All posts

ChatGPT prompt library: how to build one you'll actually reuse

By Shaig Khaligli · Updated · 8 min read

PromptsProductivity

A ChatGPT prompt library you'll actually use is small and boring on purpose. Save a prompt only after you've reached for it a few times, write it as a template with blanks for the parts that change, name it by the job it does, and keep it somewhere you can reach from the chat box in a second or two. A notes app is enough to start. ChatGPT's custom instructions and Projects handle the context you want every time; the library is for the tasks you do some of the time.

The failure I'd watch for: saving a couple hundred prompts from a viral thread, never opening the file again, and going back to typing from memory. What follows is how I'd avoid that. I build TidyChats, which has a prompt feature, but nearly everything here works in any notes app, and I'll keep the product part in one section at the end.

When a prompt deserves saving

My rule of thumb: save a prompt the third time you type something like it. By the third time, it's clearly part of how you work.

A few other signs a prompt is worth keeping:

  • It took effort to get right. You rewrote it until the output stopped needing fixes. That effort is what you're saving.
  • It encodes a decision. "No bullet points, under 150 words, written for a non-technical manager" is a choice you don't want to remake every time.
  • The output format matters. Tables with specific columns, a JSON shape, a commit message style.

What I wouldn't save: prompts that only make sense with one specific chat's context, and anything you copied because it sounded clever but never actually ran.

A structure that holds up

Good saved prompts tend to share the same parts, in roughly this order. Both companies publish prompting guides (OpenAI's and Anthropic's), and the basics here are common advice: be clear about the task, give context, and say what the output should look like.

  1. Context: who you are or what this is for.
  2. Task: the one thing you want done.
  3. Input: the material to work on, clearly separated.
  4. Constraints: length, tone, things to avoid.
  5. Output format: exactly what should come back.

Here's an example of a saved prompt in that shape. It's an illustration, not a magic formula:

I'm a product manager writing for engineers on my team.

Rewrite the update below so it's clear and direct.

Update:
"""
[paste update here]
"""

Keep it under 120 words. No marketing language. Keep any numbers exactly as written.

Return: a one-line summary, then the rewritten update.

The triple quotes around the input aren't required, but separating your instructions from the pasted text makes it harder for the model to confuse the two.

If your prompts get long, a quick check in a token counter shows how much of the context window they use, and the prompt formatter can help tidy the structure.

Turn the parts that change into variables

A saved prompt becomes reusable once you mark the parts that change. Instead of "Rewrite this for engineers," write "Rewrite this for {{audience}}." Now one prompt covers engineers, executives, customers and a client's legal team.

Here's the example from above as a template:

I'm a {{my role}} writing for {{audience}}.

Rewrite the text below so it's clear and direct.

Text:
"""
{{text}}
"""

Keep it under {{word limit}} words. Tone: {{tone}}.

Return: a one-line summary, then the rewritten text.

Some tips that keep variables from getting out of hand:

  • Stick to one to four variables. If a prompt needs eight blanks, it's probably two prompts.
  • Use names a tired version of you will understand. {{audience}} beats {{a1}}.
  • Give defaults where one answer is usual. If you write for engineers 90% of the time, make that the default.

The double-brace style is just a convention. In a plain notes app, you'd copy the prompt and replace the braces by hand. Some tools fill them in for you.

Name prompts by the job

You'll find prompts by scanning a list or typing a few letters, so names matter more than categories. I like verb + object + detail:

  • "Rewrite: update for a non-technical reader"
  • "Review: SQL query for performance"
  • "Summarize: meeting notes into decisions and to-dos"
  • "Draft: cold email, short"

Starting with the verb groups similar prompts together when sorted, and you can type "Review" to see all your review prompts at once.

Keep categories few and broad, the same way you'd keep folders: Writing, Code, Research, Email, maybe one for your main job. Tags can carry the cross-cutting details (formal, short, Spanish) if your tool supports them.

Version the prompts that matter

Prompts drift. You add a line to fix one bad output, then another, and six weeks later the prompt is twice as long and you're not sure which lines still help.

For your most-used prompts, a light versioning habit helps:

  • Keep a short changelog line at the bottom. Something like "v3: added 'keep numbers exactly as written' after it rounded figures."
  • Test changes on the same input. Run the old and new version on one saved example and compare. If the new one isn't clearly better, keep the old one.
  • Keep the previous version until the new one has earned its place. Delete it after a week or two of use.

You don't need to do this for every prompt. Do it for the five or ten you use daily.

Use chains for multi-step work

Some tasks are really a sequence. Writing a post might be outline, then draft, then critique, then revise. Each step works better as its own prompt than as one giant instruction.

