By the ConnectLabz Systems team · Updated 23 September 2026
Here is the short answer to Claude skills vs Gemini Gems: this year they stopped being the same kind of thing. A skill is now a file format that Claude, Gemini’s command-line tool and ChatGPT can all read. A Gem is a saved assistant that lives inside Google’s app. So the real choice is where your marketing playbook lives, and who owns it.
Maybe you already feel why that matters. One client has a Gem that has quietly started ignoring the tone file you attached. The same client has a custom GPT from back in May, and now there’s a Claude skill a peer swears by. You’re about to paste the same brand rules into a fourth place.
Each copy drifts a little. Each new tool means rebuilding it again. Below you’ll find what each option actually is in 2026 and the one question that settles it. Then a test you can run this afternoon to see whether your playbook would survive a move.
Key Takeaways
- Skills are now an open file format. Claude, Gemini CLI, ChatGPT and Codex, Cursor and GitHub Copilot all read the same skill folder.
- Gems and custom GPTs are assistants you save inside one company’s app. You pick one by hand, and it lives on that company’s servers.
- Ownership is the deciding question. Keep the playbook as a file you control, and treat Gems and GPTs as front ends you type into.
- A Gem still wins for some teams. If everyone works in Google Workspace and wants an assistant that reads shared Drive files, start there.
Does Gemini Have Skills Like Claude?
Yes, in one place. Google’s Gemini CLI, the command-line version of Gemini, reads the same skill folders Claude does, because both follow the open Agent Skills standard. Gemini’s app does not. There, the closest thing is a Gem, a saved assistant you pick by hand. So the answer depends on which Gemini you use.
Most marketers only ever meet one Gemini: the app in the browser, with the list of Gems down the side. That one has no skills folder. Its sibling runs in a terminal, and its documentation (last updated 30 April 2026) describes something familiar. At the start of each session it scans a few set folders, loads every skill’s name and description, and only pulls in the full instructions when a task matches. Google’s docs even name the tool it calls to do that, activate_skill.
Which is, almost step for step, how Claude behaves. No coincidence there. Anthropic released the skill format as an open standard. The standard’s site now names Gemini CLI, Codex, ChatGPT, GitHub Copilot, VS Code and Cursor as readers.
ChatGPT sits somewhere in the middle. Its custom GPTs behave a lot like Gems. On its business plans, at the time of writing, it also reads skills in that same folder format. Check what your own plan includes before you build around that.
Which Gemini are you using? Typing into gemini.google.com means you have Gems. Running Gemini in a terminal means you have skills.
What Each One Actually Is
What is a Claude skill?
A Claude skill is a folder with a SKILL.md file inside: a name, a one-line description of when to use it, and the instructions, plus any examples or scripts the job needs. Claude reads only the name and description until a task matches, then loads the full playbook. Ownership stays with you.
Why the loading order matters: it’s bigger than it sounds. Anthropic’s own explainer, Skills explained (5 March 2026), puts the metadata at roughly 100 tokens and the full instructions at under 5,000, loaded only when needed. In practice you can keep twenty marketing playbooks installed, such as a brief template and an ad-review checklist. None of them crowds the conversation until you ask for that kind of work.
What is a Gemini Gem?
A Gemini Gem is a saved version of Gemini with standing instructions and optional knowledge files, which can include Google Drive documents and NotebookLM notebooks. Open the Gem you want from your Gem list and it behaves that way until you leave it. It lives inside Google’s app.
Setting one up takes minutes, and that’s the appeal. Later, a catch shows up. A thread in Google’s Gemini help community describes a Gem ignoring its attached document once the instructions grow long. Six months of client feedback does exactly that to a brand playbook.
What is a custom GPT?
A custom GPT is a saved ChatGPT assistant with instructions, knowledge files and optional actions. OpenAI’s documentation puts the instruction limit at 8,000 characters and the knowledge limit at 20 files, which is roomy for a persona and tight for a full brand playbook with examples.
Eight thousand characters is about 1,300 words. A real voice guide with do-and-don’t examples, a banned-phrase list and three approved sample paragraphs runs past that before you’ve added the offer rules. People end up splitting it into uploaded files and hoping the GPT consults them.
Put those three side by side and a pattern appears. Two of them are places you visit. One is a file you carry.
Claude Skills vs Gemini Gems: The Real Difference Is Who Owns the Playbook
A useful way to choose is the rule we call own the file, rent the front end. Keep your playbook in a file you control, one that any skills-compatible assistant can read. Use a Gem or a GPT as the window you type into, never as the only place the playbook exists.
Models change every quarter. Your brand rules don’t. A client’s banned claims, the tone they signed off on, the way they want a landing page structured: that knowledge is the asset, and the assistant is the engine running it this month. Put the playbook inside one vendor’s app and switching engines means rebuilding the asset. Kept in a file, switching means pointing a different tool at the same folder.
We should admit something here. An earlier version of this very page walked you through copying a skill into a GPT and trimming it down into a Gem by hand. It was honest advice for 2025. Much of it is unnecessary now, and we’d rather say so than leave it up.
We build the same way. Every AI marketing system we sell ships as one library of skill files with thin notes for each assistant; our Content Creator System keeps its brand-voice playbook as a skill folder, for example. When a client changes a banned phrase on a Monday, then, it changes once, in one file, and every tool reading that folder picks it up by the afternoon.
