Custom GPTs: Retirement Dates and What to Move Where

Custom GPTs, versions of OpenAI ChatGPT configured for a specific job, are scheduled to retire on December 11, 2026, with a later date for specified workspaces. If your company depends on them, preserve the instructions and source files now, then choose the replacement by the job: a plugin for a reusable method, a Project for ongoing shared context, or a rebuilt integration for custom actions. Migration is only finished when the people doing the work can use the replacement.

Verified against OpenAI's live documentation on October 30, 2026. OpenAI's retirement FAQ now states a standard retirement date and named exceptions. Your account or workspace notice governs any different timeline communicated to you.

Custom GPTs Retirement: What Changed by Plan

OpenAI is retiring the custom GPT format across ChatGPT plans. The important distinction is between creating a GPT, editing an existing one, getting access to migration, and the date the GPT stops being available. Those are separate events.

Sources: OpenAI's creation and editing guidance and retirement schedule. The general retirement date is announced; a separate migration-availability date for every plan is not. Where no separate date is announced, do not substitute the Enterprise rollout target.

Creating Custom GPTs Is Separate From Keeping Existing Ones

A missing Create button does not mean an existing GPT has already stopped working. OpenAI permits continued use until the applicable retirement date, subject to access rules. Once migrated, however, the original becomes read-only. Finish essential edits first. OpenAI's current editing rules.

The Enterprise milestones include an admin notice on September 11, 2026, a migration-experience target of September 22, and an October 1 banner target for workspaces that had already opted into delayed banners. These were rollout targets, not promises that every account could migrate on those days. OpenAI's Enterprise timeline.

Check where each GPT was created. A public GPT built in an affected Enterprise workspace follows that workspace's transition even when your colleague opens it from a personal account. Recording only the user's subscription will miss that dependency.

Custom GPTs for Business: Choose by the Work You Need to Preserve

Keep the useful method, then choose its next home. Do not make the whole inventory follow one replacement just because every item currently has a GPT link.

My recommendations for common workplace workflows:

• A proposal reviewer used across unrelated opportunities: choose a plugin. Its value is a reusable review method and approved reference material.
• A client account assistant with changing briefs, meeting notes and drafts: choose a Project. Its value is the accumulating context of that particular account.
• An operations GPT that changes records in another system: inventory the custom action first. Choose an available app only if it performs the required operation; otherwise plan an integration rebuild.
• A company assistant intended to live inside your own product: evaluate an API implementation. OpenAI says GPTs are for use inside ChatGPT, and directs embedded-product builders to the API.

The underlying capabilities are documented in OpenAI's plugins guide, Projects guide and GPTs guide.

Think of a plugin as a portable instruction manual and a Project as a workroom. The manual tells you how to review a proposal. The workroom holds the changing evidence about one client. A connection that actually updates a business system is another component to account for.

For a team lead, prioritize GPTs whose failure would interrupt a deliverable or business operation. For a builder, prioritize integrations and behavior checks. For a workspace owner, prioritize the inventory and replacement access. Someone who only uses another person's GPT needs the creator's replacement guidance, not a copy of the migration instructions.

Move a Reusable Method to a ChatGPT Plugin

ChatGPT plugins can package reusable instructions, reference material and connected apps. A skill is the instruction component; a plugin can contain skills without requiring an external app connection. OpenAI's ChatGPT plugins page shows the product, while the setup guide explains its controls.

For the proposal-review example, preserve the review criteria and approved examples. Keep client-specific evidence out of the general method unless it belongs in every review. This gives the replacement a clear job instead of reproducing an assistant that accumulated unrelated duties.

Follow OpenAI's Migration Workflow When It Is Available

Use the account or workspace that owns the GPT. These steps follow OpenAI's migration FAQ and editor instructions:

1. 1

   Finish the version you want to preserve

   Open Explore GPTs, then My GPTs, choose the GPT and select Edit GPT where available. Apply essential changes with Update. Migration takes the latest published version; unpublished edits stay behind. Publishing does not require public sharing.

