Claude added Skills. ChatGPT has had GPTs since 2023. Gemini has Gems. Three different attempts at the same basic idea: let users package an AI behavior — instructions, reference files, sometimes code — and reuse it across sessions. The implementations are surprisingly different, and the right pick depends almost entirely on which platform you live in.
If you're heavy ChatGPT, build a GPT. If you're a developer or live in Claude Code, write a Skill. If your data lives in Google Workspace, use a Gem. There isn't really a head-to-head winner — each is the best at what its platform optimizes for.
Pricing (because that always comes up)
- ChatGPT GPTs — Free to use other people's GPTs. Building a GPT requires ChatGPT Plus ($20/mo) or higher.
- Claude Skills — Free to use, included in Claude Pro ($20/mo). Author Skills via files; no separate subscription.
- Gemini Gems — Free to use basic Gems with a Google account. Advanced Gems (with Workspace integration) need Gemini Advanced ($20/mo via Google One AI Premium).
ChatGPT GPTs — the easiest to build, the largest ecosystem
GPTs have been around since November 2023, and the ecosystem is the deepest of the three. The GPT Store has hundreds of thousands of GPTs, most of them garbage but the top ones genuinely useful — Consensus for academic search, Wolfram for math, custom code reviewers, niche professional tools.
Building a GPT is friendly. The wizard walks you through:
- System prompt (the personality/instructions)
- Knowledge files (up to 20 files, ~100MB each)
- Actions (custom API integrations via OpenAPI specs)
- Capabilities (web browsing, image generation, code interpreter)
For non-developers wanting to ship a useful custom AI to coworkers or customers, GPTs are still the easiest path. The UI is clean, sharing is one URL, and the ecosystem effects mean people will actually find it.
Drawbacks: GPTs are locked to ChatGPT. They don't work in Claude or anywhere else. If you eventually want to migrate platforms, you're rebuilding from scratch.
Claude Skills — the developer-friendly version
Skills are Anthropic's bet on a different shape. Instead of a wizard UI, Skills are files: a markdown description, optional reference files, optional scripts that run in Claude Code. Drop a Skill folder into your Claude.ai workspace or Claude Code config, and Claude loads it automatically when relevant.
What this unlocks: Skills are version-controlled by default (they're text files you commit to git), they work in both Claude.ai chat and Claude Code CLI, and they can include executable scripts that Claude can run. That last part matters — a GPT can describe a workflow, but a Skill can actually do it.
Example Skill I use daily: a code-review Skill that has my code style preferences, runs a script that pulls the diff from the current branch, and gives Claude a structured review prompt. The same Skill works in Claude.ai when I paste code, and in Claude Code when I'm in a project. One artifact, two surfaces.
The Skills marketplace launched in early 2026 — Pro/Max users can browse and install Skills published by other developers. Quality varies, but the killer Skills are genuinely useful.
Drawbacks: writing a Skill from scratch requires being comfortable with markdown and (for advanced cases) writing scripts. Non-developers will hit a wall the GPT wizard saves them from.
Gemini Gems — for the Google Workspace native
Gems are Google's contribution to the custom-AI category. The structure is similar to GPTs: name, description, instructions, optional knowledge files. You build them in the Gemini sidebar; they appear in your Gem list.
The differentiator: Gems integrate deeply with Google Workspace. A Gem can be configured to always read from a specific Drive folder, summarize a Gmail label, pull events from a Calendar, or work over a specific Sheets workbook. This is what neither GPTs nor Skills can match — neither has a native bridge into your actual Workspace data.
If your work data is in Google (most knowledge workers' is), Gems become surprisingly useful. A "weekly status update" Gem that reads your Calendar + recent Gmail and drafts the update is hard to replicate cleanly in ChatGPT or Claude.
Drawbacks: outside Workspace, Gems are weaker than GPTs and Skills. The customization options are fewer, no executable scripts, and the model (Gemini) isn't as strong as Claude or GPT-5.4 on most generation tasks.
The honest comparison
- Easiest to build for non-developers → GPTs. The wizard saves you from caring about file structure.
- Best for sharing publicly → GPTs. The Store has the network effect.
- Best for developer workflows → Skills. Text files, git, scripts, works in Claude Code.
- Best Workspace integration → Gems. Native Drive/Gmail/Calendar access.
- Most powerful overall → Skills, because of the executable-scripts capability.
- Best ecosystem of pre-built options → GPTs by far. Skills marketplace is newer.
What about open-source / platform-agnostic alternatives?
Tools like Open WebUI and LibreChat let you build custom personas with system prompts, knowledge files, and tool integrations against any LLM. For self-hosted, multi-model setups, these are the right call. They're more work to set up and lack the platform-specific integrations (GPT Store, Skills marketplace, Workspace), but they don't lock you in.
The verdict — pick by platform, not by feature
If you're heavy ChatGPT and want custom AIs your coworkers or customers can use: build GPTs. The wizard saves you a lot of effort and the Store gives you reach.
If you're a developer or live in Claude Code: write Skills. Version control, cross-surface (chat + CLI), and the script-execution capability make them strictly more powerful for developer workflows.
If your work data is in Google Workspace and you want AI tools that operate on your real data: build Gems. The Workspace integration isn't optional for this use case.
If you want to be platform-agnostic: self-host Open WebUI or LibreChat. More setup, no lock-in.
One honest observation: most heavy AI users I know end up with one primary system (because of where they live) and ignore the others. There's no need to learn all three — pick by your dominant platform and build there.
FAQ
What are Claude Skills?
File-based packages of instructions, reference files, and optional scripts that extend Claude's behavior. Work in both Claude.ai and Claude Code.
Difference between Skills and GPTs?
GPTs are built in a wizard, live in ChatGPT only. Skills are files, version-controlled, work across Claude surfaces. Skills can include executable code; GPTs can't.
Are Gemini Gems useful?
Yes if you live in Google Workspace — native integration with Drive, Gmail, Calendar. Outside Workspace, GPTs and Skills are more capable.
Can I share a Claude Skill?
Yes — via git, shared folder, or the Skills marketplace.
Which should I use in 2026?
Pick by platform. ChatGPT heavy → GPTs. Developer/Claude Code → Skills. Workspace native → Gems.