Devil's Advocate & Resynthesis

A meta-prompt technique for self-critiquing and improving AI responses

Quality Through Self-Critique

This advanced prompting technique instructs the AI to act as its own critic, identifying weaknesses in its initial response and then producing an improved version. It's like having a built-in editorial review process that catches errors, fills gaps, and enhances clarity.

The Complete Meta-Prompt

Add this to the end of any prompt where quality is critical:

Role:

You are a meticulous reviewer. Your job is to critique the answer you just produced, then rewrite it to be stronger.

Task:

Devil's Advocate Critique — challenge your previous answer. Identify:

  • Logical gaps, weak assumptions, missing alternatives
  • Outdated or uncertain facts
  • Ambiguity, overgeneralization, or unexplained jargon
  • Missing edge cases, risks, or counter-examples
  • Places where evidence or citations are needed

Resynthesize — rewrite the answer so it addresses every issue you found. Improve clarity, correctness, and usefulness. Where applicable, note uncertainties and provide sources or concrete examples.

Rules:

  • Be candid and specific; don't be defensive
  • If you can't verify a claim, flag it or reframe it
  • Prefer concise, actionable language over fluff
  • Keep the original user's intent and constraints

Output format (use exactly these headings):

Critique (Devil's Advocate):

[bullet list of weaknesses]

Revised Answer:

[the improved answer, self-contained]

What Changed & Why (1–3 bullets):

[brief justification of key improvements]

Quality bar:

The Revised Answer should be strictly better—more accurate, clearer, and more useful—than the original.

How This Technique Works

1

Initial Response Generation

The AI first produces its best answer to your question using standard reasoning and knowledge.

2

Critical Analysis (Devil's Advocate)

The AI then switches roles to become a tough reviewer, actively searching for problems:

What Gets Scrutinized:

  • • Logical gaps: Are there flaws in reasoning or unsupported leaps?
  • • Weak assumptions: Are claims based on shaky foundations?
  • • Missing alternatives: Did it ignore other viable approaches?
  • • Outdated facts: Is information current and verifiable?
  • • Ambiguity: Is the language clear and specific enough?
  • • Edge cases: What scenarios or exceptions weren't considered?
3

Resynthesis & Improvement

Armed with the critique, the AI rewrites its answer to address every identified issue, producing a strictly better response.

Improvements Include:

  • ✓ Filling logical gaps with supporting reasoning
  • ✓ Adding caveats or alternative perspectives
  • ✓ Clarifying ambiguous language
  • ✓ Adding concrete examples or evidence
  • ✓ Flagging uncertainties honestly

Example Output Format

Critique (Devil's Advocate):

  • • The original answer assumed all users have high-speed internet, ignoring rural/developing market scenarios
  • • Didn't address the cost implications of the recommended approach for small businesses
  • • Used technical jargon ("API endpoints," "serverless architecture") without explanation
  • • Failed to cite sources for the "70% of companies" statistic

Revised Answer:

[A completely rewritten answer that addresses all the critiqued issues — more comprehensive, clearer, better supported with evidence, and acknowledging limitations where appropriate]

What Changed & Why:

  • • Added section on low-bandwidth considerations to make advice more universally applicable
  • • Included cost-benefit analysis with specific price ranges for different business sizes
  • • Replaced jargon with plain language explanations and added a glossary

Key Benefits

Higher Accuracy

Catches factual errors, weak logic, and unverified claims before they reach you

Better Completeness

Identifies missing perspectives, edge cases, and alternative approaches

Enhanced Clarity

Eliminates jargon, ambiguity, and overgeneralization for clearer communication

When to Use This Technique

High-Stakes Content

Executive reports, client deliverables, published content, legal documents

Complex Analysis

Technical explanations, research summaries, strategic recommendations

Fact-Sensitive Work

When accuracy is critical and errors could have serious consequences

Educational Content

Training materials, documentation, tutorials where clarity is paramount

Build Quality Assurance Into Your AI

Implement self-critique mechanisms for consistently high-quality AI outputs

Schedule a Consultation Back to Home