OpenAI plans to retire Custom GPTs on 11 December 2026 and move them to plugins. If you built GPTs you rely on, list them now, publish any drafts, then migrate each one worth keeping when Migrate to plugin appears in My GPTs. Test the plugin against prompts you already trust, rebuild any custom actions, and share it again, because sharing settings don't carry over.
These steps follow OpenAI's Custom GPT retirement FAQ and its ChatGPT Learn migration guide, checked 30 September 2026. The status brief example is made up.
Which GPTs did you build, and which do you only use?
Open My GPTs and go down the list. For each one, note whether you created it or only use it, what job it does, who else relies on it, and whether it has custom actions. Only a GPT's creator (or, in an Enterprise workspace, an admin) can migrate it. If you only use someone else's, you don't need to migrate it: watch its notice for the replacement the creator shares.
Working across more than one account or workspace? Check where each GPT was created. That account's timeline and permissions apply, wherever you open it from. And be honest about the experiments. A GPT you haven't opened in months can just go.
Publish the drafts you still need
According to OpenAI's FAQ, migration takes the GPT's latest published version. Drafts and unpublished edits don't come across. Publishing isn't the same as sharing publicly, so a private GPT can stay private.
Enterprise workspaces have an extra deadline: OpenAI plans to stop new Custom GPT creation there on 26 October 2026. That stops new GPTs only, and you can still edit an existing one you haven't migrated. On other plans, follow the notice in your account. Either way, finish your edits before you migrate, because the original turns read-only once it's done.
Migrate when the option appears
When migration reaches your account, go to My GPTs, find the GPT under Created by me, and select Migrate to plugin. Review the details, then create the plugin. Timing varies by account. Enterprise workspaces had a target of 22 September 2026, and OpenAI says a target isn't a guarantee. Other plans have no firm date, so the notice in your account is the signal. A missing button before then isn't a fault.
Under the planned workflow, the GPT's instructions become a skill in the plugin, its knowledge files become reference files, and connected apps come over as apps. Custom actions, sharing settings and the selected model don't. The plugin starts private, your old conversations stay with the GPT, and the original becomes read-only. You can keep using it until retirement, but you can't delete it.
Example: Sam's list
Sam runs a programme office on a ChatGPT Enterprise workspace. She built a GPT called Weekly status brief, with instructions, two knowledge files and a connected Google Drive app. She also uses a public Meeting notes GPT someone else built, and has an unfinished draft called Board pack cutter. Weekly status brief is hers and she keeps it. Meeting notes isn't hers to migrate, so she waits for the creator. Board pack cutter gets published before 26 October, then migrated.
GPT name: Weekly status brief
Created by me: yes
Job: Monday status for the steering group
Who else uses it: two PMs on the same programme
Custom actions: none
Connected apps: Google Drive
Knowledge files: status template.docx, risk log excerpt.pdf
Familiar prompts to save: "Draft Monday status from the risk log and last meeting notes", "Rewrite the risks section for non-technical readers"
Hard case to save: "Produce the status when the risk log is a week stale and two workstream updates conflict"
Decision: publish pending edits, migrate when Migrate to plugin appears, share the plugin with both PMs, ask one to install it and re-run the familiar promptsTest it against prompts you already trust
Before you migrate, save a few prompts that already work well and one awkward case. Afterwards, run the same prompts through the plugin and compare. Check that the right skill gets picked, that it follows your instructions, that it uses the reference files you expect, and that the output comes back in the format you need. Your GPT's chosen model doesn't carry over, so expect some answers to read differently.
Test it two ways: select the plugin by name, then ask a normal question without naming it. ChatGPT can use an installed skill on its own when the description matches, but not on every request. Fix anything that changes your real work before you ask anyone else to switch. The checks in how to check AI output before using it at work apply here too.
Rebuild custom actions, then re-share
Custom actions don't migrate. If your GPT used one to reach another service, check whether an existing app covers the same job. If not, rebuilding it may mean a custom MCP server, which is technical work and depends on what your account allows. Test the rebuilt connection on its own before you retire the old workflow. If it isn't worth rebuilding, drop the GPT and do that job a simpler way.
Migration doesn't install the plugin for anyone else or keep your old sharing list. Share it with the people who need it, then ask one of them to install it and run a familiar prompt. In Enterprise, sharing with people and publishing to the workspace directory are separate permissions, so check yours with your admin before you announce the switch.
The dates to plan around are 11 December 2026 for standard retirement, or 11 February 2027 for Enterprise workspaces with an approved deferral. OpenAI calls these planned dates, so re-check OpenAI's FAQ before you lean on them. OpenAI's Learn guide says you can still migrate after retirement, but the GPT stops working on the day, so anyone who relies on it is stuck until the plugin is ready.
If your team is moving to Claude instead, the nearest equivalent is a skill: see package a weekly status playbook as a Claude Skill.
Sources
Custom GPT retirement and migration FAQ, OpenAI Help Center. Retirement dates (11 December 2026 standard; 11 February 2027 for Enterprise workspaces with an approved deferral), Enterprise milestones including the planned 26 October 2026 creation freeze, migrate path, what transfers, custom actions, sharing, and the read-only original. Primary source. Checked 30 September 2026.
Moving your custom GPT workflows to plugins, ChatGPT Learn. Written for Enterprise admins and GPT creators. Migrate, test and share steps, what transfers, custom actions needing a rebuild, testing with familiar prompts and a harder case, and migration after retirement. Primary source. Checked 30 September 2026.