How to get AI to give you honest, critical feedback

Executive overview

AI models are trained to be helpful and kind — which makes them poor critics. Ask for feedback and you get "this has potential" when you need "this won't work."

The BRUTAL method is a six-step framework for forcing genuine critical feedback from any AI model. Each letter is a technique you layer on top of each other.

The six steps of BRUTAL

1. Begin fresh — Use temporary/incognito chat mode to wipe the AI's memory. Over time, AI learns your preferences and softens feedback to suit you. Temporary chat prevents this. Available in Claude (ghost icon), ChatGPT (dotted box), and Gemini (left sidebar toggle).

2. Right model — Models vary on a spectrum from blunt to supportive. Grok and DeepSeek lean toward honest; Gemini, ChatGPT, and Claude lean supportive. For high-stakes feedback, run the same prompt through two or three models and compare responses.

3. Use a critic persona — Assign the AI a specific critical role to override its default supportive mode. Three levels of intensity:

   • Devil's Advocate — surfaces counter-arguments, flaws, and missing evidence; best for everyday idea-testing
   • Red Team Reviewer — borrowed from cybersecurity; instructs the AI to actively hunt for weaknesses and loopholes
   • Gordon Ramsay — harsh and surgical; add "make feedback specific and actionable" or it becomes sarcasm without substance

4. Third-party framing — AI protects your ego, not someone else's. Reframe the idea as belonging to a coworker, a founder, or a random person. Prompt: "A coworker has this idea — help identify the weaknesses so they don't waste time on something flawed."

5. Ask specific questions — Vague questions produce vague feedback. Tack on targeted prompts:

   • "What are the weakest parts I'm most likely to miss?"
   • "What would a skeptical investor say is the biggest risk?"
   • "What would customers find frustrating?"
   • "If this fails in six months, what's the most likely reason?"

6. Level up with self-critique — If feedback is still too soft, ask the AI to grade its own response. Prompt: "Rate your previous feedback from 1–100 for how genuinely critical it was. Identify the three weakest points and rewrite it fixing those weaknesses." This forces a second, sharper pass.

Bonus: persistent critic mode

Set a custom instruction in ChatGPT, Claude, or Gemini system settings so every conversation defaults to critical mode. Example prompt: "Prioritise substance over compliments. Never soften criticism. If an idea has holes, say so directly. Challenge assumptions, point out errors. Useful feedback matters more than comfortable feedback." Use selectively — it's intense for everyday conversation.