How to save your custom GPTs and Gemini Gems before they are retired

Nothing you built is gone yet, and you have time. If you made a Gem, Google says it will turn it into a skill for you automatically when Gems are switched off for personal accounts in November 2026. If you made a custom GPT, the move to OpenAI's plugins is one you start yourself, and OpenAI's help centre gives 11 December 2026 as the scheduled retirement date. Either way, spend a few minutes now saving your instructions and files somewhere you control.

We checked everything below on OpenAI's and Google's own help pages on 2 October 2026. Both say dates can shift, so treat them as today's plan.

What is actually happening

Custom GPTs. OpenAI is retiring custom GPTs and recommends moving them to plugins, which bundle reusable instructions (called skills), reference files and connected apps. Its retirement FAQ says custom GPTs "are scheduled to retire on Dec 11, 2026", that the change affects all ChatGPT plans, and that Enterprise workspaces with an approved deferral get until 11 February 2027. It also tells you to follow the notice for your own account, "including any different timeline communicated to you". Personal accounts (Free, Go, Plus and Pro) already cannot create new GPTs, though you can still edit the ones you made.

Gemini Gems. Google is replacing Gems with skills: saved instructions you call up by typing "/" and the skill's name, or that Gemini applies by itself when it judges one relevant to your request. Google's help centre says support for Gems ends in November 2026 for personal Google accounts, March 2027 for Workspace business, enterprise and non-profit accounts, and June 2027 for education accounts. That page gives no day. Google's Opal FAQ puts 17 November on the switch-off for Gems made with Google Labs, and TechCrunch reports that the notice inside the Gemini app gives the same date for ordinary Gems.

Step 1: save a copy of each one today

Do this even if you plan to accept the automatic conversion: a copy on your own computer survives a migration that goes wrong.

For a custom GPT, work on the web, which is where OpenAI says editing happens:

Open Explore GPTs, then My GPTs, choose the GPT and select Edit GPT.
In the editor, copy the name, description, instructions and conversation starters into a document.
Note which capabilities were switched on, such as web search or image generation, and which model it recommended.
Gather the knowledge files. The OpenAI help pages we read explain uploading them but not downloading them again, so find your original copies now, while you remember where they live.

For a Gem, Google gives the steps on its help page:

On gemini.google.com, open the sidebar, go to Gems and click Edit next to the Gem.
Copy the name, description and instructions.
Under Knowledge, open each file and click Download.

Also save two or three favourite prompts, each with an answer you liked. You will want them to test whatever replaces the original.

Step 2: know what will not come across

On the OpenAI side, the FAQ says the instructions become a skill inside the new plugin, knowledge files are copied in as reference files, and connected apps come along. These do not:

Custom actions. The connections a GPT made to outside services "do not transfer through the migration workflow", and rebuilding one may need technical setup.
The chosen model. "The GPT's selected model does not carry over."
Sharing. The plugin starts private, and your GPT's users do not get access automatically.

Before you press migrate. Migration uses the last published version of your GPT, so publish any edits you want to keep first. Publishing does not mean sharing it publicly. After migration the original turns read only until it retires, so make your final changes first.

Your past conversations with a GPT stay in your history. OpenAI says they "will remain accessible after retirement", so you do not need to export them, although they do not move into the plugin.

If you use a GPT someone else built, you cannot migrate it yourself. Watch for a notice on it, or word from its creator about a replacement.

On the Google side, Gems convert along with their "supported files". Google lists what does not work in skills for now: most of the default tools that Gems could use, including Create video, Create music, Canvas, Deep research and Guided learning. Files from GitHub are not supported. Gems made with Google Labs, the mini-app kind, switch off along with Opal, and Google says Opal workflows will not migrate automatically, so copy their prompts by hand. Sharing skills with other people, adding files from Google Drive and adding notebooks are promised "in the coming weeks".

Step 3: choose where your assistant lives next

Stay where you are. If the assistant is mostly instructions plus a few documents, the official route will probably serve you well. For a GPT, go to My GPTs and select Migrate to plugin once the option appears. For a Gem, you can wait for the automatic switch or rebuild it as a skill now through Settings, then Skills, then Create manually. Adding files to a hand-made skill takes extra steps, and Google's help page walks through them. Google says skills are available to people over 18 on personal Google accounts.

On a personal ChatGPT plan? OpenAI says migration and plugin access may differ by account. Its help page on skills, checked on 2 October 2026, lists them for Business, Enterprise, Healthcare and Edu users and does not mention Free, Go, Plus or Pro. If no Migrate to plugin option shows up for you, the copy you saved in Step 1 is what carries your GPT forward.

Then test it. OpenAI warns that a migrated plugin "may respond differently", and a skill deserves the same check, since Gemini can pick one up on its own. Run your saved prompts and compare the answers with the old ones.

Rebuild somewhere else. Your saved instructions and files will work in any app where a model can follow standing instructions and read reference files.

Rebuilding in MultiChats: what works and what does not

MultiChats has no import for GPTs or Gems, so this is a rebuild by hand. On the Pro and Pro+ plans, a folder can hold both pieces a custom assistant is made of: its instructions and its files.

Create a folder named after your assistant.
Open Folder memories inside it, add a memory, paste in your instructions (a memory holds up to 10,000 characters) and select it. Every new chat you start in that folder gets it automatically. Chats that were already there do not.
Add your knowledge files to the folder's Files section. Every chat in the folder can use them, so you never attach them again. Pro allows 20 files per folder and Pro+ allows 100, and the folder shows how many words of room are left.
Start a chat in the folder and choose a model. You can switch models partway through, so try your saved test prompts on a few.

Before you start, know the limits. A folder has no instruction box of its own, so instructions live in a memory or in a file. A folder cannot be shared with other people, which makes this a fit for assistants you built for yourself. There are no actions, so a GPT that fetched or posted things through another service will not do that here. And memories and folder files are paid features: on the free plan you can create folders, but you cannot use either one.

If your instructions need tightening on the way, our guide to writing better prompts shows how to make them specific enough to behave consistently.

The dates to keep

Now: save the instructions, starters and files of every GPT and Gem you rely on.
Before mid-November: decide about each Gem. Leave it and it converts automatically, or rebuild it by hand if you want to test it first.
Before 11 December: publish your final GPT edits, migrate if the option has reached you, and test the plugin against your saved prompts. Treat any custom actions as a separate job.

Dates can move, so look at the help pages linked above again before each one.