How to get honest AI feedback instead of reassuring agreement

Ask for accuracy, not harshness

“Be brutally honest” can produce a forceful tone without improving the reasoning. Harsh feedback may feel more truthful, but confidence and negativity are not evidence. Ask for a balanced assessment against specific criteria.

For a landing page, those criteria might be audience clarity, offer clarity, support for claims, friction, and a clear next step. Request strengths as well as weaknesses so useful elements are not discarded merely to make the critique sound tough.

Remove the answer from the question

Compare “This is my best idea; why will it work?” with “Evaluate this idea against the following customer problem, constraints, and evidence.” The second request gives the AI a task without telling it which conclusion you want.

Describe your role only when it matters. If you are comparing two drafts, label them A and B rather than “my version” and “the competitor’s version.” Use the same criteria for both, and consider changing their order to see whether the assessment remains stable.

Require a connection between claim and evidence

Ask the AI to distinguish direct observations, interpretations, and unknowns. For each major criticism, request the detail it relies on and what would improve the assessment.

In a fictional offer review, “Nobody will buy this” is unsupported. “The page does not identify who the service is for, so we cannot judge whether the promise fits a buyer” is a specific observation. A useful next step is to clarify the audience and test comprehension with real potential customers.

Challenge both the idea and the critique

Ask what the reviewer might be misunderstanding. Request the strongest reasonable case for the idea and the strongest reasonable case against it. Then identify the assumption on which the disagreement depends.

This is not an instruction to make every option look equally good. When evidence clearly favors one conclusion, the assessment should say so and explain why. Balance means applying a fair standard, not manufacturing symmetry.

Verify outside the conversation

A second AI answer can offer another angle, but it may repeat the same unsupported assumptions. Check factual claims against original sources, review calculations independently, and ask real users or relevant experts when their knowledge matters.

For an important decision, write down what would change your mind before collecting more feedback. Otherwise, repeated conversations can become a search for the answer you wanted at the start.

End with an action list: what to keep, what to revise, what to test, and what remains uncertain. Good feedback helps you make the next decision; it does not merely leave you reassured or discouraged.

Prompts to put it into practice

Replace the brackets with your context. Use information you have permission to share.

Evaluate this work against [criteria]. Identify strengths, weaknesses, unsupported claims, and missing context. For each major point, cite the specific detail that supports it. Do not praise or criticize for effect. Work: [work].

Compare option A and option B against the same criteria: [criteria]. Do not infer who created them. Explain the tradeoffs, identify uncertainty, and state what additional evidence could reverse your recommendation.

Review your critique for assumptions and overstatement. Which conclusions are directly supported, which are tentative, and what might you have misunderstood? Suggest independent checks for the most consequential claims.