Everyone who uses AI daily has had this moment. You spent twenty minutes refining a prompt, it produced exactly the right output, and you moved on. Weeks later you need it again, you cannot find it, and the version you reconstruct from memory is measurably worse.
The instinct is to blame yourself for not saving it. Usually you did save it — into a notes app, where it is now indistinguishable from a shopping list.
Why notes apps fail at this specific job
A notes app is optimised for retrieval by memory: you search for a word you remember writing. Prompts break that assumption, because you do not remember the words. You remember the job — 'the one that made product descriptions sound less like a brochure'.
Searching for 'product description' returns nothing useful, because your prompt never contained that phrase. It contained instructions about tone, length, and what to avoid. The thing you can recall and the thing you stored have no overlap.
What a prompt library needs that a notes app lacks
Retrieval by task rather than by content. That means a category or tag describing what the prompt is for, attached at the moment you save it, when you still know why it matters.
Copy-readiness. A prompt you have to reformat before using is a prompt you will rewrite instead. It should come out exactly as it went in.
Room for your own, not just a curated set. Collections of somebody else's prompts are a starting point. The ones that actually earn their place are the ones you refined against your own work.
A tagging scheme that survives contact with reality
Keep it shallow. Two levels at most: what the prompt does, and where you use it. 'Writing / email', 'image / product', 'code / refactor'. Deep hierarchies feel organised for about a month and then collapse, because you stop being able to guess which branch you filed something under.
Name prompts by outcome, not by technique. 'Rewrite in plain English' is findable. 'Chain of thought v3' is not, three weeks later.
Save it when it works, not later
The single habit that matters: the moment a prompt produces the output you wanted, save it immediately with a tag. Not at the end of the session, not when you tidy up on Friday. The window where you can still explain why it worked is about thirty seconds wide.