The same "skill" — a persona, a workflow, a set of rules — can live in four different places in your AI stack, and most people accidentally pick the wrong one. A raw system prompt, a Claude Project's instructions, a ChatGPT Custom GPT, or a Gemini Gem all do roughly the same job: they tell the model who to be before you ask it anything. But they behave very differently once you actually start using them day to day.
Last updated: July 2026 · By the KissMySkills team
Four places, one job
Every major AI chat app now gives you a way to "install" persistent instructions so you don't have to retype your context every session. The mechanism looks different in each tool, but the underlying idea is identical: a block of text sits above your conversation and shapes every reply. Here's where that text can live.
1. The raw system prompt
This is the original, most portable version. Any tool built on top of an API — Claude's API, the OpenAI API, a custom app, a Slack bot, an internal tool your engineering team wired up — lets you set a system-level instruction that governs the whole conversation. It's the most flexible option because you (or whoever built the tool) control exactly how it's injected, how long it persists, and what else gets layered on top of it. The catch: you need a tool that actually exposes a system prompt field. The default ChatGPT.com or Claude.ai chat window doesn't hand you one directly — you have to go through Projects, Custom GPTs, or the API playground to get that level of control.
2. Claude Projects → Project instructions
Inside Claude.ai, a Project is a workspace: you give it a name, add "Project instructions," and optionally upload knowledge files (specs, brand guides, past decisions). Every new chat you start inside that Project inherits the instructions and can reference the uploaded files automatically. This is superb for a specific, ongoing body of work — "everything related to our Q4 planning," "our brand voice and templates," "this specific client's account." The instructions and the knowledge stay scoped to that one Project. Start a chat outside it, or in a different Project, and none of it follows you. That's a feature when you want separation between contexts, and a limitation when you want the same persona everywhere.
3. ChatGPT Custom GPTs and custom instructions
ChatGPT has two separate mechanisms. "Custom instructions" (in Settings) apply globally to every chat you start — closer to a personal system prompt, but only one at a time. "Custom GPTs" are the ChatGPT equivalent of a Project: you configure a name, instructions, and optionally knowledge files and actions, then that GPT becomes its own standalone entry point you can return to, share with teammates, or publish to the GPT Store. A Custom GPT is genuinely useful when you want a repeatable, discoverable tool other people on your team (or the public) can find and use without knowing anything about prompt engineering. The tradeoff is lock-in: it only runs inside ChatGPT, and if your org later standardizes on Claude or Gemini, that GPT configuration doesn't move with you.
4. Gemini Gems
Gems are Google's version of the same pattern inside Gemini — a saved persona with custom instructions you can return to from the Gemini app or side panel, with optional file grounding via Google Drive. Functionally they sit between a Custom GPT and Claude's Project instructions: persistent, named, reusable, but native only to the Gemini ecosystem. If your team lives in Google Workspace, Gems are the natural home for a skill because they integrate cleanly with Docs, Sheets, and Drive files you already have.
Comparing the four
| Location | Best for | Portability |
|---|---|---|
| Raw system prompt | API-based tools, internal bots, maximum control | Highest — works anywhere a system prompt field exists |
| Claude Project instructions | One ongoing body of work with shared files/context | Scoped to that Project only |
| ChatGPT Custom GPT | Shareable, discoverable tools inside ChatGPT/GPT Store | Locked to ChatGPT |
| Gemini Gem | Teams working inside Google Workspace / Drive | Locked to Gemini |
The part everyone misses: the skill itself is portable, even when the container isn't
Here's the point that actually matters. A well-written skill file is just plain text — usually markdown — describing a role, its rules, its output format, and how it should reason about your requests. It's not code. It's not tied to an API. It doesn't know or care whether it's sitting inside a Claude Project, a ChatGPT Custom GPT, a Gemini Gem, or a raw system prompt field. You can paste the exact same file into all four locations and get the same configured specialist in every one.
That reframes the whole decision. You don't have to pick a platform and commit. You write (or buy) the skill once, and place it wherever you happen to be working that day — Claude for deep analysis, ChatGPT for a quick GPT Store share with your team, Gemini because you're already in a Doc. The lock-in problem people worry about isn't really a skill problem; it's a container problem, and it disappears the moment the skill is a portable file instead of a platform-specific configuration.
One markdown file that turns any chat model into a chief of staff — board packs, OKRs, decision memos. Paste it into a Claude Project today, a Custom GPT next month, no rewrite needed.
View Ryan — Chief of Staff AI Skill →Practical placement guide
- You're doing one long-running project with lots of shared context — use Claude Project instructions plus uploaded knowledge files. The scoping is a feature here, not a bug.
- You want to hand a tool to your whole team with zero setup — a Custom GPT is the friendliest option, since teammates just open a link.
- You already live inside Google Docs/Sheets/Drive — a Gem keeps the skill grounded in files you're already editing.
- You're building on the API, or want the exact same behavior across multiple tools — start from the raw system prompt version and drop it into whichever wrapper you need.
- You're not sure yet which tool your team will standardize on — this is the strongest case for buying or writing skills as plain files, so you're not redoing the work later.
Configures your AI as a senior marketing automation specialist — lifecycle flows, segmentation logic, and campaign QA. Drop it into a system prompt, a Project, a Custom GPT, or a Gem interchangeably.
View Finn — Marketing Automation Specialist AI Skill →The other advantage of treating a skill as a file rather than a platform feature: version control. When you improve the instructions — tighten a rule, fix a formatting quirk, add a new capability — you edit one file and re-paste it everywhere it's installed, instead of maintaining four slightly-drifted copies inside four different platform UIs.
Keyword clusters, technical audits, and content briefs from a senior SEO specialist persona — install it as a system prompt for your dev tools or a Project instruction for ongoing content work.
View Serge — SEO Specialist AI Skill →Plug in how you (or your team) actually use AI chats and get a real monthly token-waste and dollar estimate for re-prompting vs. a persistent skill file. No signup, no email — just the math from this post applied to your numbers.
Try the calculator →Frequently asked questions
Can I use the same skill file in Claude, ChatGPT, and Gemini?
Yes, as long as the skill is written as plain markdown or text instructions rather than tied to a specific tool's features (like a Custom GPT action or a Claude-specific tool call). Paste it into Project instructions, a Custom GPT's instructions field, a Gem's configuration, or a raw system prompt — the persona and rules carry over identically.
Which location should I choose if I'm not sure yet?
Default to whichever chat app you already use daily, and don't overthink the container — the skill itself is what does the work. If you might switch tools later, keep the original file saved somewhere outside any single platform so you can re-paste it anywhere.
Do Custom GPTs or Gems store the skill more "permanently" than Project instructions?
Not meaningfully — all three save your instructions and reload them every time you return to that Project, GPT, or Gem. The real difference is scope and shareability, not durability: a Custom GPT can be shared with a link or published publicly, a Gem stays inside your Google account, and Claude Project instructions stay scoped to that one Project alongside its files.
In summary:
System prompts, Claude Project instructions, ChatGPT Custom GPTs, and Gemini Gems are just four different containers for the same idea — the container decides scope and shareability, not portability. A skill written as a plain file, like Ryan (Chief of Staff) or Finn (Marketing Automation), works in all four, so you buy or write it once and place it wherever you're working that day.