Good AI prompts are not magic phrases. They are clear briefs: context, task, constraints, source material and a definition of a useful result. The templates below are starting points—replace bracketed text with real information and verify important output before using it.
A simple prompt structure
Use five parts: role/context → task → inputs → constraints → output format. Add examples when tone or structure matters.
15 prompts worth saving
- Customer reply: “Draft a concise reply to [customer issue]. Use only the facts below. Do not promise anything not stated. End with the next action.”
- Proposal outline: “Turn these discovery notes into a proposal outline with problem, scope, deliverables, assumptions, exclusions and next steps.”
- Meeting actions: “Extract decisions, owners, deadlines and unresolved questions from these notes. Do not invent missing owners.”
- Research plan: “Create a research checklist for [decision]. Separate facts we need, sources to consult and assumptions to test.”
- Competitor comparison: “Using only the supplied source material, compare [A/B/C] by [criteria]. Mark unknowns explicitly.”
- FAQ draft: “Turn these real customer questions into an FAQ. Keep answers factual and flag anything needing policy/legal review.”
- Product description: “Draft a clear description from these verified features. Avoid superlatives and unsupported claims.”
- Social variations: “Create three platform-appropriate drafts from this approved source text; preserve the facts and CTA.”
- Email subject lines: “Generate ten accurate subject lines for this email. No fake urgency or misleading claims.”
- Process checklist: “Convert this procedure into a numbered checklist with prerequisites, actions, validation and rollback points.”
- Risk review: “List ambiguities, dependencies and failure modes in this plan. Do not rewrite it yet.”
- Plain-English rewrite: “Rewrite for a non-technical reader without removing important conditions or caveats.”
- Data summary: “Summarise the supplied table. Separate observations from interpretations and identify missing context.”
- Interview questions: “Create questions that test [skill/outcome], with evidence to listen for. Avoid questions about protected characteristics.”
- Quality check: “Review this draft against [criteria]. Quote the exact section causing each issue and suggest a correction.”
What not to put into a prompt
Do not upload confidential, regulated or customer data until you understand your organisation’s rules and the service’s current data controls. Redact or use synthetic examples when testing.
Make prompts repeatable
Save the prompt beside its acceptance checklist and a good example. A repeatable workflow is more valuable than a clever one-off prompt.
Next: see our small-business AI shortlist and compare ChatGPT vs Claude.
