Custom GPTs Stop Running December 11 — the Small-Business Migration Plan Before ChatGPT's Retirement

Flat-vector illustration of a robot figure carrying document files from an old fading cabinet into a fresh glowing container, with a calendar icon and arrow icons marking the path, on a dark navy background with electric-blue accents

On December 11, 2026, OpenAI's custom GPTs — the build-your-own ChatGPT assistants that millions of businesses use for everything from drafting estimates to answering customer questions — are scheduled to stop running entirely. OpenAI's migration FAQ puts it plainly: at retirement, "custom GPTs stop running and leave the GPT directory." The replacement is a new format called plugins, and getting there is not automatic: parts of your GPT will not survive the move unless you plan for them.

The transition affects all ChatGPT plans, not just enterprise accounts. If a small business in Temecula or Murrieta built a GPT two years ago to write follow-up emails or quote siding jobs, that tool has a scheduled shutdown date — and the pieces that made it actually useful (its custom actions, its selected model, its conversation history) do not transfer on their own. Here is the full timeline, what survives, what breaks, and a concrete migration plan you can run before the December deadline.

What Is Actually Happening

Custom GPTs were introduced in November 2023 as ChatGPT's no-code way to package instructions, reference files, and connected apps into a reusable assistant. This year OpenAI has been consolidating that world: it replaced the App Directory with a Plugin Directory on July 9 when it launched ChatGPT Work, and it has already retired the official DALL·E GPT. The retirement of custom GPTs entirely is the next step in that consolidation, and the replacement — plugins — can bring reusable instructions, reference files, and connected apps together so your guidance works alongside the tools a task needs, across Chat, Work, and Codex.

Part of the change has already landed. As MIXED News reported, OpenAI's GPTs help page now says new GPT creation and publishing are no longer available on personal accounts — Free, Go, Plus, and Pro — while existing GPTs remain usable. In other words: you cannot build a new custom GPT on a personal plan anymore; you can only manage the ones you already have, until December 11.

The Timeline: Four Dates That Matter

OpenAI's migration FAQ gives firm milestones for affected Enterprise workspaces, and says other plans may follow the same transition timeline, with in-product notices telling you when each step applies to your account:

September 11, 2026 — admin notice sent to help Enterprise organizations prepare.
September 22, 2026 — target date for the migration experience and user banner. OpenAI stresses this is a target, not a guarantee that the option is available in every workspace.
October 26, 2026 (planned) — creation of new custom GPTs ends. Any drafts you need to migrate must be published before this cutoff; publishing does not require sharing publicly.
December 11, 2026 — scheduled retirement. Custom GPTs stop running and leave the GPT directory.

If you use several accounts or workspaces, the FAQ has a detail many people miss: the timeline and permissions of the account where the GPT was created apply — not the account you happen to use it from. A public GPT built inside a company's Enterprise workspace follows the Enterprise transition even if you open it from a personal account, and access to the original does not guarantee access to its replacement.

What Survives the Move — and What Breaks

Under the planned migration workflow, three things carry over: the GPT's instructions become a skill inside the new plugin, its knowledge files are copied into the plugin's reference files, and its connected apps are added as apps. That covers a simple GPT — a persona, a knowledge base, a set of files — reasonably well.

What does not transfer is the longer list, and it is where working business tools quietly break:

Custom actions. The FAQ is direct: "GPT custom actions do not transfer through the migration workflow." Any GPT that calls an external API — pulls job data, checks inventory, writes to a CRM — needs that integration rebuilt, possibly through a custom MCP server and technical setup.
The selected model. The GPT's chosen model does not carry over. In Enterprise workspaces, workspace defaults apply.
Conversation history. Existing GPT conversations do not move into the plugin.
Sharing settings. Migration does not carry over sharing or give existing users access. The plugin starts private; making it public is a separate submission process, and rebuilding sharing for a team means re-inviting everyone.

Two more mechanics matter. Migration uses the GPT's latest published version — drafts and unpublished edits do not transfer, so finish essential edits and publish before migrating (publishing does not require public sharing). And after you migrate, the original GPT remains usable until retirement but becomes read-only, and its creator can no longer delete it.

OpenAI also warns that a migrated plugin "may respond differently," and recommends testing familiar prompts plus at least one harder case before relying on it — checking that it selects the right skill, uses the expected reference material, and produces answers in the required format.

The Migration Plan for a Small Business

OpenAI's own advice is to start now: identify the GPTs you rely on, whether you created them or use someone else's, and allow time to test a replacement. Turned into a working checklist for a five-person shop:

Inventory every GPT your business touches. Include ones you only use, not just ones you built. For each: who made it, what it does, and whether it has custom actions. If it is someone else's GPT, you cannot migrate it yourself — check its notice and the creator's guidance about a replacement, and remember access to the original does not guarantee access to the new one.
Export the knowledge that lives inside it. Copy out the instructions and every reference file. If the GPT's knowledge base is the only place your pricing rules, FAQ answers, or product specs are written down, this retirement is your reminder that they belong in a document you own, not inside a rented feature.
Save test prompts before anything changes. A handful of familiar prompts — a real estimate request, a real customer question, your hardest edge case — so you can compare the replacement's output side by side instead of guessing whether it got worse.
Publish, then migrate, then test. Finish essential edits, publish the GPT (privately is fine), run the migration when your account has the option, and test the plugin with your saved prompts before the December deadline. Only a GPT's creator or a workspace admin can migrate it, and drafts cannot be migrated.
Rebuild custom actions first, not last. If your GPT talks to another service, that integration is manual work — check whether an available app supports the task, or budget for a custom MCP server. Test the rebuilt connection separately; OpenAI explicitly says a rebuilt integration should not be assumed to provide every capability of the original.

The Temecula and Murrieta Angle

Plenty of local businesses quietly run on a custom GPT: the contractor whose GPT drafts scope-of-work language, the salon that built one to answer the same twenty booking questions, the shop using one to polish review responses. For most of them the migration will take an afternoon — export files, publish, migrate, test. The ones that get burned are the ones with custom actions wired into scheduling, invoicing, or a CRM, and the ones who never notice the in-product notices until the tool stops on December 11.

There is also a bigger lesson in this retirement, and it is the same one every platform consolidation teaches: a workflow that lives entirely inside someone else's product can be sunset by a FAQ post. The durable version keeps the business logic — your pricing rules, your service descriptions, your customer questions — on assets you own, like your website, and uses platform tools as connective tissue rather than as the foundation. If you want the full picture of building on owned infrastructure instead of rented features, our guide to self-hostable, open-source AI employees covers the independent lane, and our earlier ChatGPT small-business program guide covers the ChatGPT-native side. This December 11 deadline is simply the newest reason to know which side of that line each of your tools stands on.

The Bottom Line

Custom GPTs stop running on December 11, 2026 — that is scheduled, published, and firm for Enterprise, with the same transition rolling out to every other plan through in-product notices. The businesses that come out clean are the ones that inventory now, export their knowledge, save test prompts, and rebuild custom actions early. And if you would rather have this handled — migrating what you built, or rebuilding the workflow on your own site so no platform's retirement can switch it off — book the free 15-minute call. PepeWebTech builds AI-powered sites and automations for small businesses at flat monthly pricing published on our pricing page, with no long-term contracts — we earn your business every month. Our blog library has over a hundred more guides on getting AI to actually work in a small business.