A Gemini Gem is a saved persona you deliberately switch into, and a Claude Skill is a packaged procedure Claude pulls in by itself when your request matches it. If you want a reusable expert you talk to on purpose, build a Gem. If you want a repeatable method Claude applies without being reminded, build a Skill. Most articles line these two up as competitors, and that framing is the reason people stay confused: they are not the same category of thing, and knowing which problem each one solves matters more than knowing which vendor wins.
What a Gemini Gem actually is
Google defines Gems as "customized versions of Gemini that help you tackle repetitive tasks or get deep expertise in new areas." In practice a Gem is three things saved together under a name you choose: a set of freeform instructions written in plain English, an optional set of uploaded Knowledge files, and the name itself. You open the Gem, you chat inside it, and Gemini behaves according to those instructions for the whole conversation.
Google's own guidance on writing the instructions is refreshingly concrete. It suggests covering four things: Persona, meaning the role the Gem plays, Task, meaning what you want it to do, Context, meaning the background it should assume, and Format, meaning how the output should be structured. There is also a button that asks Gemini to rewrite and expand your draft instructions for you, which is useful if the blank box is intimidating.
Two practical constraints catch people out. Creating, editing and deleting Gems is a web-app job at gemini.google.com; once saved, a Gem shows up in the mobile app and in the Gemini side panel across Google Workspace, but you cannot build one from your phone. And Google's help page notes that for now, Gems cannot be used with Gemini Live, so the voice conversation mode does not inherit your Gem's instructions. The feature requires a personal Google Account and users aged 13 or older, or the applicable age in your country.
What a Claude Skill actually is
Anthropic defines Skills as "folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized tasks." That word folders is the whole difference. A Skill is a directory containing a SKILL.md file with a short YAML header at the top, plus any extra markdown files, reference documents or executable scripts you want to bundle alongside it. You zip the folder and upload it.
The mechanism underneath is called progressive disclosure, and it is the most interesting engineering choice in either product. Anthropic's developer documentation breaks the loading into three levels: the name and description sit in Claude's context at all times at roughly 100 tokens per Skill, the body of SKILL.md loads only once the Skill is triggered and is meant to stay under about 5,000 tokens, and bundled files load only when Claude actually opens them. Scripts run through bash and only their output enters the conversation, never the code itself.
The single biggest difference: you choose a Gem, but a Skill chooses itself. Claude matches your request against each Skill's description field and loads the ones that fit. That is why you can have many Skills installed at once and only be in one Gem at a time.
One naming note worth getting right, because it causes real confusion when you search for documentation. Anthropic's consumer help pages call the feature Skills, while the developer documentation at platform.claude.com calls it Agent Skills. Same feature, two labels depending on which surface you arrived from. Neither name replaced the other, so do not read it as a rebrand.
Gemini Gems vs Claude Skills, side by side
| Dimension | Gemini Gems | Claude Skills |
|---|---|---|
| What it is | A saved persona: name, instructions, optional Knowledge files | A packaged procedure: SKILL.md plus optional scripts and reference files |
| How it starts | You pick it manually and chat inside it | Claude triggers it automatically when your request matches its description |
| How you build it | Fill in a form in the Gemini web app, plain English | Write a markdown file, zip the folder, upload it |
| Can it run code | No, a Gem stores instructions and files only, it cannot carry a script of yours | Yes, bundled scripts execute in a sandboxed environment |
| Using several at once | One Gem per conversation | Many installed together, each costing about 100 tokens until triggered |
| Where it works | Gemini web, mobile app, Workspace side panel; not Gemini Live | claude.ai, Claude Code, the Claude API, AWS and Microsoft Foundry |
| Sharing | Share by email or link, with Viewer or Editor access | Personal plans: individual to your account. Anthropic's support pages say Team and Enterprise owners can provision Skills org-wide; its developer docs still say they cannot |
| Syncs between surfaces | Yes, a saved Gem appears everywhere you use Gemini | No, upload separately to claude.ai, the API and Claude Code |
Who can use each one
Gems are the more generous of the two. Google gated them behind a paid subscription when they launched in 2024, then announced in March 2025 that Gems "are starting to roll out for everyone at no cost in the Gemini app." Google's current help pages list no subscription requirement at all, only a personal Google Account and a minimum age of 13, or the applicable age in your country.
Claude splits the feature in two, and Anthropic's own pages disagree slightly on the boundary, so read this carefully. The Skills Anthropic builds and maintains, covering PowerPoint, Excel, Word and PDF creation, fire automatically for users across plans with no setup. Uploading your own custom Skill is where the two sets of documentation part company: the help center lists Free, Pro, Max, Team and Enterprise, while the developer documentation lists custom Skill uploads on claude.ai as Pro, Max, Team and Enterprise, in both cases with code execution turned on. Treat custom Skill authoring as a paid-plan feature and you will not be disappointed.
- Enable code execution first, then switch the Skill on under Customize, Skills.
- On Team and Enterprise, an owner has to switch on both code execution with file creation and Skills at the organization level before anyone sees the option.
- Anthropic's help center and developer docs describe the menu path with slightly different labels, which is a fair warning that this UI is still moving. Look for the settings section governing capabilities or features.
- Skills uploaded in one place do not appear in another. A Skill added to claude.ai is not visible to the API, and Claude Code reads Skills from folders on your own machine.
Sharing is where Gems pull clearly ahead
Google added proper sharing to Gems, and it works the way sharing works everywhere else in Google's products. You share a Gem by email or by copying a link, general access runs from Private through Anyone with the link to Public, and each recipient gets either Viewer access, which lets them use the Gem and see its instructions and files, or Editor access, which lets them change the instructions, change the files, reshare it and delete it. You can set an expiry on someone's access.
