How to build and organise your own prompt library
If you use AI chat tools often, you end up retyping the same kinds of request. A prompt library is a collection of tested prompts with blanks you fill in. It saves time and gives more consistent answers.
What makes a prompt reusable
A one-off prompt mentions the details of today’s task. A reusable prompt separates the fixed instructions from the parts that change, and marks those parts as placeholders in square brackets.
One-off: “Summarise this article about watercolour paper for my blog readers.”
Reusable: “Summarise the text below for [audience]. Use [number] bullet points. Keep the reading level simple. Do not add facts that are not in the text. Text: [paste here]”
Five parts of a template
- Role or context: who the answer is for or what situation you are in.
- Task: one verb that says what to do: summarise, rewrite, compare, list, explain.
- Input: the material, clearly separated, for example after a line that says “Text:”.
- Constraints: length, tone, reading level, what to avoid.
- Output format: a table, bullet points, a short email, plain paragraphs.
Starter templates
- Explain simply: “Explain [topic] to someone with no background in it. Use one everyday analogy and keep it under [number] sentences.”
- Compare options: “Compare [option A] and [option B] for [purpose]. Make a table with rows for cost factors, effort, risks and when each is the better choice. Say what you are unsure about.”
- Feedback on a draft: “Read my draft below. List the three biggest problems with structure or meaning, in order of importance. Do not rewrite it. Draft: [paste]”
- Turn notes into steps: “Turn these rough notes into a numbered checklist. Keep my wording where possible. Notes: [paste]”
- Generate questions: “I am preparing to [situation]. List [number] questions I should ask, grouped by theme.”
How to store and label them
Any tool that holds plain text works: a notes app, a document, or a folder of Markdown files. For each entry, keep:
| Field | Example |
|---|---|
| Name | Draft feedback, structure only |
| Use when | Before editing a long post |
| Template | The full text with [placeholders] |
| Last tested | A note of the tool you tested it with |
| Notes | “Works better if I paste one section at a time” |
Group entries by job (writing, research, planning, code) rather than by tool, since the same template usually works across several tools.
Testing a template
- Run it on three different inputs.
- Check each answer for invented facts, wrong tone and ignored constraints.
- Tighten the wording where the tool drifted, then run it again.
- Only add it to the library once it behaves on all three.
Common mistakes
- Saving prompts you used once and never checked.
- Placeholders that are unclear later, such as [thing]. Name them: [product name].
- Pasting private or confidential material into a tool without checking its data policy.
- Trusting the output because the prompt is good. Always review the answer.
Questions
Do prompts need special keywords?
No. Clear, plain instructions work better than tricks. Say what you want, for whom, and in what format.
Why does a saved prompt behave differently over time?
The underlying models change. Re-test your most used templates now and then and adjust them.