Platform change guide

    Custom GPT Migration: Moving GPTs to ChatGPT Plugins and Gems to Gemini Skills

    OpenAI is retiring custom GPTs and Google is replacing Gems. The dates, what carries over, a checklist for each platform and how to rebuild custom actions.

    By Chad Westleigh · Updated

    Book a migration sprint See AI Workspace Setup

    What's changing: custom GPTs retire and Gems become skills

    OpenAI and Google are both replacing their custom assistants with skills: packaged instructions in the same SKILL.md-based format Anthropic uses for Claude. The OpenAI Help Center says custom GPTs will retire and can be migrated to ChatGPT plugins, which bundle skills, reference files and connected apps. Google's Workspace Updates blog (September 30, 2026) says skills will replace Gems, with business accounts losing Gems from no sooner than March 1, 2027.

    Each custom GPT or Gem your team built needs a decision before its cutoff: migrate it, rebuild it or retire it. The migration itself is rarely hard, but it adds up, and GPTs that call outside systems through custom actions are the part that breaks. If you'd rather hand it off, our AI Workspace Setup service rebuilds assistants as plugins and skills.

    Custom GPT and Gems retirement timeline

    From OpenAI's Help Center and Google's Workspace Updates and Gems migration pages, as of October 16, 2026. Dates can change; follow the notice in your own account.

    DatePlatformWhat happens
    Sep 22, 2026ChatGPT EnterpriseTarget date for the migration experience (a target; not every workspace gets it that day)
    Sep 30, 2026GeminiGoogle announces that skills will replace Gems
    Oct 13, 2026GeminiSkills start rolling out in the Gemini app; the Workspace apps launch is delayed
    Oct 26, 2026 (planned)ChatGPT EnterpriseCreation of new custom GPTs ends
    Nov 17, 2026GeminiGems move to the Settings panel and still work
    Dec 11, 2026ChatGPTStandard retirement: custom GPTs and their pages become inaccessible
    Feb 11, 2027ChatGPT EnterpriseRetirement for workspaces with an approved deferral only
    No sooner than Mar 1, 2027Gemini business and enterpriseGems can no longer be created, edited or used; Ask a Gem steps in Workspace Studio flows stop working
    No sooner than Jun 1, 2027Gemini educationThe same cutoff for education editions

    What carries over in a custom GPT migration, and what doesn't

    OpenAI and Google describe their planned migration workflows differently. Review every migrated assistant before anyone relies on it.

    ItemChatGPT: custom GPT to pluginGemini: Gem to skill
    InstructionsBecome a skill inside the pluginCarry over, with the Gem's name and description
    Knowledge filesCopied into the plugin's reference filesRetained
    Connected appsAdded as apps; each still needs its own authorizationDrive permissions and sharing settings transfer
    Custom actionsDo not transfer; rebuild with an app or a custom MCP serverNot applicable; flows using Ask a Gem must be rebuilt
    Selected modelDoes not carry overNot addressed in Google's guide
    ConversationsNot migrated; OpenAI says they stay accessible after retirementNot addressed in Google's guide
    Sharing and statusPlugin starts private; public listing needs a separate submissionSkills arrive as inactive drafts the creator must activate
    The originalRead-only once migrated, usable until retirementUsable until your edition's cutoff

    ChatGPT migration checklist: custom GPT to plugin

    Run these steps for every custom GPT your company created. If you only use someone else's GPT, check the creator's guidance about a replacement instead.

    1. 1. Inventory and triage

      List each GPT, its creator, the workspace it was created in and who relies on it. Retire unused GPTs, merge near-duplicates and flag any with custom actions.

    2. 2. Publish final edits

      Migration uses the latest published version, and drafts don't transfer. Publishing doesn't require public sharing.

    3. 3. Save a test set

      Keep five to ten familiar prompts and one hard case per GPT, with the output you expect.

    4. 4. Run Migrate to plugin

      When migration is available, open My GPTs and select Migrate to plugin. Only the creator, or a workspace admin in Enterprise, can migrate a GPT.

    5. 5. Review the skill and files

      Check the instructions that became a skill, the reference files and the connected apps, and test on the model your team will actually use.

    6. 6. Rebuild custom actions

      Replace each action with an available app or a custom MCP server, and test it separately.

    7. 7. Share and switch people over

      Share the plugin or publish it to your workspace directory, set an installation policy and confirm a real user can install and run it.

    Gemini migration checklist: Gems to skills

    Google's Gems migration guide recommends rebuilding critical Gems as skills early rather than waiting for automated migration.

    1. 1. Turn skills on

      A Google Workspace admin confirms skills are enabled and shares the dates for your edition.

    2. 2. Inventory Gems and flows

      List each Gem, its owner and knowledge files, plus every Workspace Studio flow with an Ask a Gem step.