Google flags the obvious hazard plainly: if you have uploaded files to a Gem, do not share the Gem with anyone you would not want reading those files. Access to the Gem carries access to its Knowledge. That is a real consideration if you loaded a Gem with anything from your own records.
Claude's story depends entirely on your plan. Anthropic's developer documentation still says custom Skills on claude.ai are individual to each user and cannot be centrally managed by admins, but its own support pages have moved past that: since May 2026, Team and Enterprise owners can upload a Skill under Organization settings and have it appear for everyone in the organization, enabled by default, with individual members free to toggle it off. On personal Free, Pro and Max accounts the older description still holds, so a colleague who wants your Skill has to upload the zip themselves. Sharing does exist through the API, where custom Skills are workspace-wide, and through Claude Code plugins, but neither of those is a consumer path.
Where this comparison honestly breaks down
The comparison collapses the moment you try to declare a winner, because the two features answer different questions. A Gem answers "who should the assistant be right now." A Skill answers "how should the assistant do this particular job, whenever it comes up." Ask a Gem to reliably run a five-step process on a file and you are relying on the model to follow written instructions faithfully every time. Ask a Skill to hold a consistent tone across an open-ended conversation and you are using a trigger-based mechanism for something that has no trigger.
The nearer comparison to Gems is not Skills at all, it is Claude Projects, which also hold instructions plus uploaded files and which you also enter deliberately. And the nearer comparison to Skills on Google's side is the tooling and extension layer, not Gems. Any article that presents Gems and Skills as two implementations of one idea is flattening that.
Rule of thumb: if the thing you want to save is a voice, a role or a body of reference material you will consult in conversation, build a Gem. If the thing you want to save is a checklist, a house format, or a procedure with steps that must run the same way every time, build a Skill. If you want both, build the Gem first, because it takes five minutes and no file structure.
The security question people skip
Skills carry a risk Gems do not, precisely because they are more capable. Anthropic warns to use Skills only from sources you created yourself or obtained from Anthropic, because a Skill can direct Claude to run code and invoke tools in ways that do not match its stated purpose. The documentation singles out Skills that fetch data from external URLs as particularly risky, since fetched content can carry instructions of its own. The advice is to treat installing a Skill like installing software.
- Downloading a community Skill zip is closer to running a script than to pasting a prompt. Read what is inside it.
- A shared Gem cannot execute anything, but it can carry files the sharer did not mean to expose, and it can carry instructions you did not read.
- Neither feature is a place to store anything you would mind another person seeing if a link leaked.
Neither one is memory, and both are locked to one vendor
This is the part that trips up people who arrive at Gems or Skills hoping to solve the bigger problem. A Gem remembers what you wrote in its instructions and whatever you uploaded to its Knowledge, and nothing else. It does not accumulate what you tell it over months. A Skill remembers a procedure, not you. Neither one is a growing record of your own life, your documents, your receipts, your notes and your decisions.
They also do not travel. A Gem lives inside Google's app and cannot be opened in Claude. A custom Skill does not even move between Anthropic's own surfaces without a second upload, let alone to a different vendor. So the configuration work you put into either one is effectively a deposit into one company's product. That is fine for a writing-tone Gem you can recreate in ten minutes. It is a worse deal for a knowledge base you spent a year assembling.
That gap is what MemX is built for. Instead of re-uploading the same files into a Gem, a Project and a Skill folder, you capture things once, photos, PDFs, voice notes and chats, and MemX reads them with OCR and transcription so you can ask a plain-language question later and get the answer plus the document it came from. It is private by architecture, with per-user isolation, customer-managed encryption keys and encryption at rest. It does not replace either Gems or Skills; it is the layer neither of them is trying to be.
A sensible way to decide
- You already use Gemini and want a reusable assistant today, free, in five minutes: build a Gem.
- You want to hand a colleague a working assistant with a link: build a Gem, and check the file warning before you share it.
- You pay for Claude and keep repeating the same instructions about how you want work formatted: build a Skill so you stop repeating them.
- Your task involves a deterministic step, such as processing a spreadsheet the same way every month: a Skill can carry a script, a Gem cannot.
- You want the assistant to know your own accumulated documents across every tool you use: neither feature does that, and no amount of configuration will make it.
01Are Gemini Gems and Claude Skills the same thing?
No. A Gem is a saved persona you switch into manually, holding instructions and uploaded files. A Skill is a folder of instructions and optional scripts that Claude loads automatically when your request matches it. The closer Claude equivalent to a Gem is a Project.
02Do you need to pay for Gemini Gems?
No. Gems launched as a paid feature in 2024, and Google said in March 2025 they were rolling out to everyone at no cost. You need a personal Google Account and must be 13 or older, and Gems can only be created or edited in the web app.
03Can free Claude users create custom Skills?
Anthropic's own documentation disagrees with itself. The help center lists Skills on Free, Pro, Max, Team and Enterprise with code execution enabled, while the developer docs list custom Skill uploads on claude.ai as Pro and above. The Skills Anthropic builds, for PowerPoint, Excel, Word and PDF, run automatically either way. On Free, check Settings first.
04Can you share a Gem with someone else?
Yes, by email or link, with Viewer or Editor access and an optional expiry date. Google warns that anyone you share a Gem with can see the files uploaded to it, so check the Knowledge section before sending the link.
05Do Gems or Skills remember your past conversations?
Not in the way people expect. A Gem only knows its instructions and its uploaded Knowledge files, and a Skill only knows the procedure it packages. Neither accumulates a personal record of you over time, and neither works outside its own vendor's products.
