The short answer: OpenAI is retiring custom GPTs on December 11, 2026 and replacing them with plugins. The built-in migration moves instructions, knowledge files and connected apps across — but drops custom actions, the chosen model, conversations and sharing settings. The date that matters most is October 26, not December 11: migration copies the latest published version, so publish a final version of everything you rely on before then, then migrate, test and re-share.
If you built a custom GPT for your business — a quoting helper that knows your price list, an onboarding assistant trained on your policies — OpenAI has put an expiry date on it. The plan is published in OpenAI’s own retirement and migration FAQ, which we read directly, and cross-checked against secondary walkthroughs from Digital Matters and Twister Hub. You have time and a migration path — but the button does not move everything, and what it drops (the connections to your other software) is what hurts.
The Deadline Everyone Quotes Is the Wrong One
Every headline says December 11. That is the day the old GPTs stop running, but it is not the date that decides how painful this is. OpenAI’s FAQ adds a second date: October 26, 2026 (planned), when new GPT creation ends for affected Enterprise workspaces. It sounds like a builder’s problem, but the migration copies the GPT’s latest published version — drafts and unpublished edits do not transfer — and once you migrate, the original becomes read-only, so you cannot fix instructions afterwards.
The real sequence, then: finish your edits before you migrate, publish that version (publishing does not require sharing it publicly), and take your time pressing the button. There is no prize for migrating early — the original keeps working until retirement. The one thing you must not leave late is publishing.
These dates are subject to change — the Enterprise dates already moved once (September 25 → October 26). Treat the FAQ as current. And vendors switch tools off whenever it suits them: the Atlas shutdown story is what an unplanned migration looks like.
What the Migration Button Carries — and What It Drops
When the migration option reaches your account, go to My GPTs and select Migrate to plugin. Exactly what survives the trip:
Three consequences deserve more than a table row. First, custom actions are the hard part. If your GPT calls your booking system or CRM through an action, that connection does not come across. OpenAI says rebuilding it “may require a custom MCP server and technical setup” — a developer’s job, not a click. Your plan decides whether you can even build one: Free and Plus cannot, Pro is read-and-fetch only, Business and Enterprise get read and write.
Second, sharing resets completely. The plugin starts private; nothing installs it for anyone and nobody who used your GPT gets the replacement automatically. Going public needs a separate submission process OpenAI has not documented yet. One trap: public GPTs created in an affected Enterprise workspace follow the Enterprise timeline even when used from a personal account — what matters is where the GPT was created, not which plan you are on.
Third, a correction to the secondary coverage. Several guides say past conversations are lost. OpenAI’s FAQ is more specific: your existing conversations with your GPTs will remain accessible after retirement; they just do not move into the plugin. When a secondary source and the FAQ disagree, trust the FAQ.
Not Every GPT Deserves a Plugin
Here is the call the migration button does not make for you: a plugin is the wrong container for some GPTs. A GPT that is really just a well-written prompt with no files and no connections may be better recreated as a standalone skill — a point Jen Taylor makes in her DevDay migration write-up: for a focused workflow like a copyeditor, she prefers a standalone skill; a plugin only earns its complexity when you package related workflows together.
So sort your GPTs into three piles before you touch the button. Button-migratable: instructions plus knowledge files, no custom actions — publish the final version and migrate. Developer rebuild: anything with custom actions — the API work mostly survives but authentication and permissions need a rebuild and a security review; start this pile first, it is the bottleneck. Retire: GPTs nobody opens or that duplicate something ChatGPT now does natively.
The discipline is the same as evaluating any AI tool for your business: judge the workflow, not the container. If you built the GPT following a guide like our custom GPT build walkthrough, you already have the instructions and files — the asset was never the GPT, it was the configuration work.
Move a Custom GPT to a Plugin, Step by Step
OpenAI’s FAQ describes the migration; this is the practical order for small teams:
- Inventory what you have. Open My GPTs and list the ones your team genuinely relies on. For each, note: who created it, who uses it, whether it is published, and its files, apps and custom actions. The workload planner below shows how big the job is.
- Finish and publish the final version. Migration copies the latest published version only; drafts and unpublished edits are left behind. Publishing does not mean making it public. Do this before the October 26 milestone if you are in an affected Enterprise workspace — and as soon as possible everywhere else.
- Save your source material outside ChatGPT. Copy the full instructions, download every knowledge file, and document the conversation starters and any actions. This is the master copy: whatever OpenAI does to the container next, your asset is the text and the documents.
