Typing prompts from scratch each time works, and it never gets better. A saved library turns every good prompt into something you keep — the difference between using AI and having a way of working.
What every reusable prompt needs
Four parts, in this order. Miss one and the results become unpredictable.
- The job. One line saying what you want done.
- The input. Clearly marked off, with a gap to fill in each time —
<PROFILE>,<NOTES>,<JOB>. - The rules. What it must not do. No guessing, no invented facts, quote the
source, write
not statedwhen something is missing. This part is the difference between a prompt that works and one that produces convincing fiction. - The output format. Exactly what should come back — a search string, a table, five bullets. If you do not say, you will reformat it by hand every time.
What to cover
| Area | Prompts worth having |
|---|---|
| Profiling | Meeting notes to job profile; job brief to scorecard items |
| Widening | Job title variations; tool and skill sets; target companies |
| Searching | Requirements to a search string; search string to X-Ray; fixing a search that fails |
| Screening help | Pulling quotes as evidence; questions to ask one specific candidate |
| Outreach | First message from facts; follow-ups; rejections by stage |
| Summarising | Themes across a list; interview notes into scorecard evidence |
Twenty prompts covers all of that comfortably. A hundred is a library nobody searches, and finding the right one costs more than writing it again.
Test it before you save it
Five cases where you already know the right answer. For widening prompts, five jobs whose alternative titles you know. For extraction prompts, five documents you have read.
If it fails on two out of five, it is not ready — and fixing it now is much cheaper than finding out across forty candidates.
Keep the test cases with the prompt. They are how you check it still works after the AI gets updated, which is the most common reason a prompt that worked stops working.
Note what you change
Plain text, three things at the top of each prompt: date, what changed, why.
The reason is diagnosis. When results get worse, you need to know whether the prompt changed, the AI changed, or your inputs changed. Without a history you are guessing — and the usual guess, rewriting the prompt, often makes it worse.
Download 20 ready-made prompts · all templates
Start with three
Do not sit down to write twenty. Write the three you have already typed from memory more than twice this month. Libraries that get used grow out of repetition, not planning.