Good prompts stay hidden in personal chat histories, so every person solves the same problem again. A shared library fixes that only if it is small, tested on real work and owned by someone. These four steps collect what already works, make it reliable and keep it current.

Step 1 — Collect what already works

Goal: Find the prompts people actually reuse.

  1. Ask each person for the two or three prompts they return to every week.
  2. Note the task, the input it needs and what a good result looks like.
  3. Group duplicates and near-duplicates into one task each.
  4. Pick the ten tasks that happen most often or matter most.

You finish with: A shortlist of ten recurring tasks with example inputs and outputs.

Step 2 — Standardize and test

Goal: Make each prompt produce consistent results.

  1. Rewrite each prompt with the role, context, steps and output format spelled out.
  2. Remove anything that depends on one person’s chat history.
  3. Test each prompt on three real examples and fix it until results are consistent.
  4. Mark which prompts need approved reference files or a human review.

You finish with: Ten tested prompts with notes on inputs and review.

Step 3 — Publish where the work happens

Goal: Put the library inside the tools people already use.

  1. Store each prompt as a shared project, custom instruction or template in your AI workspace.
  2. Attach approved reference files where a prompt needs them.
  3. Write one line per prompt: what it is for and when not to use it.
  4. Show the team the library in a short walkthrough on their own tasks.

You finish with: A published library the team can find in seconds.

Step 4 — Own and improve it

Goal: Keep the library accurate as work and tools change.

  1. Name one owner for the library and one per team if it grows.
  2. Collect improvements people make and fold the good ones in monthly.
  3. Retire prompts nobody uses.
  4. Retest prompts when the underlying model or tool changes.

You finish with: An owned, current library with a regular review.

Common pitfalls

  • Collecting hundreds of prompts nobody tests.
  • Prompts that only work inside one person’s chat history.
  • A library stored somewhere nobody looks.
  • No owner, so outdated prompts keep being used.

Read as Markdown Configure your AI