Ask AI to evaluate its own output, then improve it. A simple loop that polishes writing.
Same writing task. One takes the first draft. One asks for self-critique.
AI writes a reasonable email. It's fine. Generic professional tone, gets the job done, but nothing special.
Acceptable, but could be better.
Draft: [Initial email]
Critique: "The opening feels abrupt. The reason given is vague. It lacks a concrete alternative..."
Revised: [Warmer, clearer, more helpful email]
Self-identified problems. Actively improved.
AI doesn't automatically give you its best work on the first try. When you ask it to critique its own output, it notices things it missed — awkward phrasing, unclear structure, wrong tone. The revision addresses those specific issues.
This works because style and tone are subjective. AI can judge whether something sounds too formal, too wordy, or too cold — there's no "correct answer" to verify, just appropriateness to evaluate. That's something AI can do on its own.
Self-critique works for style and quality — not for correctness. Research shows AI cannot reliably detect its own errors in math, logic, or facts. If you ask AI to "check your reasoning" on a math problem, it often makes things worse. For tasks with objectively right or wrong answers, use external verification instead (calculators, code execution, fact-checking). This technique is for polish, not proof.
Ask AI to write, then critique, then revise. Be specific about what to critique — tone, clarity, structure. Two or three rounds usually gets you there. Works for writing; skip it for math and reasoning.