Guide
How to Organize AI Prompts
A prompt library fails in one of two ways. You cannot find the prompt you know is there, or you find four versions of it and do not know which one worked. A few rules, set once, fix both: how to title a prompt, how many categories to keep, what tags are for, and when to use variables instead of a copy.
Last updated: October 10, 2026
Organize for the moment you look for a prompt
Think about how you get a prompt back. You do not browse folders. You type a few letters and pick from what comes up. In PromptTray that is Spotlight (Control Shift K). It looks through titles, categories, tags and the text of each prompt, and it ranks a match in the title first.
So the title does most of the work, the category narrows the list, and tags catch what the other two miss. The rules below come in that order.
Titles: verb, object, what changes
Start every title with the verb you would type when you need the prompt: Summarize, Rewrite, Review, Draft, Compare. Then the object. Then, in curly braces, the part that changes each time.
- Good: “Summarize meeting notes into action items”
- Good: “Write a launch tweet for {product}”
- Weak: “Email prompt v2”, “good one”, “NEW marketing”
Leave out version numbers, dates and the name of the AI app. A version number says there is an older copy you have not deleted. The app name is wrong the first time you use the prompt somewhere else.
Categories: a handful, named after jobs
A category answers one question: what kind of work is this? Writing, Research, Coding, Marketing, Support. It is not a topic, a client or an AI app.
Two tests keep the set honest. You can file a new prompt in two seconds without thinking, and every category holds at least three prompts. A category with one prompt is a tag that got promoted.
Keep the number small. PromptTray’s free plan allows 5 categories, and that is a useful limit even when you can have more: with too many, you start reading the list of categories and stop seeing the prompts.
A prompt saved from the message box lands in General. Treat General as the inbox, and move a prompt to its real category when you give it a proper title.
Tags: for what cuts across categories
Tags carry what a category cannot: the client, the output format, the medium (image, video), the stage of a project. A prompt can have several, and one tag can run through all your categories.
Two or three tags per prompt is enough. A tag you have used once does not help you find anything, so reuse the ones you have before adding a new one. Mark the prompts you use every week as favorites: Spotlight shows favorites when you open it with nothing typed, and it has a Favorites filter of its own.

Variables: one template instead of five copies
Most duplicates are the same prompt saved again with a different product, audience or tone pasted in. Keep one copy and mark what changes with curly braces:
“Write a launch tweet for {product}. Lead with the benefit for {audience} and keep it under 240 characters.”
When you insert a prompt like this, PromptTray asks for those words in a short form and puts the finished text in the message box. That works on every plan. Premium also keeps a set of answers as a named preset, for the client or product you write for most.

A rule of thumb: two prompts that differ by a few words are one prompt with variables.
Versions: change the prompt, do not save a second one
When you find better wording, edit the prompt you have. A second copy called “v2” means that six months later you have to work out which one you trusted.
The worry is losing wording that worked. PromptTray Premium keeps earlier versions of each prompt, so you can go back to one. Without history, keep one test copy with “(test)” at the end of its title, use both for a week, then delete the one that lost.
An example library
Twelve prompts in four categories, each one clear from its title alone. General stays empty between saves:
- Writing: “Rewrite for clarity”, “Summarize meeting notes into action items”, “Draft a reply to {customer}”
- Marketing: “Write a launch tweet for {product}”, “Landing page hero generator”, “Create a content calendar”
- Coding: “Debug React state issue”, “Review code for risks”, “Write test cases for {function}”
- Design: “Cinematic shot of {place}”, “Critique a product screen”, “Write UX microcopy options”
Tags do the rest. A “launch” tag runs through Marketing and Writing, and “image” and “video” mark what goes to an image or a video tool.
Prune once a month
Ten minutes, three passes. Delete what you have not used since the last prune. Merge near-copies into one prompt with variables. Check that every category still passes the two tests.
The same care helps when you bring prompts in from a doc. PromptTray’s Import Hub splits pasted text, or a .md, .txt, .json or .csv file, into prompts and flags the ones you already have, so the library starts without duplicates.
If you work in more than one AI app, the guide to managing prompts across AI tools covers what stays the same in each app and what changes.
FAQ
How many categories should a prompt library have?
A handful: four to six for most people. With many more, you read the list of categories before you get to a prompt. Name them after kinds of work, and keep one, such as General, as the inbox for prompts you have not filed yet.
Should I organize prompts by AI tool?
No. A good chat prompt works in ChatGPT, Claude and Gemini with little or no change, so a folder per tool leaves you three copies to maintain. Organize by job, and tag the medium (image, video) where it matters.
Are tags better than folders?
They do different jobs. A prompt sits in one category, which says what kind of work it is for. Tags mark what cuts across categories, such as a client or a format, and a prompt can have several.
How do I stop saving duplicate prompts?
Use variables for the parts that change, edit a prompt in place when you improve it, and prune once a month. Most duplicates are one prompt with a different name or product pasted in.
Which prompts are worth saving?
The ones you have used twice. A prompt written once for a one-off task is history, not a library, and saving everything is how a library stops being searchable.