How Do You Convert a Custom GPT or Gemini Gem Into a Skill Before They Retire?

OpenAI and Google are both retiring their custom assistants this fall. Here is how to move your work into a format you actually own.

By Robb Almy, Founder of AI Coach 4 All

Imagine a bookkeeper who spent a Saturday afternoon building a custom GPT to sort receipts into the right expense categories. It worked. She used it every week for months. Then one morning she opened ChatGPT and found a notice telling her that custom GPTs are being retired on December 11, 2026. Right underneath it sat a friendly button that said "Migrate to plugin."

One click. Problem solved, right?

Not quite. That button moves her work into a new container, but it is still OpenAI's container. And if she also built a Gemini Gem to draft client emails, Google just put that one on a clock too.

The quick answer is this. Before either deadline hits, copy out your instructions and files, open the Skill builder in your AI tool, hand it everything you already built, and test the result. A Skill is the portable format that Claude reads, that OpenAI's new plugins are built on, and that Google's new Gemini skills can import. Build the Skill first and you can run it almost anywhere, including inside a ChatGPT plugin if you still want one.

Key Takeaways

• OpenAI will retire custom GPTs on December 11, 2026, and after that date the GPTs and their pages become inaccessible.
• Google will begin automatically migrating Gemini Gems to skills on November 17, 2026, and reports suggest Gem editing may lock as early as October 13.
• OpenAI's "Migrate to plugin" button is fast, but custom Actions, sharing links, large files, and your model choice don't all make the trip.
• A plugin keeps your work inside OpenAI's ecosystem, while a Skill built on the open standard travels between Claude, ChatGPT, Gemini, and dozens of other tools.
• Converting takes an afternoon and no code. Gather your materials, hand them to a Skill builder, test side by side, and keep a copy you own.

What Is Actually Happening to Custom GPTs and Gemini Gems?

If you built a Custom GPT or a Gemini Gem sometime in the last two years, you did something smart. You took a repetitive task, whether that was drafting client emails, categorizing expenses, or answering the same customer questions, and you taught an AI tool to handle it the same way every time. That is exactly the right instinct.

The catch was always the container. A Custom GPT only runs inside ChatGPT. A Gem only runs inside Gemini. Neither one was built with the idea that you, the business owner, might someday want to switch tools or keep working after the platform changed the rules. This fall, both platforms changed the rules.

On the OpenAI side, the company announced on September 11 that it plans to retire custom GPTs across ChatGPT plans and offer a migration path to plugins. According to OpenAI's retirement and migration FAQ, custom GPTs are scheduled to retire on December 11, 2026. Enterprise workspaces with an approved deferral get until February 11, 2027. On the retirement date, custom GPTs and their GPT pages become inaccessible. Personal accounts (Free, Go, Plus, and Pro) already can't create or publish new GPTs, and anyone who tries to build one is now nudged to create a plugin instead.

On the Google side, the Gems manager inside Gemini now carries a notice that Google will begin automatically migrating Gems to skills starting November 17, 2026. You can keep using your Gems until they migrate. Code in the Gemini app, analyzed by Android Authority, points to Gem creation and editing shutting off on October 13, 2026, though that date comes from app code rather than an official Google announcement. There is also a catch. Skills used through Gemini Spark require a paid Google AI Pro or Ultra subscription, and skills currently require a personal Google account, with support for work and school accounts coming later. If your Gem lives inside a business Google Workspace account, as many small business owners' do, you are watching this one from the sidelines for now.

I run a good chunk of my own business on Skills, from blog drafting to bookkeeping to rental property accounting. So when these deadlines landed, my first question wasn't "which button do I click?" It was "where do I want this work to live for the next five years?"

What Does OpenAI's "Migrate to Plugin" Option Actually Do?

If you remember ChatGPT plugins from 2023, forget them. The 2026 version is a different thing. A plugin is now a package that can hold one or more skills, connections to outside tools, and reference material. According to OpenAI's developer documentation, plugins live in a single directory shared by ChatGPT and Codex, its coding agent.

When you click "Migrate to plugin," your GPT's instructions become a skill inside the new plugin. Your knowledge files are copied in as reference files, and connected apps can come along too.

