Somebody owning it and something recording which entries have actually been used. Ownership supplies the pruning nobody volunteers for, and use is what a reader is really asking about when they open it. A collection with neither becomes an archive within a couple of quarters.

Where should it live?

Wherever the work already happens, which is usually a workspace or the automation platform itself. A separate destination is one more thing to remember, and the collections that stay maintained are consistently the ones that did not require anybody to go somewhere special.

Quietly and predictably. Entries accumulate faster than review, near-duplicates appear because nobody found the original, and confidence falls until people stop looking. Nothing breaks and no one announces it; the collection simply stops being opened, and the knowledge it held returns to individual notes. That is where it started, which is why the failure is so easy to miss.

Glossary

Prompt library

A prompt library is the shared collection a team draws on, which is what turns a good way of asking from one person's habit into something the organisation holds.

In plain terms

The shared place the good ones live. Somebody works out a reliable way to ask for something, and instead of it staying in their notes, it goes somewhere everyone can find it. That is the whole idea, and it fails in the ordinary way shared collections fail.

Why it matters

Because otherwise the capability leaves when the person does. A team where three people each have their own private way of asking has three quality levels and no way to improve any of them, and the version that worked best is the one nobody else has seen.

How it works

The collection is the easy part and the arrangement around it is not. Somewhere to put things takes an afternoon. What decides whether it survives is whether anybody owns it, whether entries are found when needed, and whether people know which ones have actually been used in earnest rather than merely added.

Decay is the normal outcome and it has a recognisable shape. Entries accumulate faster than they are reviewed, several near-duplicates appear because nobody found the existing one, and the collection becomes something people search once, fail to find what they want, and stop returning to. At that point it is an archive rather than a library.

Provenance matters more than volume, which is the counter-intuitive part. Ten entries somebody has actually relied on are worth more than a hundred contributed hopefully, because the reader's real question is whether this worked for somebody, not whether it exists. Collections that record use rather than just contents stay useful for longer.

It sits naturally where the work already is rather than in a place of its own. Teams keep these in the workspace they document in, or as saved steps inside an automation platform, precisely because a separate destination is one more thing to remember. The best-maintained ones are usually the least ceremonious.

A folder, and a library

The failure here is so consistent that it is worth treating as the default rather than the risk. A team recognises that good ways of asking are being lost, creates a shared place, and contributions arrive enthusiastically for a few weeks. Nothing then removes anything. Entries that stopped behaving well after a version change stay; near-duplicates accumulate because the search did not surface the original; and the ratio of things worth using to things merely present falls steadily. The moment that decides the outcome is unglamorous and passes unnoticed: somebody opens the collection, does not find what they need, writes their own, and does not add it. From there the collection is an archive of what people used to do, and the knowledge it was built to hold is back in individual notes, which is where it started.

Seen in the wild

A workspace assistant answering from your own content, where the shared way of asking lives beside the documents it is used on.
Notion AI
Saved steps inside an automation platform, where a way of asking is reusable because it is part of a scenario rather than a note somewhere.
Make
A knowledge platform that detects conflicting content, merges duplicates and flags stale material, which is the maintenance problem this collection has.
Guru

Common misconceptions

People assume

The hard part is collecting them.

In fact

Collecting is an afternoon. Keeping the collection worth opening is the standing work: pruning duplicates, retiring entries that no longer behave, and making the good ones findable. A collection nobody prunes becomes a place people search once and abandon.

People assume

More entries make it more useful.

In fact

Past a point they make it less useful, because finding the right one gets harder and confidence in any of them falls. What a reader actually wants to know is which entries somebody has relied on in earnest, and volume dilutes that signal rather than adding to it.

Telling them apart

Prompt library vs Prompt template

Prompt library

The collection, with ownership, findability and pruning.

Prompt template

One reusable request with its variable parts marked.

The template is the artefact; the library is the arrangement that keeps it current and findable. Most teams build the first and skip the second.

Questions

What makes one last?: Somebody owning it and something recording which entries have actually been used. Ownership supplies the pruning nobody volunteers for, and use is what a reader is really asking about when they open it. A collection with neither becomes an archive within a couple of quarters.
Where should it live?: Wherever the work already happens, which is usually a workspace or the automation platform itself. A separate destination is one more thing to remember, and the collections that stay maintained are consistently the ones that did not require anybody to go somewhere special.
How does one fail?: Quietly and predictably. Entries accumulate faster than review, near-duplicates appear because nobody found the original, and confidence falls until people stop looking. Nothing breaks and no one announces it; the collection simply stops being opened, and the knowledge it held returns to individual notes. That is where it started, which is why the failure is so easy to miss.

Key takeaways

The collection is easy; the arrangement around it is the work.
Decay is the normal outcome, and it looks like duplicates plus falling confidence.
Provenance beats volume: what a reader wants is what somebody relied on.
The best-maintained ones sit where the work already is.

Tools that use this

Notion AI
The shared way of asking living beside the documents it is used on.
Make
Saved steps inside a scenario rather than notes kept somewhere else.
Guru
Duplicate merging and stale-content flagging, the same maintenance problem.