Skills, Gems and GPTs Side by Side
| Claude skill | Gemini Gem | Custom GPT | |
|---|---|---|---|
| Where it lives | A folder you keep | Inside the Gemini app | Inside ChatGPT |
| How it starts working | Loads itself when a task matches its description | Chosen from your Gem list | Picked from your GPT list |
| What else can read it | Every tool that supports the standard (Gemini CLI, Codex, Copilot and more) | Only Gemini | Only ChatGPT |
| Room for a long playbook | Instructions load only when needed; examples sit in separate files | Long instructions can crowd out attached files (reported by users) | 8,000-character instructions, up to 20 files |
| Live company documents | Whatever the tool running it can connect to | Google Drive and NotebookLM | Uploaded files and actions |
| Moving it elsewhere | Copy the folder | Rebuild it | Rebuild it |
Read the last row twice. Everything above it is a feature comparison that will shift with the next product update. That bottom line decides how much rework your future self does.
When a Gem or a GPT Is the Right Call
Owning the file is the right default. And yet a Gem is sometimes the better tool, and pretending otherwise would be selling you something.
- Your whole team lives in Google Workspace. A Gem that reads a shared Drive document your account manager updates every Friday is doing real work, and nobody has to learn where a skill folder lives.
- A colleague needs a quick helper today. Either one takes minutes and needs no files, no terminal and no explanation.
- You want to share it publicly. A custom GPT can go in the GPT Store; a skill folder can’t be browsed like that.
- The task is a one-off. Building a portable playbook for something you’ll do twice is effort without a payoff.
There’s a quieter version of this, too. A skill folder that sits on one person’s laptop, that nobody else on the team knows how to open, is not ownership. Call it what it is: a file with a single point of failure. If that’s where you’d end up, a shared Gem may genuinely be the safer choice until the playbook has a proper home.
How to Test Whether Your Playbook Is Portable (in an Afternoon)
Portability is also a quality test. A playbook that only works in one assistant usually leans on something it never wrote down, and a second assistant finds that gap fast.
- Write the playbook as one SKILL.md with a description that says exactly when to use it (“Use when writing a landing page for a local-service client”).
- Move examples and banned phrases into separate files beside it, so the instructions stay short.
- Load it in Claude and run one real brief.
- Load the same folder in a second skills-compatible tool, such as Gemini CLI or ChatGPT on a plan that supports skills.
- Run the identical brief and put the two outputs next to each other.
- List every place the second tool went wrong. Each one is a step your playbook implied instead of stating. Write it in, and the first tool usually gets better too.
Whatever tool runs it, keep the review habits from our own work: one checklist file and one banned-phrase list. No statistic goes out without a source link, and a person approves anything before it’s published. Models differ in tone. Standards shouldn’t.
What Skills Still Get Wrong
Skills have limits, and a few of them bite marketers specifically.
A vague description is the most common one. If it says “helps with marketing”, the skill rarely triggers when you need it, so write descriptions the way you’d brief a new hire on when to call you. Installing dozens of skills at once makes behaviour less predictable, because more of them compete to match each request. And a skill downloaded from a stranger can carry instructions you never intended to run. Read any third-party skill before you install it, the same way you’d read a contract.
None of that argues against skills. What it argues for is a small, well-described library you actually understand.
Conclusion
The question was never really Claude versus Google. It’s whether the playbook your business depends on sits in a file you can carry, or inside an app you rent. Once that’s settled, choosing an assistant gets a lot less dramatic: you pick whichever front end suits the task this month, and the playbook comes with you.
You now know the difference that decides it. What you don’t know yet is whether your own playbooks would survive a move, or which of them should become files first. See the five ConnectLabz systems as the skill files you would own →
FAQ
Can Claude skills be used with ChatGPT?
Yes, if your plan includes them. As of September 2026, OpenAI supports skills in this folder format on ChatGPT’s business plans and in Codex. Elsewhere you can paste a skill’s instructions into a custom GPT or a project, though it won’t load itself automatically the way it does in Claude.
Are Claude skills only for Claude Code?
Not at all. Skills work in the Claude apps as well as in Claude Code, and a growing list of other tools reads the same folders, Gemini’s command-line version among them. What changes between tools is how you install the folder, never the playbook inside it.
What is the difference between Gems and GPTs?
Both are saved assistants inside one company’s app. A Gem runs in Google’s Gemini and can read Google Drive files and NotebookLM notebooks. A custom GPT runs in ChatGPT, holds up to 20 uploaded files and can use actions. Neither can be read by the other company’s tools.
Gemini Gems vs Claude Projects: which should I use?
They do slightly different jobs. A Gem carries standing instructions you switch into, while a Claude project is a workspace holding chats and documents about one initiative. For recurring marketing work, the durable part belongs in a skill file either way, with the Gem or project as the place you work.
Do I need to know how to code to use skills?
You don’t. Writing a skill is closer to writing a brief than to programming: it’s a plain-text instruction file sitting in a folder. You’ll need code execution switched on in Claude’s settings, and some skills bundle scripts, but you can write and use a useful marketing skill without writing any code.
