AI Chatbots
Custom GPTs Are Retiring on December 11, 2026: What Happens to Yours and What to Do Now
The short answer: OpenAI is retiring custom GPTs on December 11, 2026 (some Enterprise workspaces with an approved deferral get until February 11, 2027). On that date GPTs stop running and their pages go away. You can turn each one into a ChatGPT plugin with one click: your instructions become a skill and your knowledge files come along, but custom actions, the selected model and past chats don't carry over. If your GPT calls your own tools or serves your customers, plan a rebuild, not a click.
The date comes from OpenAI's custom GPT retirement FAQ, and the migration details below from its guide to moving GPTs to plugins. Below, I sum them up, then give you my opinion as someone who has built more than 50 chatbots: for a lot of businesses, this deadline is a good moment to ask whether that assistant should live inside ChatGPT at all.
A custom GPT is a kind of AI chatbot that only exists inside ChatGPT. If you want the basics of how chatbots work outside of it, this guide explains what an AI chatbot is. The rest of this page assumes you already have a GPT and a deadline.
Further down: the one question that tells you whether "Migrate to plugin" is enough for your GPT.
What happens to custom GPTs on December 11, 2026?
They stop running and leave the GPT directory. The original GPT stays usable until then. After that date, the GPT and its page are no longer accessible, according to OpenAI.
The key dates, as OpenAI publishes them:
| Date | What happens | Who | |---|---|---| | October 26, 2026 | New GPT creation ended | Affected Enterprise workspaces (not FedRAMP) | | December 11, 2026 | Custom GPTs retire: they stop running, pages go away | All plans, unless your workspace notice says otherwise | | February 11, 2027 | Later retirement date | FedRAMP workspaces and Enterprise workspaces with an approved deferral |
Check the notice in your own account or workspace: OpenAI says timelines can differ by plan.
What carries over when you migrate a GPT to a plugin?
Your instructions and files come over; your custom actions, model choice and conversation history don't. Here's the full list from OpenAI's migration guide:
| Part of your GPT | After "Migrate to plugin" | |---|---| | Instructions | Become a skill inside the plugin | | Knowledge files | Come over (review them after migration) | | Connected apps | Stay part of the plugin | | Custom actions (your own API calls) | Do not transfer. Rebuild them with a supported app connector or a custom MCP server | | Selected model | Does not carry over; workspace defaults apply | | Conversation starters | Don't transfer one-to-one | | Previous chats | May not copy: save the prompts and context you need | | The original GPT | Becomes read-only once migrated | | Sharing | The new plugin starts private: share it again yourself |
Two more details that catch people out:
- Only the creator (or a workspace admin) can migrate, and the GPT must be published. If you use someone else's GPT, ask its creator to migrate it.
- The plugin may answer differently. OpenAI says so itself: model, output quality and instruction-following can change. Test before you tell your team it's done.
Converting a GPT to a plugin doesn't consume credits, per OpenAI; on credit-based pricing, editing the plugin afterwards does.
How to migrate a custom GPT: 6 steps before the deadline
Make a list, save everything, migrate, rebuild the actions, test, then share. In order:
- List every GPT you own and what each one does. Mark the ones that use custom actions or that customers use.
- Save the instructions and knowledge files somewhere outside ChatGPT. Copy the useful prompts from past chats too.
- Migrate: My GPTs → Created by me → Migrate to plugin. Admins can do several at once with the Admin plugin; bulk migration doesn't copy sharing settings.
- Rebuild the custom actions as an app connector or an MCP server. This is the only technical step, and the one that breaks quietly if you skip it. If the action calls your own tools, an automation platform can host it: n8n has an MCP Server Trigger that exposes a workflow as tools (here's how n8n compares with Make).
- Test with your usual tasks plus one hard case. Try it both by picking the plugin and by asking normally, since skills are picked automatically from the request.
- Share it again with the right people, and name one owner for future changes.
Is "Migrate to plugin" enough for your GPT? (my take)
If your GPT is a personal or internal helper with instructions and files, yes: migrate it and move on. If it calls your own systems or talks to your customers, the deadline is telling you something.
Here's the question I'd ask: who uses it, and what does it touch?
- Internal helper, no actions (a writing assistant, a policy Q&A for your team): the one-click migration covers you. Do it this month, test it, done.
- Uses custom actions (it checks orders, books slots, writes to your CRM): you have to rebuild that part anyway. Since you're rebuilding, decide where it should run.
- Customer-facing (you sent customers a GPT link): your customers need a ChatGPT account to use it, the plugin starts private, and OpenAI can change the rules again, as it just did. A chatbot on your website or WhatsApp is yours: your channel, your data, your branding, and no account needed for your customers.
That last case is where I'd stop patching. When we build a chatbot for a business, it runs on their site or on WhatsApp, connected to their tools through Make or n8n, with the AI model as one replaceable part. If OpenAI retires a feature, you change one module, not the whole assistant. If you go that way, here's how building an AI chatbot works, step by step.
What this costs is the natural next question. A simple bot on a no-code tool runs tens of dollars a month; one with real integrations costs more to build. I put real numbers in what an AI chatbot costs in 2026.
Common questions about the custom GPT retirement
Are custom GPTs being discontinued?
Yes. OpenAI's FAQ schedules the retirement for December 11, 2026, with plugins as the replacement. Some Enterprise and FedRAMP workspaces get February 11, 2027.
Can I still use my GPT until December 11?
Yes, the original keeps working until retirement. If you migrate it, the original becomes read-only and you edit the plugin from then on.
What replaces custom GPTs?
Plugins, which bundle skills (your instructions), files and connected apps. Custom actions must be rebuilt as an app connector or an MCP server.
Do I lose my conversations?
Past chats don't move into the plugin, and OpenAI advises saving the prompts and context you need. Don't count on old GPT chats as your archive.
Don't wait for the last week. The click takes seconds; rebuilding custom actions and testing that the plugin behaves like the old GPT is what takes time. Start with the GPTs that touch customers or your own systems.
Have a GPT that customers or your team rely on, with actions you'd rather not rebuild inside ChatGPT? Tell me what it does and I'll tell you whether a plugin is enough or whether it should become a website or WhatsApp chatbot you own. You can also see our AI chatbot service.
Keep reading
Not sure if your content is helping or hurting you?
Get a free content & SEO audit. We'll tell you exactly where you stand — no jargon, no fear-mongering.
Get my free audit