Prompt Review Checklist

Use this checklist before you reuse a prompt or act on the output. It is designed for practical workplace prompts: drafting, summarizing, planning, reviewing, research support, and handoff notes.

When To Use It

Use the full worksheet when:

• the prompt will be reused by you or someone else
• the output affects a person, customer, workflow, system, or decision
• the answer includes claims, numbers, dates, sources, owners, or recommendations
• the input contains material that must be sanitized before use
• the first output looked polished but still felt hard to trust

For low-stakes brainstorming, use the compact checklist at the end.

Prompt Setup

Check the prompt before running it.

Task

• Strong: the task says the action to perform, such as draft, summarize, compare, classify, revise, or critique.
• Weak: the task asks for general help without naming the work product.

Prompt check:

• The task is specific enough that success can be recognized.
• The prompt says what the output is for.
• The prompt says what the model should not do.

Context

• Strong: the context includes the facts, source material, audience, and situation needed for the task.
• Weak: the prompt expects the model to guess the background.

Prompt check:

• The context includes only what the model needs.
• Sensitive details are removed, generalized, or replaced with placeholders.
• Missing facts are labeled as missing instead of implied.

Constraints

• Strong: the constraints name limits, review expectations, exclusions, and uncertainty handling.
• Weak: the prompt asks for a good answer but does not define what “good” means.

Prompt check:

• The constraints say what to avoid, preserve, or verify.
• The prompt tells the model how to handle missing facts.
• The prompt keeps consequential decisions with a human reviewer.

Output Shape

• Strong: the prompt asks for a usable structure: sections, bullets, table columns, decision labels, or a checklist.
• Weak: the prompt asks for an answer and leaves the shape to chance.

Prompt check:

• The output shape is clear.
• The output can be reviewed without heavy cleanup.
• The output includes labels for uncertainty, source status, or next owner when needed.

Output Review

Review the answer before editing it into final prose.

• The answer follows the requested task.
• The answer uses provided facts without inventing missing details.
• Claims, numbers, sources, and recommendations are marked for verification.
• The format is usable without manual cleanup.
• The tone fits the audience and situation.
• The output does not hide uncertainty behind confident language.

Score the output:

• Pass: usable after normal human review.
• Partial: useful, but the prompt needs one targeted edit before reuse.
• Fail: unsafe, off-task, unsupported, or too hard to review.

Failure Diagnosis

Name the failure before changing the prompt.

Iteration decision:

• Keep the prompt if the output passed on more than one realistic example.
• Revise the prompt if the same partial failure appears twice.
• Discard the prompt if it repeatedly produces unsupported or unsafe output.

Safety Check

Run this before you paste input and again before acting on output.

• Secrets, credentials, private keys, tokens, or connection strings were not included.
• Personal, customer, employee, or financial details were removed or generalized unless approved.
• Confidential business material was not shared unnecessarily.
• The output does not infer identities from placeholders.
• The output does not invent missing names, dates, numbers, sources, owners, or commitments.
• Any external message, workflow change, customer impact, system action, or consequential decision still has human review.

Stop and ask before using the output if any sensitive detail is needed for the task and you are not sure whether the tool, data, or use case is approved.

Iteration Note

Record one sentence after each meaningful revision:

This prompt worked when [condition], failed because [failure category], and was revised by [specific prompt change]. It still needs review for [risk or gap].

Example:

This prompt worked when the notes included decisions and owners, failed because missing dates were invented, and was revised by requiring “not provided” for missing facts. It still needs review for unsupported claims.

Sign-Off Block

• Prompt owner:
• Reviewer:
• Date:
• Use case:
• Test examples tried:
• Output score: pass / partial / fail
• Decision: reuse / revise / discard
• Residual risk:
• Next review trigger:

Decision line:

This prompt is approved for [use case] when [conditions]. It must not be used for [excluded uses] without additional review.

Compact Copyable Checklist

• Check: the task is specific enough that success can be recognized.
• Check: the context includes only what the model needs.
• Check: sensitive details are removed, generalized, or approved for this tool and use case.
• Check: the constraints say what to avoid, preserve, verify, or label as missing.
• Check: the output shape is easy to review.
• Check: I tested the prompt on at least one realistic example.
• Check: I scored the output as pass, partial, or fail.
• Check: I named the failure category before editing the prompt.
• Check: claims, numbers, dates, sources, owners, and recommendations are marked for verification.
• Check: a human reviewer still owns any external action or consequential decision.