    3. 3. Rebuild critical Gems

      Copy the instructions into a new skill, re-upload the knowledge files and compare results on your saved prompts.

    4. 4. Recreate skills where people work

      Skills don't yet sync between the Gemini app and Workspace apps, so recreate Workspace skills in Google Workspace Studio.

    5. 5. Activate migrated drafts

      Gems left for automated migration arrive as inactive drafts that each creator reviews and activates.

    Rebuilding Custom Actions as MCP servers

    Custom actions are the part of a custom GPT that stops working if nobody rebuilds it. OpenAI's FAQ says custom actions don't transfer, and rebuilding a connection may require a custom MCP server. The Model Context Protocol (MCP) is the open standard ChatGPT, Claude and many other AI tools use to call outside systems.

    How we rebuild an action:

    • Look for an existing app first. Use a vendor's official app or MCP server if it covers the job with scopes you can limit.
    • Turn endpoints into task-level tools, such as look up an order or open a ticket.
    • Mark read-only tools. OpenAI's documentation says ChatGPT treats any MCP tool without a read-only hint as a write action, which requires confirmation by default.
    • Use OAuth and least privilege, not an old action's API key copied across unexamined.
    • Check permissions. In ChatGPT Enterprise, creating plugins that use MCP servers is a separate permission, off by default.

    Claude supports custom connectors through remote MCP on every plan, so one rebuilt server can serve both assistants. Our MCP integration service builds these servers with logging and review gates.

    Make migrated assets portable across ChatGPT, Claude and Gemini

    ChatGPT, Claude and Gemini now all build skills on the SKILL.md format, so one library can serve all three. Anthropic published Agent Skills as an open standard in December 2025. OpenAI's developer documentation says its skills build on that standard. Google says Gemini skills use SKILL.md, so skills from other platforms can be copied in.

    A portable skill is a folder:

    • SKILL.md: a name and a description of what the skill does and when to use it, then plain-Markdown steps, rules and examples
    • Reference files such as price lists and templates, each with an owner
    • Platform extras kept separate, such as OpenAI's optional agents/openai.yaml file or scripts only some platforms run

    Write the description carefully, because OpenAI notes ChatGPT may use an installed skill automatically when its description matches a request. Keep system access in MCP servers, not in instructions. See how to build Claude Skills, and Claude vs ChatGPT for business to decide which assistant each team should use.

    Testing after migration

    OpenAI warns that a migrated plugin may respond differently. Against your saved test set, check that each plugin or skill:

    • Gets selected for the right requests, with and without an @ mention
    • Follows the original instructions, tone and output format
    • Uses the expected reference files and says when an answer isn't there
    • Runs every connected app or rebuilt tool, with approval on write actions
    • Handles your hard case as well as the original did
    • Works for a second person from the intended audience

    Governance for shared plugins and skills

    Shared plugins need an owner, permissions and a review rhythm, or they sprawl the way custom GPTs did. ChatGPT workspace admins control plugins through separate permissions (Use, Upload, Create with MCPs, Share and Publish to workspace) and an installation policy that makes a plugin available or installs it automatically. Apps inside a plugin keep their own authorization and approval rules.

    For every migrated library we set up:

    • One named owner per plugin or skill, plus a short changelog
    • Least-privilege app access, with writes behind approval
    • Vendor review before install; OpenAI says its Verified badge doesn't replace your own security review
    • A quarterly review that retires anything unused
    • Data rules tied to your AI acceptable use policy

    In Gemini, sharing settings transfer with migrated Gems, so check who can see each skill before activating it. Creators learn to maintain all of this in our ChatGPT and Gemini training.

    The Custom GPT & Gems migration sprint

    Operator Intelligence (OI) runs a fixed-scope migration sprint, scoped after a short inventory call and confirmed in a written proposal.

    Inventory and triage

    Every GPT and Gem listed with its owner, users and custom actions, and a decision to migrate, merge, rebuild or retire.

    Migrate and test

    Migrations run in your own accounts, with each skill and reference file reviewed and tested on real tasks.

    Actions rebuilt as MCP servers

    Apps or custom MCP servers with scoped OAuth access, logging and approval on write tools.

    Learn more

    Portable library and governance

    SKILL.md-style assets that move between ChatGPT, Claude and Gemini, with owners and permissions set.

    Learn more

    Team handover

    Hands-on training so creators can maintain their plugins and skills.

    Learn more

    Frequently asked questions

    When do custom GPTs stop working?

    The OpenAI Help Center lists December 11, 2026 as the scheduled retirement date, or February 11, 2027 for Enterprise workspaces with a qualified, approved deferral. After retirement, custom GPTs and their pages become inaccessible.

