Claude Projects vs Gemini Gems

A Gemini Gem and a Claude Project both hold standing instructions and reference files, so they look alike on a feature list. They are shaped for different jobs. A Gem is a reusable assistant: a name, instructions and optional knowledge files that you pick from a list whenever you want Gemini to behave that way, as Google puts it a shortcut for a topic you return to with similar instructions. A Claude Project is a place: a workspace with its own chats, knowledge, instructions and memory, where one body of work accumulates. If you want the same behaviour on tap across many unrelated conversations, a Gem is the closer fit. If you want the conversations about one piece of work kept together, a project is, and Gemini’s own equivalent of that is its notebooks, covered below.

If you have already compared Claude with ChatGPT, the shape will be familiar: a Gem sits nearer a custom GPT, set out in Claude Projects vs Custom GPTs, and Gemini notebooks nearer ChatGPT Projects, in Claude Projects vs ChatGPT Projects.

At a glance

• Unit. Gem: a configured assistant you open to start a chat. Claude Project: a workspace that contains chats.
• Knowledge. Both take uploaded files. A Gem can also take Google Drive files, which it reads in their latest version, and notebooks.
• Instructions. Both. A Gem’s instructions are the Gem; a project’s instructions apply to every chat inside it.
• Memory. Claude gives each project its own memory space. The Gems help pages describe no Gem-specific memory.
• Sharing. Gem: Viewer or Editor by email, a link, and general access settings. Claude Project: Can view or Can edit, inside an organisation on Team and Enterprise.
• Other apps. Claude: connectors, available in private projects. Gemini: Connected Apps at account level; the Gems pages do not describe attaching apps to a Gem.

What a Gem is

Google’s page on using Gems↗ (opens in a new tab) describes creating one at gemini.google.com: a name, instructions, and optional files under Knowledge, uploaded from your device or added from Drive. Gemini can rewrite rough instructions into fuller ones, and you can preview the Gem before saving it. A Gem made on the web also appears in the Gemini mobile app and the Gemini side panel in Google Workspace. Google ships premade Gems too, such as a brainstormer, a coding partner and a writing editor, whose instructions it publishes as examples.

You use a Gem by selecting it and starting a chat. Each chat is its own conversation that happens to begin from the Gem’s set-up, which is why one Gem can serve a hundred unrelated questions: a Gem for editing in house style works on any document you bring it.

What a Claude Project is

Anthropic’s guide to creating and managing projects↗ (opens in a new tab) describes a project as a set of chats sharing a knowledge base and project instructions, which Claude uses in every chat in the project. The chats live in the project, and you can move existing chats in or out. Two details matter for the comparison: Claude cannot see a project’s name or description, only its instructions and knowledge, and context is not shared across chats in a project unless it is added to the knowledge. Projects are on every plan, with a cap of five on free accounts.

So a project is scoped to a body of work, not to a behaviour. A project for one client holds that client’s documents and rules; you would not use it to edit an unrelated essay. How to scope one well is in Claude Projects best practices.

Knowledge files

Both take uploaded files, and both are best with a few current, clearly named ones. The differences are in freshness and citations. A Gem that uses a Drive file reads its most recent version, so edits to the file show up in the Gem without re-uploading; you can also add a notebook as a source. Gem responses cite the knowledge files unless you tick Disable knowledge citations. A Claude Project holds copies: you replace a file when the original changes. On paid plans, when project knowledge approaches the context limit, Claude switches to searching it, which the help centre says expands capacity by up to ten times.

Instructions

Google’s tips for writing Gem instructions organise them around persona, task, context and format: who the Gem is, what it does, the background it needs, and how answers should look. That suits a Gem, whose instructions are its whole identity and must cope with whatever anyone brings to it.

Project instructions do a narrower job. The project already knows what it is about from its knowledge; the instructions set the house rules for that work: audience, tone, which document wins in a conflict, what to do when unsure. They tend to be shorter and more specific, because the scope is already fixed.

Sharing

Google’s page on sharing a Gem↗ (opens in a new tab) gives two levels. A Viewer can use the Gem and see its instructions and files; an Editor can also change them, share the Gem and delete it. You can add an expiry date for a person’s access and transfer ownership. General access can be private, your organisation, anyone with the link, or public, with the last two usually unavailable on work and school accounts, where sharing follows the organisation’s Drive settings. Shared Gems are saved to a Drive folder. Google’s warning is worth repeating: anyone with access can see the Gem’s instructions and uploaded files, so do not share it with anyone who should not see them.

Claude project sharing is a Team and Enterprise feature inside an organisation, with Can view and Can edit, and each person’s chats stay private unless they share them one at a time. What that does and does not hand over is in can Claude Projects be shared?

Connectors and connected apps

On the Claude side, Anthropic’s page on connectors↗ (opens in a new tab) says connectors let Claude read from and act in other apps with the permissions of the person using them, that each person authenticates separately, and that connectors are only available in private projects. Custom connectors reach any service with an MCP server.

Gemini has Connected Apps, switched on in settings for the account rather than for a Gem. Google’s page on custom connected apps↗ (opens in a new tab) describes adding an app by its MCP server URL, currently for personal accounts in the US, in English, and says Gemini asks for manual confirmation before write actions. The Gems help pages describe instructions and files, not attaching a particular app to a Gem, so treat apps as something your account brings to every chat, Gem or not.

Gemini notebooks: the closer match

If what you want from Gemini is the project shape, look at notebooks rather than Gems. Google’s page on notebooks in Gemini Apps↗ (opens in a new tab) describes a focused space for one project, with sources, custom instructions and continuous chats in which Gemini remembers the sources and instructions. Notebooks sync with Gemini Notebook, the product formerly called NotebookLM, compared in Claude Projects vs NotebookLM. A fair pairing is Gem with a reusable Claude setup such as a skill, and Gemini notebook with a Claude Project.

Where each fits

• The same behaviour applied to many unrelated requests, such as editing in house style or drafting replies to a policy: a Gem.
• A behaviour you want colleagues to use as-is, including people who will never edit it: a Gem shared as Viewer.
• One client, product area or piece of research returned to over weeks, with its documents and decisions in one place: a Claude Project, or a Gemini notebook.
• A team working from the same briefing inside one organisation, each in their own chats: a shared Claude Project on Team or Enterprise.
• Reference files that change often and live in Drive: a Gem, which reads the latest version.

What neither one tracks

Gems and projects both shape how an assistant answers. Neither records the work that comes out of the answers: what is open, who has it, and who changed it. When that list matters, it belongs somewhere both people and assistants can read and write. On fenbs that is a board of tasks in To Do, Next Up, In Progress and Completed. Claude connects over MCP with the Claude connector, and Google’s terminal assistant with the Gemini CLI set-up, each acting under the role of the person who connected it and each change recorded with who made it. Standing rules sit once on the board as AI context, so they are not rewritten into a Gem in one product and a project in the other.