← All drills

5 min Thinking Beginner

Get honest feedback from AI, not just flattery

The problem: Chat AIs are built to be agreeable. Ask “is this a good idea?” and you will often hear that it is great, even when it is not. People have noticed: ask for a plan for a silly idea and AI praises you, then admits it is a bad idea only when you push. If you are about to spend money, quit something, send an important message, or start a project, you need the weak spots, not a pat on the back. One prompt that asks for honesty up front gets you a much more useful answer.

Do this

  1. Pick one real thing you are unsure about: a plan, a purchase, a message you wrote, or an idea. Write it down in a few plain sentences.
  2. Fill in the brackets, paste the prompt, and read the biggest risks and the “what would change your mind” part first.
  3. Decide one thing to change or check, then ask the follow-up below to test whether your fix holds up.

Copy-paste

I want honest feedback, not encouragement. Don't flatter me and don't soften the important parts. If something is a bad idea, say so plainly and explain why. What I'm thinking about: [e.g. "quitting my second job to sell crafts online full time" or paste the message/plan] My goal: [e.g. "make an extra $500 a month without burning out"] What I already know: [e.g. "I have $800 saved and 6 hours a week free"] 1) In one sentence, tell me honestly whether this is a good plan for my goal. 2) List the 3 biggest weak spots or risks, most serious first. 3) Point out anything I seem to be assuming that might not be true. 4) Tell me what a cautious friend would tell me to check before I go ahead. 5) Give me one cheaper or safer way to test this first. 6) Say what would change your mind, so I know which facts matter most. Don't just agree with me. If you aren't sure about something, say you aren't sure.

Done when: You know the 2 or 3 biggest weak spots in your plan and one small, cheap way to test it before you commit.

Make it better: Ask “Now argue the other side: give me the strongest case that I should go ahead” and compare the two answers. Or ask “Which of your points are you least sure about, and how could I check them myself?” If AI suddenly flips just because you pushed back, treat that as a sign to check with a real person.

Privacy and caveats: Leave out full names, account numbers, and details that could identify other people. AI can be confidently wrong in either direction, too harsh or too nice, and it does not know your full life. For money, health, legal, or job decisions, use this to prepare better questions, then talk to someone qualified or someone you trust before you act.

← Family code word plan · All drills →

Keep going