The Most Dangerous Thing AI Can Do? Tell You You're Right
Here's something that took me years in journalism to learn, and it's exactly what trip up beginners with AI: the tool is incredibly agreeable. Show it a position—any position—and it will flesh it out, strengthen it, dress it up in impressive language. It will find supporting evidence, anticipate objections, and hand you back a polished version of what you already believed. That feels like thinking, but it's actually the opposite. It's echo chamber thinking, and AI does it beautifully.
That's where adversarial prompting comes in. Instead of asking AI to build your case, you ask it to tear it down. The sequence is simple but requires discipline: state your position clearly, then explicitly tell the AI you want the strongest possible counterargument. Use prompts like "Give me the best case against this position" or "Argue as if you were an expert who disagrees with everything I just said." You're not looking for a token opposition—you want the AI to steelman the opposing view, the version that would actually make you reconsider if you heard it from a human expert.
Now here's the harder part: what you do with the output. When AI surfaces objections you didn't anticipate, or articulates a flaw in your reasoning that you genuinely hadn't considered, your job is to sit with that discomfort. Don't immediately prompt it to defend your original position—that defeats the purpose. Don't dismiss the AI's objections as artificial or overblown. Take notes. Let the challenge reshape your thinking, even if you ultimately stick with your original view. The goal isn't to abandon your position; it's to stress-test it.
The temptation to dismiss inconvenient AI responses is real, and I've caught myself doing it. But here's what I've learned: if an AI-generated objection would make you reconsider if a colleague raised it, you owe it the same respect when the machine produces it. Adversarial prompting isn't about distrusting AI—it's about distrusting yourself just enough to make your thinking stronger.
Published on PromptResponse: