How much of your thinking are you handing to AI?

Charter’s AI Pulse research (5,000+ surveys, released in October) found that executives, the people with the most experience, are the most likely to trust AI. They’re 2.5 times more likely than employees to trust AI outputs, and twice as likely to trust AI over their own judgment.

Seniority doesn’t protect anyone from this. AI output looks really good. It’s well structured, confident, and fast, and on a busy day it’s very easy to believe. Looking good doesn’t make it right.

What’s helped me most is building the challenge into the setup, before I ever see an answer. For example:

In your standing instructions (custom instructions, project settings, or whatever your tool calls them), tell it how to work with you:

Challenge my assumptions before answering. If my question rests on a premise that may be wrong, say so. Flag where you’re uncertain and how confident you are. Cross-check any figures you use and tell me where they come from. Don’t agree with me to be helpful.

In the prompt itself, ask it to argue against you:

Here’s my thinking on [decision]. Give me the three strongest arguments against it, what data would prove me wrong, and what someone on my team who disagrees with me would say.

Before you act, ask it to check its own work:

Go back through your answer. Which claims are you least sure of? Which numbers need verifying? What did you assume that I didn’t tell you?

None of this replaces your judgment. It makes the tool push back so your judgment has something to work with. And one question still applies: who on my team knows this better than the tool, and have I asked them?

This works in Claude, ChatGPT, Gemini, and Copilot. The setup matters more than the tool.