- Do the prep while waiting for the button. Migration is rolling out at different times; a missing option may just mean it has not reached you yet. Steps 1–3 need no button, so do them now.
- Migrate one low-risk GPT first. Pick a simple staff tool, migrate it, and review what arrived: the skill, the reference files, the apps. Reconnect the apps — each needs its own authorization.
- Rebuild custom actions as MCP integrations. Check whether an available app already covers the task before building. Rebuilds need a developer and a security review.
- Test before trusting. Run your normal prompts plus one harder case. OpenAI warns the plugin may respond differently — compare, do not assume.
- Re-share deliberately. The plugin starts private. Have one intended user install and test it first. Then update every place the old link lived — staff docs, client emails, your website, handouts. If the GPT was customer-facing and a plugin no longer fits, our Tidio vs Lindy vs custom GPT comparison looks at alternatives for small-business customer support.
On cost: in Enterprise workspaces, using a migrated GPT without apps in Instant Chat has no incremental cost; creating and editing plugins through a conversation uses Work or Codex, and actions in Work consume credits. If the migration makes you reconsider your plan tier, see our API vs Plus vs Business breakdown and the guide to AI API spend limits. Our automation guide covers staging a tooling rollout without breaking daily work.
Plan Your Custom GPT Migration Workload
How big is this job, really? Most teams overestimate the clicking and underestimate the rebuilding and the re-sharing. Plug in your own numbers — the assumptions are our estimates, labelled as such, not OpenAI’s figures. Defaults are examples only.
How the estimate is built (our calculation): each relied-upon GPT costs about 2 hours for inventory, final edits, publishing, migration and testing — an example figure, not an OpenAI number. Each one with custom actions adds 8 hours for the developer rebuild and security review, plus 30 minutes per user for re-sharing. The weekly figure divides the total by the days left until December 11, 2026.
What We Could Not Confirm
Honesty requires the gaps. OpenAI’s FAQ is thorough, but four things are not in it: a creation cutoff for non-Enterprise plans; how an Enterprise workspace qualifies for the February 2027 deferral; how public plugin submission works and whether Free accounts can install public plugins; and whether Plus and Pro creators get the migration button. And one meta-warning: these pages changed once in September and may change again before December. Re-check the FAQ the week you act, not the week you read this.
The Short Version
A custom GPT was never a product you owned — it was a configuration inside somebody else’s app, and the vendor changed the container. The instructions you wrote and the documents you uploaded are the asset: keep them somewhere you control, and the next migration is an upload rather than a rebuild. Publish before October 26, move the simple GPTs with the button, hand the connected ones to a developer, re-share deliberately, retire the ones nobody opens. Treat December 11 like a planned outage — our AI tool outage checklist (inventory, backups, alternatives, communication) works here too.
Frequently Asked Questions
When is OpenAI retiring custom GPTs?
December 11, 2026, per OpenAI’s migration FAQ — the date the GPTs and their pages become inaccessible. Enterprise workspaces with an approved deferral get until February 11, 2027. The dates are subject to change, so re-check the FAQ before planning around them.
What happens if I do nothing before December 11?
Your GPTs stop running and their pages go dark. Unless you copied out the instructions and files, you rebuild from memory. Conversations stay accessible after retirement, per the FAQ — but the workflows are gone.
Does the migration move my conversations into the plugin?
No. The FAQ says your existing conversations will remain accessible after retirement, but they do not move into the plugin — the new plugin starts with no history.
My GPT uses a custom action. What do I do?
Budget for a rebuild, not a migration. Custom actions do not transfer; OpenAI says rebuilding “may require a custom MCP server and technical setup.” Whether you can build one depends on your plan: Free and Plus cannot, Pro is read-and-fetch only, Business and Enterprise support read and write. Involve a developer and a security review, and test the rebuilt integration before switching users over.
I only use someone else’s GPT — do I need to migrate anything?
No, and you cannot: permission to use a GPT does not let you migrate it. Check the notice on the GPT and the creator’s guidance about a replacement — and confirm you can install and use the replacement before the original retires.
Is it worth migrating a simple GPT with no files or connections?
Often not as a plugin. A focused, instructions-only workflow may work better as a standalone skill — cheaper to share and maintain — or the instructions may just live on as a saved prompt. Migrate the workflows worth maintaining; retire the rest instead of moving busywork into a new container.
Shurah is the founder of AI Tools Daily, tracking pricing, licensing and policy changes across AI tools so readers can make decisions without wading through marketing claims themselves.
