Claude / Getting Started & Prompt Design
Claude: evaluate a response against a small rubric
Define three observable criteria before generating a draft: factual coverage, audience fit and an actionable next step.
Environment and inputs: Claude user guide · Feature and source access depend on the active account, app and organization settings. Optional file or app steps have a text-excerpt fallback. No fixed model version, price or usage quota is assumed.
What you will make
A practical Claude guide to evaluate a response against a small rubric. Use the sample input and prompt, then check the result. Use a fictional support-document review as the practice setting. Keep passages and your interpretation in separate output fields.
Prepare the source and scope
Define three observable criteria before generating a draft: factual coverage, audience fit and an actionable next step. Open Claude and start a focused conversation, or use the appropriate existing project for recurring work.
Try this prompt
In Claude, provide the prepared input, then send the following request. For real work, replace fictional details with approved facts.
Score this service announcement against the rubric below. Quote the sentence supporting each score, identify one missing fact and suggest a targeted revision. Do not rewrite the whole announcement.Review the result
Check each quoted sentence exists and each score uses the supplied criterion rather than a new preference. Record each correction and its supporting source.
If the answer is wrong or incomplete
An assistant reviewing its own draft may miss the same error twice. Include at least one independent human check. Put the source excerpt and the request in separate labeled blocks. For longer reviews, retain paragraph labels so corrections remain traceable.
Practice and completion check
Repeat with one incomplete input. Check the answer against the review criteria, remove unsupported claims and save the prompt with its source version. This workflow has not been verified in a customer deployment.