2. 2

   Run the available migration

   In My GPTs, select Migrate to plugin and review the migration details. If the option is absent, check the owning workspace and its notice, then ask the administrator to check plugin access. Do not assume a rollout target guarantees the control is present.

3. 3

   Inspect what transferred

   Instructions become a skill, knowledge files become reference files, and connected apps are added as apps. The selected model does not transfer; Enterprise defaults apply. Custom actions need separate replacement work.

4. 4

   Test the method and the integration

   Compare familiar prompts and a difficult case. Check whether the correct instructions and sources are used, whether the deliverable has the required format, and whether connected operations work. A plausible written answer is insufficient evidence that a record was updated correctly.

5. 5

   Share, install and check the recipient experience

   An individually migrated plugin starts private. Where available, use Share Plugin to prefill the original audience, review it and select Save. Ask an intended recipient to install and try the replacement. Existing GPT access does not automatically become plugin access.

For larger Enterprise inventories, OpenAI now documents bulk migration through the ChatGPT Admin plugin. Retrieving the full workspace inventory requires a workspace Owner. Bulk work is asynchronous, so review the results before treating it as complete. This is distinct from individual migration by an eligible creator or administrator. OpenAI's bulk-migration guidance.

For a new reusable workflow, OpenAI directs users to install Plugin Creator, start a conversation with @plugin-creator, and describe what to build. Use the documented creation path where available, then test it. Plugin Creator setup.

Projects vs Custom GPTs: Move the Ongoing Context

ChatGPT Projects fit work that keeps accumulating evidence. For a client account, that means the approved brief, current files and continuing discussions stay together. For a recurring reporting setup, follow the ChatGPT Projects guide.

To recreate that workflow, select New project in the sidebar, name it, upload the relevant source files, and add the working instructions through Project settings in the more options menu. Project instructions apply inside that Project and override global custom instructions. These are OpenAI's documented Project setup steps, not a GPT-to-Project migration tool.

For a client-report GPT, rewrite the instructions around the account: which evidence is authoritative, what the report must contain, and which decisions require review. Preserve the useful rules without assuming every line from the old assistant belongs in the new workroom.

GPT conversations cannot be moved into a Project. OpenAI distinguishes them from eligible chats that offer Move to project. Start a new Project chat and add the relevant source material. Existing GPT conversations remain accessible after retirement, but that does not make them imported Project history. Project chat restrictions, conversation-retention guidance.

Do Custom GPTs Have Memory?

Custom GPTs do not use saved memory or previous conversations; each conversation starts fresh. Projects can use their chats and files as context for later work. That distinction is why an evolving account file often belongs in a Project. GPT memory behavior, Project context.

Are Custom GPTs Private?

A GPT's sharing scope depends on its settings, and its builder cannot view users' individual conversations. A shared Project has a different collaboration model: members can see its chats and files, and download its files. Review the audience before placing old working material in it. GPT privacy, shared Project visibility.

For example, a reusable staff-policy assistant and a manager's discussion of an individual case should not become one shared workroom by default. Choose the access boundary before copying the material.

Custom Actions Need an Integration Decision

A GPT that updates a business system needs more than its prompt preserved. Custom actions do not transfer through the migration workflow. OpenAI recommends checking for a suitable app; a replacement may require a custom MCP server, a connection that exposes tools to the assistant. Available integration options depend on the account and workspace. OpenAI's action-migration guidance.

For an order-management assistant, write down the operation it performs: reading an order, changing its address, or submitting a cancellation are different requirements. Then check each required operation against the replacement. An app that can retrieve an order is not proof that it can change one.

Assign this GPT to a technical owner early. Have that owner verify the destination record, allowed operations and failure behavior. Keep a human-operated version of the process available until the replacement passes those checks.