Notice what sits at the center of that plugin. A skill. OpenAI's own plan for your GPT is to turn it into a Skill and then put that Skill in an OpenAI box. That tells you which part actually matters.

Why Might Migrating to a Plugin Be Harder Than It Looks?

For a simple GPT, the migration can be quick. A hands-on review in Visual Studio Magazine described the process as straightforward. But the same review, along with a growing pile of threads in OpenAI's own developer community, points to several places where small business owners can get stuck.

Custom Actions don't convert. If your GPT connected to outside software through Actions, such as a booking system, a CRM, or a spreadsheet, those connections aren't automatically converted. OpenAI says they may need to be rebuilt as an app or a custom MCP server. For most small business owners, "build an MCP server" is exactly where the do-it-yourself road ends.

Your files may not fit. Builders report that knowledge files their GPT accepted without complaint were rejected as too large during migration.

Your sharing link stops working the way it did. A migrated plugin starts private. People who used your GPT don't automatically get access, and sharing settings don't transfer. If you shared a GPT with clients or students, or put its link on your website, in a newsletter, or on a QR code, that audience is cut off until you rebuild access.

There is no easy undo. Once you migrate, the original GPT becomes read-only and future edits happen in the plugin. Your selected model doesn't carry over, and your past conversations stay behind.

It may not behave the same. OpenAI itself cautions that a migrated plugin may respond differently. And unlike opening a GPT on purpose, ChatGPT decides when a plugin applies, and OpenAI notes it won't necessarily run on every request. Some builders report weaker instruction-following after migrating.

It still lives in one house. Even a flawless migration leaves your work inside OpenAI's plugin directory. That is the same problem that got us here.

None of this makes the plugin button a bad choice. If you work entirely inside ChatGPT, it is the fastest bridge available. Just don't mistake a bridge for a foundation.

Why Is a Skill the Better Destination?

Skills would be just another walled garden if only one company supported them. The evidence says otherwise.

Anthropic introduced Skills in October 2025 as folders of instructions, scripts, and resources that Claude loads only when a task calls for them. In December 2025, Anthropic published Agent Skills as an open standard built for portability across platforms. Since then, the rest of the industry has followed. OpenAI's own developer documentation now states that its skills build on the open agent skills standard. Google shipped an official library of Agent Skills on the same format at its Cloud Next conference in April 2026, and Google's new Gemini skills can be created from SKILL.md files, the same file at the heart of every Skill.

By March 2026, industry trackers counted roughly 32 separate AI and coding tools that could read the same Skill format, including Codex CLI, Gemini CLI, Cursor, JetBrains Junie, AWS Kiro, and Block Goose. The same comparison found a practical difference that matters for how much you can teach your assistant. GPT instructions cap out around 8,000 characters and Gems around 4,000, while Skills carry no meaningful limit. If your instructions ever felt cramped, that ceiling was built into the format.

I've seen this firsthand. When I moved my own academic note-polishing assistant out of a GPT, I didn't start from scratch. I handed Claude the five knowledge documents that had powered the GPT, and it became a working Skill I used the same day.

The Plugin Is the Box. The Skill Is What's Inside.

Here is the shift in thinking that matters most. A Custom GPT or a Gem is really a set of instructions borrowed by a platform. You wrote them, but the platform decides where they live and how long they last. A Skill flips that relationship. The instructions, the knowledge files, the formatting rules all belong to you as a portable package that happens to run inside whichever AI tool you choose that day.

Think of it like storage. Your GPT has been sitting in a rented unit inside OpenAI's warehouse. The "Migrate to plugin" button moves your things into a newer unit in the same warehouse. Google's automatic migration does the same thing across town.

A Skill is a moving box. Pack it once, and it goes wherever you go, including into either warehouse if you decide that's where you want to work.

How Do You Convert a Custom GPT or Gemini Gem Into a Skill?

1. Copy everything out before the deadline. Open your GPT's editor, copy the full instructions into a document, and download every knowledge file. For Gems, do this before October 13 in case editing locks. OpenAI's own advice is to record your instructions, reference files, integrations, and a few familiar prompts so you can test the replacement.

