Prompts You Reuse

Stop Rewriting the Same Prompt

Most people prompt from scratch every time, including for tasks they do weekly. Every Monday they rewrite roughly the same request for a status summary, get roughly the same mediocre answer, and edit it in roughly the same way. The alternative takes twenty minutes once. Take a task you repeat, write the best prompt you can for it, use it a few times, and each time you have to correct the output, fold that correction back into the prompt permanently. After four or five rounds you have something that produces usable output on the first try. This is the difference between using AI and having a small set of tools you built.

• Pick tasks by frequency, not by difficulty — weekly beats impressive
• Every correction you make twice belongs in the prompt permanently
• Four or five improvement rounds is usually enough to reach first-try usable
• Keep them somewhere you will actually find them, not scattered in old chats

Build It With Slots

A reusable prompt is a template with obvious gaps. Everything constant lives in the template; everything variable becomes a clearly marked slot. A weekly update prompt: "You are writing a weekly update for [AUDIENCE], who care most about [WHAT THEY CARE ABOUT]. Format: three sections — shipped, in progress, blocked. Each item one line, plain language, no adjectives. Lead with anything that changes a decision. Do not include work that has not started. Raw notes: [PASTE NOTES]." Capital-letter slots are easy to spot and hard to leave unfilled by accident. Keep the slot list short — a template with eleven variables is not a template, it is a form nobody completes.

• Constant parts in the template, variable parts as clearly marked slots
• Use an obvious slot convention such as [AUDIENCE] so blanks are visible
• Three or four slots is usually the practical limit before people stop using it
• Put the pasted material last, under a labelled heading

Test It Like a Tool

A prompt you will rely on deserves a small amount of testing, and this is where a surprising number of team prompt libraries fall over. Run it on three or four real cases, not one — including a hard case and an edge case where the input is messy or incomplete. Because output varies between runs, one good result is not evidence of reliability. Look specifically at how it behaves when information is missing: does it say so, or does it invent something? Adding one line such as "if the notes do not say, write 'not stated' rather than guessing" is often the difference between a template you can trust and one that quietly produces confident fiction on a bad week.

• Test on several real cases including a messy one, not on your best example
• One good run is not evidence — variation between runs is normal
• Check behaviour when input is incomplete; that is where templates fail badly
• Add an explicit instruction for missing information: "write not stated rather than guessing"

Try It Yourself

A prompt that worked once becomes an asset the moment you add slots. Take the best prompt you have written during this course and make it reusable.

ROLE: [who the AI should answer as]
FOR: [who receives the output]
TASK: [the outcome you want, stated as a result]
LIMITS: [length · tone · format]
EXAMPLE: [optional — paste a previous output that was right]