Stop Asking AI to Agree With You

AI is very good at helping an idea move forward. That can become a problem when the idea needs questioning first.

If you explain a plan and ask, "Does this look good?", you have already pointed the conversation towards agreement. The AI may improve your wording while leaving the weak assumption untouched.

Use AI as a second pair of eyes, not a machine for confirming what you already think.

The question shapes the criticism

Compare these requests:

I think this training plan is clear. Can you improve it?

Review this training plan as a cautious participant. What is unclear, unrealistic or missing?

The first asks for polish. The second asks for examination. Neither guarantees a good answer, but they give the AI different jobs.

Give it a useful point of view

"Be critical" is broad. Tell the AI whose concerns matter.

For a website page, ask it to read as a beginner using a phone. For a purchase, ask it to test your choice against your budget and non-negotiable requirements. For a workplace plan, ask what a busy manager or hesitant member of staff may question.

Do not ask AI to imitate a named living person. Describe the knowledge and concerns the reviewer should bring.

Ask for the strongest case against the idea

Try this prompt:

I am considering [idea]. My goal is [outcome]. My main assumptions are [list them]. Do not improve the idea yet. Identify the strongest reasons it may fail, the evidence that would change the decision and any group affected that I may have missed. Separate serious risks from minor preferences. Then suggest the smallest test that would give me useful evidence.

This produces something more useful than a list of generic pros and cons. It asks what would alter the decision.

Do not turn criticism into theatre

Some prompts ask AI to be brutal, ruthless or destroy an idea. That can create dramatic wording rather than better judgement.

You want specific weaknesses, not insults. Ask for evidence, impact and a practical test. A calm objection is easier to use than a performance.

Keep the final decision

AI does not know every local condition, relationship or consequence. It may overvalue what is easy to express and miss what people know through experience.

Use its challenge to find questions. Check those questions with evidence and the people involved. Then decide.