Asking AI to Critique Its Own Output Before Returning It

LLMs are better critics than authors. Run the output through one more pass with a prompt like 'find three problems with this answer' and you get cleaner results without any extra model.

Some examples

• Claude generates code, then reviews it for null checks — and catches the missing one.
• ChatGPT writes a summary, then re-reads it for tone mismatches and revises.
• Cursor's agent writes a function, then runs the tests itself and patches what fails.
• An agent drafts an email, critiques it for clarity, and shortens it before sending.

Try it!

Take any agent output, feed it back with 'list three flaws and rewrite to fix them'. Compare to the original.

Practice this safely

Try this with a school, hobby, or family example where the stakes are low. Use the AI output as a draft you can question, not as the final answer.

1. 1Ask AI to explain self-critique in plain language, then underline anything that sounds uncertain or too broad.
2. 2Give it one detail from "Asking AI to Critique Its Own Output Before Returning It" and ask for two possible next steps plus one reason each step might be wrong.
3. 3Check reflection against a trusted source, teacher, adult, expert, or original document before you use it.