An example chain:

  1. "Outline a {{format}} about {{topic}} for {{audience}}. 5 to 7 sections, one line each."
  2. "Write a first draft from this outline. Plain language, no filler."
  3. "Critique the draft as a skeptical editor. List the 5 weakest points."
  4. "Revise the draft to fix those points. Return only the final text."

Save the steps together, in order, under one name. Without a tool, you'd paste them one after another and wait for each reply.

My opinion: chains are worth it for work you repeat the same way each time. For exploratory work, just talk to the model. A chain that forces a fixed path on a messy problem tends to get in the way.

Where to keep your library

There's no single right place. The main question is how fast you can get from "I need that prompt" to having it in the chat box.

OptionGood forDownside
Notes app or docFree, works with every AI, easy to searchCopy, paste and fill in blanks by hand
ChatGPT custom instructionsPreferences you want in every chatOne set for everything, not a menu of tasks
ChatGPT ProjectsInstructions and files for one area of workContext for a workspace, not quick task prompts
Custom GPTs, Claude Projects, Gemini GemsPackaging a recurring role or workflowSeparate per platform, more setup per prompt
A prompt extensionInserting prompts from the chat boxAnother tool to install and trust

A few details on the native options, since they get mixed up with prompt libraries:

  • Custom instructions in ChatGPT tell it things about you and how you like answers. Great for "I'm a developer, skip the basics." Not where you'd keep twenty different task prompts.
  • ChatGPT Projects can carry their own instructions and files, so a "Client work" project can hold the style guide and background once.
  • Claude Projects do a similar job on Claude and are available on all plans, though Anthropic caps Free accounts at 5 projects.
  • Custom GPTs and Gemini Gems let you save a set of instructions as a reusable assistant. Availability depends on your plan and changes over time, so check the current help pages.

My take: use those native tools for standing context, and keep a separate library for the task prompts you insert on demand. The two work well together.

TidyChats prompts (one option)

If copying from a notes app gets old, TidyChats adds a prompt library inside the chat box on ChatGPT, Claude, Gemini and Grok. Here's how it works:

  • Insert with //. Type // in the chat box and an "Insert a prompt" popover lists your saved prompts and library matches. Keep typing to filter by title, then press Enter or Tab. You can change the trigger in Settings.
  • Variables. Write {{name}} for free text, {{name:default}} for a default value, or {{tone|formal|casual}} for a dropdown. A "Fill in the prompt" form asks for each one before inserting. {{date}} and {{time}} fill themselves in.
  • Saving. Create prompts in the TidyChats window: click the TidyChats button on the right edge of the page, then Prompts on Home > "My prompts" > +, click "Save as prompt" on any message, or right-click selected text on any page and pick "Save selection as TidyChats prompt".
  • Built-in library. 122 prompts in 12 categories (Writing, Coding, Research, SEO and others). Browsing and using them is free, and "Save" copies one into your own prompts so you can edit it.
  • Chains. Build a multi-step chain in "New prompt chain", then run it with "Run" or by typing .. in the chat box. Each step waits for the previous reply to finish, and {{previous}} drops the last reply into the next step.

Typing // in the ChatGPT chat box opens the Insert a prompt popover with saved prompts and the prompt library
Inserting a saved prompt with // on ChatGPT

The limits, plainly: the Free plan saves 10 of your own prompts, and the 122-prompt library is free to use. Free includes 1 prompt chain of up to 10 steps. Premium gives unlimited saved prompts and unlimited chains with up to 25 steps, and it's one plan for all four AIs, listed on the pricing page. The prompt library tutorial walks through setup.

Questions people ask

Where can I save prompts in ChatGPT?

As far as I can tell from OpenAI's help center (October 2026), ChatGPT's built-in options are for standing instructions: custom instructions, Projects and custom GPTs. For task prompts you insert on demand, the common options are a notes app, a doc or an extension.

What's the difference between custom instructions and a saved prompt?

Custom instructions apply to your chats in general, like a standing note about who you are and how you want answers. A saved prompt is something you insert for one specific task, when you need it.

How many prompts should a prompt library have?

There's no right number, but a short list you use beats a long one you don't. Start with the handful you already retype and add more only when you catch yourself repeating something.

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

Usually, yes, since a clearly structured prompt works across models. Outputs differ, though, so run your key prompts once on each assistant you use and adjust if needed.

Are prompt packs from the internet worth saving?

They're fine as a starting point. Run them on your own real tasks, keep the few that hold up, and rewrite them in your own words with your own constraints.

Try TidyChats free

Folders, search and prompts for ChatGPT, Claude, Gemini and Grok.

Add to Chrome