Prompt Quality Guide

AI Prompt Quality Checklist

A practical checklist for reviewing whether an AI prompt is specific, useful, safe, and likely to produce a usable answer.

3 min read
Prompt Engineering prompt quality prompt checklist ai output review

The problem

Most weak prompts do not fail because the wording is bad. They fail because the task is underspecified.

A useful prompt gives the AI a job, context, constraints, examples, and a way to judge whether the answer is good. Without those pieces, the output often sounds fluent but remains hard to use.

Use this checklist before you publish a prompt, save it to a prompt library, or rely on it in a workflow.

The 7-part checklist

1. Job to be done

The prompt should say what the AI is being asked to accomplish, not just what role it should play.

Weak:

You are a marketing expert. Write a campaign.

Better:

Create a 4-email onboarding campaign for trial users who signed up but have not completed setup.

Ask:

  • What output should exist at the end?
  • Who will use it?
  • What decision or action should it support?

2. Real context

Context should include facts that change the answer.

Useful context includes:

  • audience
  • product or project
  • source material
  • current stage
  • constraints
  • examples of what has worked or failed

Avoid adding long background that does not affect the output. More context is not always better. Relevant context is better.

3. Constraints and boundaries

Good prompts say what the AI should not do.

Examples:

  • Do not invent customer quotes.
  • Do not use claims that require legal review.
  • Keep the answer under 500 words.
  • Use only the facts provided below.
  • If information is missing, ask questions before drafting.

Boundaries make the output easier to trust.

4. Output format

If the output has to be used in a real workflow, specify the format.

Useful formats:

  • table
  • checklist
  • email draft
  • JSON schema
  • decision memo
  • issue list
  • step-by-step plan

Format matters because it controls how much editing the user has to do after the AI answers.

5. Quality criteria

Tell the AI how the answer will be judged.

For example:

A good answer is specific, avoids unsupported claims, includes tradeoffs, and gives a clear next action.

Quality criteria prevent polished but shallow answers. They also make it easier to run a second critique pass.

6. Examples

One example usually improves a prompt more than another paragraph of instructions.

Examples can show:

  • tone
  • level of detail
  • structure
  • terminology
  • what to avoid

If you have a strong prior output, include it as a style reference and say what makes it good.

7. Review step

For important work, add a review step directly into the prompt.

Copy-ready review instruction:

Before finalizing, review your answer for unsupported claims, missing assumptions, vague recommendations, and places where more user context would improve the result.

This turns the AI from a drafter into a reviewer of its own answer.

Copy-ready checklist prompt

Review this prompt before I use it.

Check whether it includes:
1. A clear job to be done
2. Relevant context
3. Constraints and boundaries
4. A specific output format
5. Quality criteria
6. Examples or style references
7. A review or critique step

Return:
- What is strong
- What is missing
- The top 3 changes to make
- A rewritten version of the prompt

Prompt:
[PASTE PROMPT]

Common mistakes

  • Asking for a role instead of a deliverable
  • Giving broad goals without source material
  • Forgetting to say what the AI should avoid
  • Requesting “best practices” when you need a decision
  • Saving prompts that only work with one hidden context
  • Treating the first answer as final