's description more specific about when it should be used.
- Rebuild any custom action as an app connection or MCP server, with help from IT if it touches company systems.
For the wider picture of what plugins and skills can do today, see our ChatGPT guide and the step-by-step guide on creating a custom GPT.
Why it matters at work
If your team relies on shared GPTs, this is a deadline with consequences: a GPT that is not migrated stops working. The work is mostly small, but the actions part can take real time, and someone has to own each migration. OpenAI's guide for admins suggests naming an owner for each GPT and recording its purpose, audience, instructions, files and integrations. Our guide to rolling out AI to your team has a simple template for that kind of inventory.
Google is making the same move. On personal Google accounts for adults, Gems become read-only on 17 November 2026 and are converted to Gemini skills automatically, and the Gems are removed on 17 December 2026. Workspace business accounts follow in March 2027 and education accounts in June 2027. Google notes that most default Gem tools, including Canvas, Deep research, Create video, Create music and Guided learning, do not work with skills, and that skills have a 100 MB file limit. See our Gemini guide. Claude users can get similar results with Projects and skills, as our guide to setting up Claude Projects explains.
An example you can try
Even if you never built a GPT, you can create the replacement idea in ten minutes: a reusable instruction set for one recurring task. Use the first prompt below to turn a task you do weekly into skill instructions, then paste them into a ChatGPT skill, a Gemini skill (Customize, then Skills, then Create and Create manually) or a Claude Project. Run it on last week's real example and compare the result with what you produced by hand.
Common misconceptions about custom GPTs
- "My GPT will be converted automatically." On ChatGPT, migration is something the creator or an admin does. Google converts Gems automatically; OpenAI's process starts from My GPTs.
- "Everything carries over." Custom actions do not, and you should test how knowledge files behave once they are reference files in a skill.
- "A plugin works just like a GPT." Not quite. A skill may load automatically when relevant rather than every time, so wording its description well matters.
- "A custom GPT was a separately trained model." It was not. It used ChatGPT's normal models with saved instructions and files. That is closer to a system prompt plus documents than to fine-tuning.
Related terms
- System prompt: what a GPT's instructions really were.
- MCP: one way to rebuild custom actions.
- RAG: how assistants use knowledge files to answer.
- AI agent: where plugins and skills are heading.
- Prompt: the building block of every skill.
Weighing a move instead of a migration? See ChatGPT alternatives and ChatGPT vs Claude.
Prompts to copy
Turn a weekly task into skill instructions
I do this task every week: [describe the task, for example turn sales call notes into a client follow-up email]. Ask me up to five questions about the inputs, the output format, the tone and common mistakes. Then write skill instructions with: a one-sentence description of when to use the skill, the step-by-step method, the output format and three rules it must never break.
You get: Ready-to-paste instructions for a ChatGPT skill, Gemini skill or Claude Project
Audit a GPT before you migrate it
Here are my custom GPT's instructions and a list of its knowledge files and actions: [paste]. Tell me which parts will move into a plugin skill, which parts are custom actions that need rebuilding, and any instructions that are vague, outdated or contradictory. Then rewrite the instructions as a clean skill, keeping my intent.
You get: A migration checklist and cleaned-up instructions