AI prompt examples that make the task clear.

A simple prompt checklist

1. Task: What should the AI produce or change?
2. Context: Which source material, audience, or situation matters?
3. Constraints: What facts, limits, or exclusions must survive?
4. Output: Would a paragraph, table, checklist, or code change be most useful?

Not every prompt needs all four. “Fix the spelling in this sentence” can remain a short instruction. Extra roles and elaborate formatting are useful only when they help with the actual task.

1. Write a customer email

Rough request: reply customer package late say sorry dont promise refund need order number under 100 words

Write a reply to a customer whose package is late. Apologize, ask for their order number, and do not promise a refund. Keep the reply under 100 words.

The request names the situation, the next action, and the promise to avoid. A clearer prompt should not invent a refund policy, delivery date, or order number.

2. Plan a project launch

Rough request: launch atlas 12 october plan milestones owners risks keep short

Create a concise launch plan for Atlas on 12 October. Include milestones, owners, and launch risks.

The product name and date survive the rewrite. If team members are not provided, the AI should identify roles or ask for details rather than present invented people as assigned owners.

3. Compare two options

Rough request: compare these two options cost setup reliability maintenance table recommend main tradeoff

Compare these two options in a table covering cost, setup time, reliability, and maintenance. Recommend one and explain the main trade-off.

Explicit criteria make the comparison easier to check. Supply the actual options and evidence with this prompt; clearer wording cannot fill missing facts.

4. Summarize a document

Rough request: summarize this for manager start decision then 3 risks next actions use only doc

Summarize this document for my manager using only information from the document. Lead with the decision, then list three risks and the next actions.

The audience, order, and source boundary are explicit. If the document does not contain a decision or three risks, ask the AI to say what is missing instead of inventing it.

5. Ask for a code fix

Rough request: fix this typescript bug explain cause smallest change no new deps dont change public api add regression test

Fix the bug in this TypeScript code. Explain the cause, make the smallest necessary change, preserve the public API, and add a regression test. Do not add dependencies.

The constraints define an acceptable fix. Include the relevant code, error, and expected behaviour before submitting; a prompt improver should not guess them.

6. Research a decision

Rough request: research options for eu hosting small nextjs app pricing region limitations official sources current links flag unknowns

Research EU hosting options for a small Next.js app. Compare current pricing, available regions, and limitations using official sources. Include source links and flag anything you cannot verify.

Freshness and source expectations are stated. Use an AI tool with browsing access and check its links; a well-written request does not guarantee accurate research.

Improve the instruction, then review the result

Before sending, check every name, date, quantity, link, and exclusion. A polished sentence that changes a deadline or drops “do not add dependencies” is a worse prompt.

Rewordy’s AI Prompt action rewrites the instruction you select on your Mac. It does not answer the request. Supply missing context yourself, then submit the improved prompt to your chosen AI tool. Use Refine when you are polishing the final message for a person.