A persuasive response can feel more certain than the evidence warrants. This essay is about keeping your own judgment during a conversation, not demonstrating emotional fitness to a model.

You do not need to earn accurate answers by appearing calm, strong or agreeable. AI output is not a measure of your emotional stability, and a difficult interaction is not a personal failure.

State the kind of help you want

If you want critique, name the criteria. “Identify unsupported claims and missing steps” is clearer than “be brutally honest.” Ask for reasons and evidence rather than a verdict about you.

If you want options, ask for tradeoffs. If you want factual information, identify what needs sourcing and what the model should leave unresolved.

Make disagreement useful

Write down your current assumption before asking for alternatives. Then consider what evidence would change your mind. This makes it easier to evaluate a response without accepting or rejecting it based only on tone.

A model can agree too readily, disagree without good reason or invent support for a claim. Treat its critique as something to examine, not a final judgment.

Pause before a consequential decision

Separate exploring an idea from acting on it. Before publishing, spending money, changing access or making another consequential decision, check the facts and the practical consequences.

For medical, legal or other specialist decisions, an AI conversation does not replace qualified help. If a conversation becomes upsetting or repetitive, it is reasonable to stop and return later or talk with someone you trust.

Practice

Choose a low-stakes proposal. Ask the model to list its assumptions, strongest objections and missing information. Select one check you can carry out independently. Record whether the evidence supports changing your proposal.

You can use AI thoughtfully without making it the judge of your competence, identity or future.