An assistant inside your own website or service is a separate build decision. OpenAI's GPTs documentation directs that use case to the API. Do not promise stakeholders that a ChatGPT migration will also deliver an embedded product.

The File Limits That Can Change Your Choice

Check the source bundle before choosing a Project. A GPT can contain up to 20 knowledge files, while Project allowances vary by plan. GPT file limit.

OpenAI also limits uploads to 10 files at once. These are Project limits, not plugin-migration limits. OpenAI's Project allowances.

A hypothetical GPT using all 20 file slots would exceed a Free Project's allowance by 15 files. That is arithmetic from the documented limits, not a migration result. Check which files are still needed; if the working source bundle does not fit, choose a destination that does instead of silently dropping evidence.

What's Overhyped: Migration Does Not Guarantee Continuity

The announcement is real. The claim that every replacement will behave identically is not supported by OpenAI's guidance.

A report is not the current timetable. PCWorld reported the retirement on September 18, 2026, describing the then-Enterprise schedule and suggesting Projects. OpenAI's current pages now give the standard retirement date, specific exceptions and updated creation rules. Use those live official rules for planning.

Preserved instructions do not mean preserved behavior. Model selection, available tools and the way a skill is invoked can affect the result. Judge the replacement against the work you need completed, not whether it produces familiar-sounding prose. OpenAI's migration testing guidance.

Available across plans does not mean every plugin is available to everyone. Installation and use depend on plan, workspace, role, region and included capabilities. Check the actual replacement with an intended user. This availability statement does not establish that plugins are universally free or paid. OpenAI's plugin availability rules.

What to Do This Week

Produce a replacement list that someone can execute. The priority is preserving the work and assigning an owner before switching users.

• List your GPTs. Record the owner, originating workspace, purpose, users, notice deadline and any apps or custom actions. Include public GPTs your staff rely on but do not own.
• Export their instructions and files into your own working archive. For GPTs you can edit, copy the instruction text and conversation starters, and gather the original knowledge files from their maintained source locations. Make this a manual backup of the configuration and maintained originals. Use the documented GPT editor to review the configuration, and resolve missing source files with their owner.
• Pick the replacement for each. Mark plugin, Project, integration rebuild, API build, or no replacement for an unused workflow. Give every active item an accountable owner and a representative task to verify.
• Complete a recipient check. Before announcing the switch, have someone other than the builder access the replacement and complete the representative task. Follow OpenAI's migration and sharing guidance.

Keep using an existing GPT while its replacement is being prepared if the applicable timeline and permissions allow it. Wait to switch an untested integration; do not wait to inventory it. Someone with no GPT dependencies has no GPT workflow to migrate.

Common Questions

What are custom GPTs?

They are versions of OpenAI ChatGPT configured for a purpose, combining instructions, reference knowledge and selected capabilities. A company might use one to apply its proposal-review rules. OpenAI's GPTs guide.

Are custom GPTs still a thing?

Existing GPTs remain usable before their applicable retirement date, subject to access rules. New creation and continued use have different rules; check the plan table above and your workspace notice. OpenAI's current status.

Can I use custom GPTs for free?

OpenAI says signed-in users on all ChatGPT plans can use GPTs they have access to. That includes Free users. Personal accounts cannot create or publish new GPTs, and plugin availability is a separate question. GPT availability.

Is OpenAI phasing out custom GPTs?

Yes. OpenAI's current FAQ sets standard retirement for December 11, 2026, and February 11, 2027 for FedRAMP and qualifying approved deferrals. Follow a different account or workspace timeline if OpenAI communicates one to you. Official retirement FAQ.

Are custom GPTs gone?

No, as of the October 30, 2026 verification. At retirement, the GPTs and their pages become inaccessible, while existing conversations remain accessible. A missing creation button is a separate restriction. OpenAI's retirement guidance.

For practical guidance on keeping business AI workflows useful as products change, join the newsletter.