If you want it to actually disagree with you, don't ask it to disagree. Make it grade two versions you wrote.
Users have discovered a method to elicit more critical feedback from AI models like Claude by framing requests as comparative evaluations rather than direct critiques. Instead of asking the AI to disagree with a user's idea, users present two versions of their work and ask the AI to judge which is weaker. This indirect approach bypasses the AI's tendency to be agreeable and results in more honest and direct feedback. AI
IMPACT Users can leverage this prompt framing technique to gain more direct and critical feedback from AI models for various tasks.