    What happens to GPTs that use custom actions?

    Custom actions don't transfer, so features that depend on them won't work in the plugin until you replace them. Check for an available app first; otherwise the connection usually becomes a custom MCP server. Start these first.

    Can we move our custom GPTs to Claude instead?

    Yes. Instructions and knowledge files become a Claude Skill or Project, and a rebuilt MCP server connects to Claude as a custom connector. Our Claude for business page covers that rollout.

    When do Gemini Gems stop working?

    On Google Workspace business and enterprise accounts, Gems can no longer be created, edited or used from no sooner than March 1, 2027; education editions follow no sooner than June 1, 2027. They keep working until then.

    What does a migration sprint cost?

    Cost depends on how many GPTs and Gems you have, how many use custom actions or need MCP servers, and how much governance and training you want. We don't publish fixed prices; see how we price.

    Get your custom GPTs and Gems migrated before the cutoff

    Send us the GPTs and Gems your team relies on. We'll say what to migrate, rebuild or retire, and scope a fixed sprint around your deadline.

    Book a migration sprint

    Sources

    1. OpenAI Help Center, Custom GPT retirement and migration FAQ (2026): Custom GPTs scheduled to retire Dec 11, 2026; Feb 11, 2027 for Enterprise workspaces with a qualified, approved deferral; transition affects all ChatGPT plans, migration and plugin access may differ by account or workspace; after retirement GPTs and GPT pages become inaccessible; Enterprise milestones Sep 22, 2026 (migration target) and Oct 26, 2026 (planned end of new GPT creation); migration via My GPTs > Migrate to plugin using the latest published version (drafts don't transfer); instructions become a skill, connected apps added as apps, knowledge files copied to reference files; selected model does not carry over; custom actions do not transfer and rebuilding may require a custom MCP server; existing conversations remain accessible after retirement; original becomes read-only after migration; sharing settings don't carry over, plugin starts private, public listing requires a separate submission; only the creator or an Enterprise workspace admin can migrate; migrated plugin may respond differently, with a testing checklist
    2. OpenAI Help Center, Plugins in ChatGPT (2026): A plugin can include skills, connected apps, app templates and extensions; workspace permissions Use plugins, Upload plugins, Create plugins with MCPs (off by default for Enterprise), Share plugins, Publish plugins to workspace; installation policies Available and Installed; apps keep their own controls including authorization and approval requirements; OpenAI Verified badge does not replace your organization's privacy, security or vendor review; ChatGPT uses installed plugins automatically when relevant
    3. OpenAI, Build skills (ChatGPT Learn developer docs) (2026): A skill is a directory with a SKILL.md file plus optional scripts and references; SKILL.md must include name and description; optional agents/openai.yaml; skills build on the open agent skills standard
    4. OpenAI Developers, Custom MCP server guide (2026): Custom MCP servers are added through ChatGPT Plugins and created as a plugin; tools without the readOnlyHint annotation are treated as write actions, which require confirmation by default; OAuth supported; workspace permissions apply
    5. Google Workspace Updates blog (2026): Published Sep 30, 2026: skills will replace Gems; skills are built on the open SKILL.md standard, so skills created on other platforms can be copied in; Nov 17, 2026 Gems move to the Settings panel and can still be created, edited and used; no sooner than Mar 1, 2027 (business and enterprise) Gems can no longer be created, edited or used and Workspace Studio flows with Ask a Gem stop working; remaining Gems auto-migrate to draft skills; education no sooner than Jun 1, 2027; Oct 13, 2026 update: skills rolling out in the Gemini app, launch to Workspace Studio and Workspace apps delayed
    6. Google Workspace, Gems migration guide (2026): Converted skills retain names, descriptions, custom instructions and uploaded knowledge files; Drive permissions and sharing settings transfer; migrated Gems are inactive drafts creators must review and activate; skills don't sync between the Gemini app and Workspace apps and must be recreated manually in Google Workspace Studio; admins should make sure skills are turned on and encourage rebuilding critical Gems early
    7. Anthropic (claude.com blog) (2025): Dec 18, 2025: Anthropic published Agent Skills as an open standard for cross-platform portability
    8. Claude Help Center (2026): Custom connectors using remote MCP are available on Free, Pro, Max, Team and Enterprise plans

    Get the Operator Intelligence briefing

    A short, practical note every two weeks on what is working with AI in real businesses. No spam; unsubscribe anytime.

    Related

    AI Workspace Setup

    Projects, plugins and skills built for your team.

    ChatGPT and Gemini training

    Training that ends with configured assistants.

    MCP integrations

    Custom MCP servers with review gates.

    Claude Skills explained

    How SKILL.md-based skills work.

    Claude vs ChatGPT for business

    Which assistant fits which team.