Critical Friend Prompt

Have you ever... asked AI and got an answer like, "Yeah, all good."

Try this formula and the AI will start to take itself more seriously as a consultant.

1) "Do not agree first. Give a neutral summary of 1-2 sentences."

How to Work: Let AI Reflect What We Really Say First (Like an Understanding Check)

Strength: Reduce talk about different things / Reduce issues / Make answers "straight problems"

Weaknesses: Some tasks are slow + waste answer space, especially simple questions.

2) "Arguing for at least 3 reasons (including weaknesses / vulnerabilities / overlooked things)."

How to Work: Forcing AI to Play Job Examination / Opposing Lawyer

Strengths: Found blind spot sensitive / Fits decision-making-planning-content / business

Weaknesses: Risking "objection to objection" (AI may drama / raise unimportant arguments) and making people who want encouragement feel pressed.

3) "Offer three alternative assumptions that explain the same thing."

How to Work: Not Locked in One Description, Forcing Multiple Views

Strengths: Very good with problem solving tasks (e.g. falling peaks / images not passing / ideas) because there is a choice.

Weaknesses: If the reader / person asks does not like complexity, they are confused, "Which one is it?"

4) "Separate facts / assumptions / opinions clearly."

How to Work: Elevate from Beautiful Answers → Answers with "Knowledge Status"

Strengths: Reduce Believing AI Sleep Ear Close / Fits Sensitive Data Matters (Health, Law, Finance)

Weaknesses: AI is still "stuck," just labeled as fact, so you have to check the source when it's important.

5) "If there is not enough information, say that it is not enough and say that you need to know more (briefly)."

How to work: together, AI, guess yourself

Strengths: Reduce hallucination (compose data) / get us a "list of things to send more"

Weaknesses: If written poorly, AI will ask too much, so that the work does not walk (solve with the words "Ask only 1-3 most important questions").

3-line add-on (this one is good, but it also has a double-edged sword)

• "Must point out risks / side effects / exceptions"

Hard: Fits real jobs. Because the world always has exceptions.

Mild: If you use everything, you will make the post look too stressful.

• "Rating Confidence (0-100) + Brief Reason"

Stiff: Make sure you know what to believe, what to check.

Mild: Numbers are model sense, not real statistics (don't think 90% is 90% correct)

• "If there is a sensitive issue / information may be wrong, warn first."

Hard: Save yourself from sharing misinformation.

Mild: Sometimes a warning to waste flavor (especially light content)

Summary of Critical Friend Prompt's "Strengths"

Make the answer "exactly the problem + rational + see the risk point" more.

Reduce AI symptoms, say, because it doesn't help decision-making.

Fits Systems Thinking Tasks: Business, Content, Strategy, Problem Solving

Summary of "weaknesses" to be admitted

Slower / Longer (not suitable for short questions)

It can make the atmosphere look strong if it applies to emotions and encouragement.

Not guaranteeing the truth - just forcing "think better," having to check the source material itself when it's serious.

- This mode is more suitable for "Think Work" than "Consolation."

- Doctor / Legal / Financial Matters: Used to frame ideas, but should always check the reference source.

# AI consultant # Ask to get a good answer # Ask to the point# Trick ai # Lemon 8 Howtoo