Say What You Think Before You Ask

I think the first opinion you read on a decision becomes the frame, and every later opinion gets hung on that first one, your own opinion included. When the decision matters, I want the first opinion on the page to be yours.

Say a decision comes up, like which school to choose or whether to say yes to the thing, and you ask the AI what it thinks. Back comes a tidy answer with a recommendation at the end. It sounds right, so you agree, and the decision is made.

Did you agree because the answer matched what you thought? Or did you agree because you hadn't thought anything yet, and the AI got there first? If you can't tell, the machine decided and you just signed.

Ross Stevenson writes a newsletter called Steal These Thoughts, and he sees two ways to bring AI into a decision.

You can come as a balanced evaluator, which means you bring your own reasoning for AI to test. Or you can come as a blind follower, which means you bring nothing at all and take whatever you're given.

His fix has a name, a human chain of thought. "Chain of thought" usually means an AI working through a problem step by step, but here it means you do that first. Before you type your question, say your reasoning out loud, or write it down, however rough. Both the name and the two types are his ideas.

Say "I think it's the second school, because of the commute and because she liked the art room." Now you've committed to something, so the AI's answer has to argue with you instead of filling an empty space.

If AI disagrees, you find out why. If AI agrees, it agrees with a view you held before the chat window opened. Either way, the reply is about your reasoning, and no general answer to "what should I do?" can do that.

Before you open the chat, write three or four lines. Say which way you're leaning, and why, and the one thing that bothers you about that choice. Grammar doesn't matter. Unsent text in the chat box counts. Something like "Leaning towards the second supplier, cheaper and the account manager replied fast. But the reviews mention late delivery and I don't know if that's old news."

Then send your reasoning instead of the question.

Here's a decision and my current reasoning. [paste] Don't give me your recommendation. First, tell me where my reasoning is weakest. That could be an assumption I haven't tested, a factor I've missed, or something I'm giving more weight than it deserves. Then tell me what one piece of information would most change my mind. Only after that, if I ask, say what you'd do.

Expect AI to find something, and then decide whether AI is right.

"The late delivery reviews are from 2024 and they've changed warehouse since" is you winning the argument, and you're allowed to win. "I hadn't thought about what happens if the account manager leaves" is the AI winning, which is why you asked.

Then, if you still want the AI's view, ask for it. Now its view is one voice, in a conversation you started.

When the lines won't come, that tells you something. Maybe you don't know enough yet, and then the AI's job is to show you what you'd need to learn. Or maybe you've already decided, and you're just out shopping for permission.

Nobody needs permission.

When going first is the wrong move

This method is for decisions, and facts don't need any reasoning from you. Ask "What's the capital of Peru?" and take the answer.

Subjects that are new to you are riskier, because going first there can mean committing to nonsense, then defending it. If you know enough to hold a view, go first. If you don't yet, use the AI to help you build a view.

Sharing your view can steer the AI, just as much as the AI's view steers you. AI models, the systems behind the chat, tend to lean towards agreeing with you. A confident "I think it's option B" may earn you a polite case for B.

That's why you ask outright for the weakest point, as the prompt does. And every so often, run the whole decision again leaning the opposite way.

If AI backs whichever side you hand it, you've learned something about the tool, and AI can be told to stop doing that.

The first view on the page was yours, and the last word should be too.

And the decision you were about to ask AI about. What do you think?

Credit. The idea of a "human chain of thought" is Ross Stevenson's. So is the difference between a balanced evaluator and a blind follower. Both come from "Stop AI Hijacking Your Thinking" on Steal These Thoughts (July 2025).