2. Save three to five test prompts. Pick the requests you make most often and save the outputs you liked. These become your answer key when you test the new version.

3. Write one sentence about what it does and when to use it. This becomes the Skill's description, which is the part an AI tool reads to decide when to pull the Skill in. "Sorts receipts into my expense categories whenever I upload a receipt or invoice" beats "helps with bookkeeping."

4. Hand everything to a Skill builder. In Claude, start a new chat and say, "I want to turn my Custom GPT into a Skill." Paste your instructions, attach your files, and share your one-sentence description. Claude will organize it all into a Skill folder with the instructions and reference files in the right places.

5. Add back what you had to cut. Remember the examples and edge cases you trimmed to squeeze under the character limit? Put them back. Examples of great output and a short list of things the assistant should never do make a noticeable difference.

6. Handle outside connections separately. If your GPT used Actions, list every app it connected to. Many popular tools already offer ready-made connectors inside Claude and other AI tools. If yours doesn't, that one piece is worth getting help with.

7. Test side by side, then keep your own copy. Run your saved prompts through the old and new versions and adjust until the Skill matches or beats the original. Then download the finished Skill folder and store it somewhere you control. That folder is the asset.

Frequently Asked Questions

Do I need to know how to code to convert a GPT or Gem into a Skill?
No. You gather the instructions and files you already have, describe what the assistant should do in plain language, and your AI tool formats everything into a working Skill through a normal conversation. The only piece that may need technical help is a GPT that used custom Actions to connect to outside software.

Should I just click "Migrate to plugin" in ChatGPT?
It works as a quick bridge if you live entirely in ChatGPT, but copy your instructions and files first. After migration the original GPT becomes read-only, sharing links don't carry over, and custom Actions must be rebuilt. Building a standalone Skill first gives you a portable copy you own either way.

What happens to my Gemini Gems on November 17?
Google says it will begin automatically migrating Gems to skills on November 17, 2026, and you can keep using them until they migrate. Reports based on app code suggest editing may lock on October 13. Skills used through Gemini Spark require a paid Google AI Pro or Ultra subscription.

Will a Skill I build in Claude also work in ChatGPT and Gemini?
The format is shared. OpenAI's skills build on the open agent skills standard, and Google lets you create skills from SKILL.md files. Availability depends on your plan and account type, so check first. OpenAI currently lists ChatGPT skills for eligible Business, Enterprise, Healthcare, and Edu workspaces.

Can I keep using my Custom GPT while I build the Skill?
Yes, until December 11, 2026, when custom GPTs and their pages become inaccessible. Running both side by side is the smartest approach. Compare the outputs, adjust the Skill until it matches or beats the original, and only then stop relying on the GPT.

The Bottom Line

Go back to that bookkeeper staring at the "Migrate to plugin" button. She could click it and move on. Her GPT would keep living in OpenAI's world a little longer, until the next time the rules change.

Or she could spend one Saturday afternoon, the same kind of afternoon that built the GPT in the first place, turning it into something she owns. Every hour she spent writing instructions and testing edge cases comes with her. None of it has to be rebuilt the next time a platform reshuffles its lineup.

You have the same choice, and a calendar to go with it. October 13 for Gem edits. November 17 for Gem migration. December 11 for GPTs. Pull up your instructions today. Gather your files. Hand them to the AI tool of your choice and let it build you a Skill.

Platforms will keep changing the rules. Build the thing that doesn't care.

About the Author

Robb Almy is the founder of AI Coach 4 All, where he teaches AI and helps businesses navigate the complexities of the AI landscape. He helps businesses translate data into decisions and adopt AI responsibly. Robb helps small business owners cut through AI hype and adopt the tools that actually save time and spawn creative workflows that make a difference.

A note from Robb. I use AI tools, primarily Claude, ChatGPT and Gemini, to research, outline, and draft my content. I provide the source material such as personal stories, frameworks, business experience, and the judgment about what will actually help you. AI creates the first draft based on my own context and voice. I carefully review and edit everything before it reaches you.