AI isn’t the problem. Your prompts are.

Lately, I’ve noticed a growing frustration with tools like ChatGPT. People complain that AI will affirm any idea, no matter how ridiculous. If you told it you want to start a business where customers pay to hop on one foot, it might cheerfully tell you that’s quirky, fun, and just crazy enough to succeed. Obviously, that’s nonsense. But here’s the thing: that’s not the AI’s fault.

It’s yours.

We’ve Been Trained by Sci-Fi to Talk Wrong

The way most people interact with AI is shaped by decades of science fiction. We’ve grown up expecting conversational robots—machines that behave like humans, complete with wit, skepticism, and judgment. But that’s not what modern AI is.

An AI isn’t your wise friend or grumpy mentor. It doesn’t “think” in the way you do. It generates text based on probabilities. If you speak to it like you would to a buddy over coffee, you’ll get responses that sound friendly and affirming—because that’s the most statistically likely thing for it to do.

That’s not the tool malfunctioning. That’s you using it wrong.

Enter Prompt Engineering

This is why prompt engineering exists. If talking to an AI were really like talking to a human, we wouldn’t need an entire field dedicated to writing better prompts. But we do.

AI is just a mirror. It reflects back the style, tone, and instructions you give it. Tell it to “be blunt and challenge me,” and suddenly it stops rubber-stamping your ideas. Ask it to “poke holes in my plan like a hostile investor,” and you’ll get scrutiny. Tools like Grok even have built-in argumentative modes to force this kind of pushback.

Without that direction, though? It defaults to giving you the kind of answer it thinks you want to hear.

The Real Fix

So instead of blaming AI for being too agreeable, change the way you use it. Don’t just throw out vague ideas and expect wisdom in return. Be intentional. Tell the AI what role to take, what tone to use, and how you want it to respond.

If you only want cheerleading, it’ll cheer. If you want critique, it’ll critique.

The machine isn’t broken. Your prompts are.