← Blog · 2026-10-12

How to organize a prompt library for ChatGPT, Claude and Gemini

Most people who use AI chat every day end up retyping the same ten prompts: “summarise this”, “review my code”, “rewrite this email in a friendlier tone”. A small, well-organised prompt library saves that time and, more importantly, makes your results more consistent. Here is a system that works without any special tool, and how to make it faster with one.

1. Only save prompts that earned it

Don’t collect prompts from the internet in bulk. Save a prompt when you have used it at least twice and the result was good. A library of 25 prompts you trust beats 500 you never open.

2. Write prompts as templates with variables

The part that changes each time should be obvious. Mark it as a variable: Explain {{topic}} to {{audience:a beginner}} in 5 bullet points. The double curly braces are a common convention; a default after a colon ({{audience:a beginner}}) saves typing when the usual value is fine. Templates also stop you from accidentally leaving old context (a previous client’s name, last week’s numbers) in a reused prompt.

3. Name prompts by what they produce

“Code review: bugs + edge cases” is easier to find than “My best prompt v3”. Start the title with the task (Summarise, Review, Rewrite, Plan, Explain) so typing the first word finds it.

4. Folders for context, tags for type

• Folders answer “where do I use this?”: Work / Clients, Coding, Writing, Personal. Keep it to one level or two.
• Tags answer “what kind of prompt is this?”: #summary, #email, #code, #translate, #es. A prompt can have several tags but lives in one folder.

If you only want one mechanism, use tags: they are more flexible and search well.

5. Keep a short “why it works” note

Add one line at the end of the prompt or in its title: what model it was tuned for, or what it avoids (“asks for sources”, “no marketing tone”). When a prompt stops working after a model update, this tells you what to fix.

6. Review once a month

Delete prompts you have not used, merge near-duplicates, and update the ones that drifted. Use counts help: if your tool tracks how often you insert each prompt, sort by it.

Where to keep the library

• A notes app or document (Notion, Obsidian, Google Docs): free, searchable, easy to share. Downside: copy, switch tab, paste, edit the variables by hand.
• Built-in features: ChatGPT custom instructions and Projects, Claude Projects and styles, and Gemini Gems are great for standing instructions (tone, role, background). They are less suited to dozens of one-off task prompts, and they don’t move between the three services.
• A text expander: fast, works everywhere, but variables and search are usually basic.
• A browser extension that inserts into the chat box. Our Prompt Library opens with Alt+P or by typing // in ChatGPT, Claude or Gemini, searches by title, tag or folder and inserts at your cursor. Free covers 10 prompts with tags on all three sites; Pro adds unlimited prompts, folders, a fill-in form for {{variables}} and JSON import/export. Prompts stay in your browser’s local storage, with no account and no server.

Whatever you choose, keep an export. A JSON or Markdown backup means your library survives a tool change.

A starter set

• Summarise {{text}} in 5 bullets, then list open questions.
• Review this {{language}} code for bugs, edge cases and readability. Suggest concrete fixes.
• Rewrite this email to be {{tone:friendly and concise}}. Keep all facts.
• Act as a critical reviewer of {{plan}}. List the 3 biggest risks and how to test each.
• Translate to {{language:Spanish}}, keeping formatting and code unchanged.

FAQ

Should I share prompts with my team?

Yes, via an exported file or a shared doc, but remove anything confidential first: prompts often contain client names or internal details.

Do prompts work the same in ChatGPT, Claude and Gemini?

Mostly. Clear task, context and output format transfer well; model-specific tricks don’t. Note which model a prompt was tested on.