How to use AI to critique and stress-test your work

Executive overview

Most people ask AI to polish their work, not challenge it. A structured critique prompt forces AI to find what's incorrect, unclear, biased, or incomplete — then propose fixes.

A named, persistent critique agent acts as a mandatory second opinion before anything leaves your desk.

The core critique prompt

• Instruct AI to review for: incorrect, weakly supported, unclear, incomplete, biased, impractical, verbose, or improbable content
• Add a corrective step: "propose better alternatives and provide a revised version"
• Customise the criteria to your context — policy compliance, CFO lens, operating procedures

How Charles works (the full system)

• First step: AI infers what you were actually trying to achieve
• Quality gates: factual accuracy, logical soundness, completeness, relevance, clarity, structure, safety, practicality
• Diagnostic output: each issue rated by severity and confidence
• Fixes: problem type, location, and specific remediation suggested
• Alternative approaches offered for flagged sections

When and what to use it for

• Anything you'd want a second opinion on: emails, documents, reports, slide decks, social posts, analysis
• Works even when AI generated the original output — it will still find flaws
• Treat it like spell check: run it before anything goes to a colleague or stakeholder
• You retain judgment — accept, reject, or override suggestions as context demands