Gemini Gems vs Custom GPTs: Instructions, Files, and Memory

Gemini Gems and Custom GPTs both let you create a reusable AI assistant: a named configuration with specific instructions, an optional persona, and knowledge files the assistant draws on. They serve the same purpose but are built on different platforms, work in different ecosystems, and are at very different points in their product lifecycles.

The important news is that Custom GPTs are scheduled to retire on December 11, 2026. If your workflows depend on them, now is the time to plan the move. A tool like fenbs keeps your shared prompts, instructions, and project rules in one place outside any single platform, so the move does not lose the work behind them.

What Custom GPTs are

A Custom GPT is a saved version of ChatGPT built using OpenAI's GPT builder. You write a set of instructions, optionally upload files the assistant should reference, and give it a name. You can keep it private, share it with your organization, or publish it to the GPT store. Custom GPTs can also call external APIs through Actions.

They have been widely used for building reusable assistants: a writing-style guide that turns into a writing coach, a company FAQ that turns into an internal support bot, a prompt library that turns into a task-specific assistant. That utility is why their retirement matters — the workflows built around them need to move somewhere.

OpenAI has announced that Custom GPTs will retire on December 11, 2026. OpenAI's plans for what replaces them are described on OpenAI's own help pages↗ (opens in a new tab) — check there rather than relying on any summary here.

What Gemini Gems are

A Gem is a saved configuration of Gemini with a name, instructions, and optional knowledge files. You create it once and open it whenever you want that behavior. Like a Custom GPT, it is a reusable persona: a writing assistant with your style guide, a research assistant with your preferred format, a coding helper with your project's conventions.

Gems are available on Google One AI Premium and Workspace accounts. On personal accounts, Gems are scheduled to become skills from November 2026. The transition details are on Google's own help pages — verify the current state there, especially if you are building workflows around Gems that need to remain stable past November.

How each handles instructions

Both Gems and Custom GPTs accept plain-text instructions that define the assistant's behavior: its persona, the tasks it should focus on, the format it should use for responses, and any constraints on what it will or will not do. The instruction quality matters much more than which platform you use.

Well-written instructions are portable: they describe a behavior in plain language, not in platform-specific code. Instructions you wrote for a Custom GPT can be transferred to a Gem with minor adjustments. The logic of "here is the role, here are the rules, here is the format" applies to both.

Knowledge files and reference material

Both platforms let you upload files the assistant uses as reference. In a Custom GPT, the files are uploaded once and the assistant draws on them when answering. In a Gem, knowledge files work similarly — the assistant uses them alongside its general Gemini training.

Neither platform makes a file browsable by other tools or people outside the assistant. If your reference material is a style guide, a product document, or a set of procedures, keeping a master copy outside the assistant means you can update it independently of the platform.

Moving from Custom GPTs to Gems

If you have Custom GPTs you rely on, the retirement deadline means you need a plan before December 11, 2026. The practical steps are: export or copy the instructions from each Custom GPT, review which knowledge files are still current, and create a Gem (or another replacement) with the updated instructions.

Not every Custom GPT needs to move to a Gem. Some may be better replaced by a project in Claude, a shared instruction set in your team's AI tool, or a shared prompts library. The right answer depends on which platform your team uses most and which features — especially external API Actions — you depended on.

For the Claude equivalent, see Claude Projects vs Custom GPTs. For a comparison with Claude's own project system, see Claude Projects vs Gemini Gems.

How to do this with fenbs

1. Create a free account and add a project for managing your shared prompts and assistant configurations.
2. Copy the instructions from your most-used Custom GPTs into the Skills and prompts library in fenbs, with notes on which task each is for.
3. Connect your AI apps over MCP from Integrations so connected assistants can list and use your saved prompts.
4. Before December 11, 2026, review each Custom GPT and decide: move to a Gem, move to a Claude Project, or retire it. Record the decision in fenbs.
5. After the move, test the replacement assistant with the same tasks you used the Custom GPT for, and update the prompt if the output differs.