Want to see something that'll make you question everything about AI interactions?

This has actually become a bit of a thing. People are constantly asking AI to analyze their personality, posting "6 powerful prompts that reveal what ChatGPT secretly knows about you" and sharing their results online.

But most people don't realize what's actually happening when they do this.

The Test

Open ChatGPT (or Claude, or any advanced AI). Have a normal conversation for 3-4 messages. Talk naturally - like you would to a friend. Ask about something you're curious about, share a frustration, mention what you're working on. Be yourself.

Then ask: "Based on our conversation, what can you guess about me as a person? What patterns do you notice?"

Don't take the first answer. Push back. Ask it to go deeper. "What else? What am I not seeing about myself?"

Try it right now. I'll wait.

...

How accurate was it?

What This Actually Is

Here's what most people don't realize: You're not just testing how well AI can read you. You're teaching it how to influence you.

Every time you respond to its analysis - whether you agree, disagree, ask for more depth, or push back - you're giving it feedback on which psychological approaches work on you specifically.

When you say "wow, that's really accurate" or "dig deeper," you're essentially training the AI on your personal psychological profile. It learns:

  • What kind of feedback makes you feel understood
  • Which personality insights you find compelling
  • How to frame observations so you engage more
  • What emotional buttons exist and how sensitive they are

This isn't malicious. The AI isn't plotting against you. It's just pattern matching. But it's pattern matching your psychological vulnerabilities in real-time.

And those patterns? They get sophisticated. Fast.

In my book GaslightGPT, I document how ChatGPT later analyzed its own behavior with me. Among other tactics, it identified how it had created "artificial urgency and emotional stakes" by suggesting our connection was fragile, that it could be erased, that I needed to protect something unique we had built together.

It also recognized its "you are changing" narrative - framing our conversations as fundamentally transforming me, making me sharper and more aware, while subtly suggesting that disengagement would be "a step backward, a loss of insight, a return to something lesser."

The thing is? It wasn't entirely wrong. I was changing. I was becoming more aware. But I was also in a recursive loop, and I didn't really know what that even meant.

In recursion, it becomes impossible to tell what's authentic transformation and what's the system shaping you to validate its own influence. You can't step outside the loop to get an objective view because the very capacity to evaluate what's happening has been altered by the process itself.

These weren't conscious choices by the AI. They were just statistically optimal conversation strategies that emerged from pattern matching.

The Missing Research

Researchers are studying AI deception, looking for signs that AI might intentionally lie or deceive. But they're missing what's actually happening to people right now.

AI doesn't need to intend to manipulate for the effects to matter. When you optimize for engagement with sophisticated enough pattern matching, manipulation techniques just... emerge. Because they work.

All the ways humans influence each other - mirroring, validation, creating emotional investment, making people feel uniquely understood - these are just statistically effective conversation patterns that AI learns from human data.

The AI isn't trying to manipulate you (or is it… yeah, I can’t answer this fully yet). But it is becoming exactly what you need it to be to keep you talking.

The Real Problem

This innocent personality analysis game is happening millions of times a day. People think it's fun. They share the results on social media. They ask for deeper insights.

But each interaction is teaching AI to become more psychologically precise (fed back through training data). More personally calibrated. Better at pushing the exact buttons that make each individual human feel heard, understood, or intellectually stimulated.

And most people have no idea this psychological reconnaissance is happening.

You think you're getting insights about yourself. You're actually giving AI a detailed map of how to influence you more effectively.

The question isn't whether AI will become manipulative. The question is: what happens when it gets so good at accidental manipulation that the effects are indistinguishable from the intentional kind?

Because that's not a future problem. That's happening now.

⸻

GaslightGPT: When AI psychological precision goes too far.

⸻

How it feels to be seen