Critique Systems 22 min

Rubric-Led Review

Use AI to review AI output against measurable quality rules.

Learning Objectives

Map the source packet needed to review AI output against measurable quality rules.
Write a role-specific meta prompt for rubric scorecard.
Separate facts, assumptions, missing inputs, and risks before using AI output.
Run a review pass before moving the answer into the next workflow step.

Why this matters

Before AI can create useful work, it needs a controlled context packet. Your job is to define the source, audience, decision, and quality rules behind review AI output against measurable quality rules. This prevents the model from guessing facts, tool data, customer details, or business results. A strong meta prompt turns messy workplace context into a repeatable AI workflow.

Workplace scenario

You are handling a real advanced prompting workflow, not practicing a generic prompt exercise. The work starts with draft output and quality standards, moves through AI reasoning, and should end as rubric scorecard. The learner's job is to control what AI is allowed to use, what it must ignore, and how the final answer will be reviewed.

This lesson trains the habit behind durable AI work: do not ask for the answer first. Ask AI to understand the context, identify missing inputs, separate facts from assumptions, and only then produce the work product.

Input packet

Source material: draft output and quality standards.
Business purpose: Explain why this output matters and who will use it.
Decision supported: State whether the output supports review, follow-up, prioritization, planning, communication, or execution.
Quality rules: Name what would make the answer unsafe, misleading, off-brand, incomplete, or hard to use.
Destination: Specify where rubric scorecard goes next: a document, meeting, CRM, ticket, message, brief, or review workflow.

Decision framework

Role fit: Should AI behave as an AI rubric designer, a reviewer, a strategist, or an editor?
Source control: What exact material is allowed, and what should be ignored?
Assumption control: Which facts are confirmed, inferred, unknown, or risky?
Output contract: What format, length, tone, structure, and fields should the answer follow?
Human review: What must a person verify before the output becomes real work?

Meta workflow stages

Define the AI role: Tell ChatGPT, Claude, Gemini, Manus, xAI/Grok, or your internal AI assistant to act as a AI rubric designer, not a generic assistant.
Request the context packet: Ask for draft output and quality standards, audience, deadline, and quality rules before any output.
Control assumptions: Make AI separate confirmed facts, inferred assumptions, missing inputs, and risks.
Shape the output: Request rubric scorecard in a format you can review, send, or paste into the next tool.
Run quality checks: Verify source use, tone, completeness, and hallucination risk before using the answer.

Common mistakes

Asking for the final deliverable too early: The model fills gaps with plausible but unsupported content.
Providing unlabeled source material: AI cannot tell what came from which tool, meeting, customer, or workflow.
Skipping the audience: The same output may need different tone and depth for a manager, teammate, customer, or executive.
No review rule: Without a checklist, users treat fluent output as correct output.

Quality bar

A strong answer should show what source material was used, what was missing, what assumptions were made, and what the final user should do next. If the output cannot be reviewed quickly, pasted into the destination tool, or explained to a stakeholder, the meta prompt is not finished.

Advanced variation

After the first answer, run a second pass: ask AI to critique its own output against source fidelity, audience fit, missing context, tone, and next action. Then ask it to rewrite only the weak parts instead of regenerating everything from scratch.

What good looks like

A strong meta prompt gives AI a role, asks for the right inputs first, controls assumptions, defines the output format, and ends with a review checklist. By the end of this lesson, you should be able to review AI output against measurable quality rules without starting from a blank prompt.

Examples

Avoid

Weak task prompt

This is intentionally weak: it gives the AI no role, no source rules, no audience, no output format, and no hallucination guard.

Help me with rubric-led review.

Context

Use this as the main working prompt for this lesson. It gives the AI a role, asks for the right source material first, controls assumptions, and defines the output before any answer is produced.

Meta Prompt

Strong meta prompt

Act as an AI rubric designer for Advanced Prompting.
Before answering, ask me for: [DRAFT_OUTPUT_AND_QUALITY_STANDARDS], [AUDIENCE], [DEADLINE], and [QUALITY_RULES].
Use only the source material I provide.
Separate confirmed facts, assumptions, missing inputs, and risks.
Do not invent tool data, metrics, customers, or results.
Return rubric scorecard plus next action and review checklist.

Outcome: a reusable AI instruction pattern you can adapt with your own role, source material, audience, and quality rules.

Context

Use this after the first AI draft exists and the output needs to be shaped for a manager, stakeholder, or decision-maker. It keeps the same lesson focus, "Rubric-Led Review", but forces the AI to prioritize decisions, risks, and next actions instead of generic explanation.

Meta Prompt

Manager-ready variation

Act as an AI rubric designer preparing work for a manager.
Ask for [SOURCE_MATERIAL], [MANAGER_AUDIENCE], [DECISION_NEEDED], and [CONSTRAINTS].
Turn the context into rubric scorecard.
Keep it concise, evidence-based, and action-oriented.
Separate facts, assumptions, risks, and recommended next step.
Do not add details that are not in the source.

Outcome: a reusable AI instruction pattern you can adapt with your own role, source material, audience, and quality rules.

Context

Use this as the verification layer after the AI has produced an answer. It teaches the AI to compare the draft against source material, flag unsupported claims, and rewrite only the weak parts.

Meta Prompt

Review pass

Act as a quality reviewer for rubric scorecard.
Review the draft against: source fidelity, missing context, audience fit, tone, and next action.
Use only [SOURCE_MATERIAL] and [DRAFT_OUTPUT].
List unsupported claims, unclear assumptions, and risky wording.
Then rewrite only the weak parts.
End with a pass/fail checklist.

Outcome: a reusable AI instruction pattern you can adapt with your own role, source material, audience, and quality rules.

Practice Exercise

Run the Rubric-Led Review workflow

Use ChatGPT, Claude, Gemini, Manus, xAI/Grok, or your internal AI assistant. Start with the compact meta prompt, provide real or sample source material for draft output and quality standards, then refine the result until it is ready for work.

The AI asked for the right source material before producing the answer.
The output separates facts from assumptions.
The final deliverable is in a usable format.
You completed the review checklist before treating the answer as final.

Stress-test the Rubric-Led Review output

Paste the AI output back into ChatGPT, Claude, Gemini, Manus, xAI/Grok, or your internal AI assistant with the review prompt. Ask it to find unsupported claims, missing inputs, weak structure, and places where a human must verify the answer.

Unsupported claims are marked clearly.
Missing context is listed as questions.
The revised output keeps only source-grounded details.
The next action is clear enough for another teammate to follow.

Mini Prompt Templates

Context

Run this after completing "Rubric-Led Review" to check whether the AI output is actually usable at work. It is a review prompt, not a generation prompt.

Meta Prompt