You have already written your best prompts — then lost them in chat history. This playbook captures your ten most-repeated work prompts into a reusable, example-enriched library in Notion or a doc: the compounding asset that makes every other playbook a fifteen-second paste instead of a five-minute rewrite.
Workflow area: AI Fluency · Tier: Foundation · Hands-on time: ~30 minutes
The difference between someone who "uses AI sometimes" and someone visibly faster at their job is rarely prompting genius — it is that the second person never writes the same prompt twice. A prompt that produced a great weekly report once will do it again every week, forever, if you save it. This playbook is thirty minutes of setup that pays back every working day.
Mine your own history. Open your ChatGPT/Claude/Copilot history and scroll the last two weeks. List every task you prompted more than once. Typical office top-ten: email from bullets, thread summary, tone rewrite, meeting minutes, status update, formula explanation, document summary, translation, agenda, job-ad or post rewrite.
Pick a home with four fields. Notion database, OneNote page, Google Doc with headings — anything you open daily. Each entry gets: Name (what task), Prompt (paste-ready with [PLACEHOLDERS]), Example (one real input/output pair that worked), Notes (which tool, what to tweak).
Convert your best chats into template entries. For each repeated task, find the chat where the output was actually good, and generalize it — or make the model do the generalizing:
Below is a prompt I wrote and the output I liked. Turn my prompt
into a reusable TEMPLATE:
- Replace the specifics with [BRACKETED PLACEHOLDERS] and keep
everything that made it work (the role, rules, format demands).
- List the placeholders with a one-line hint for each.
- Keep it under 150 words.
My prompt:
[PASTE YOUR PROMPT]
The good output (for reference, do not template this):
[PASTE THE OUTPUT]
A finished entry looks like:
NAME: Weekly status update
PROMPT: Turn my work log into a status update for [AUDIENCE].
Sections: Done / In progress / Blocked / Next week. Max 150 words.
Flag risks explicitly. Do not invent progress — if the log is
thin, the update is thin.
Log: [PASTE LOG]
EXAMPLE: (link to the chat that produced the good one)
NOTES: Works in ChatGPT + Claude. For Copilot, add "use Teams
messages from this week" instead of pasting the log.
Enrich the big three with few-shot examples. For your three most-used prompts, paste one or two real, good outputs (your edited final versions, not the model's raw draft) into the template as "match the style and structure of this example." Few-shot examples are the single biggest quality lever you control — they carry your voice into every future output.
Adopt the one-in rule. Every time a fresh prompt produces something great, spend sixty seconds templating it into the library before closing the tab. Once a month, delete entries you have not used — a ten-entry library you trust beats a fifty-entry junk drawer.
A prompt library is also a privacy control — templates with [CUSTOMER NAME] placeholders remind you to genericize before pasting. Write the placeholder, not the data, into every saved entry; pb-12 covers what must never be pasted at all.
The library is the compounding asset: every playbook in this track ends with a reusable prompt, and this is where they live. Ten trusted templates, refreshed monthly, will outperform any amount of ad-hoc prompting skill.