The nudge you don't notice is the one that wins.

Yale just published something in PNAS Nexus that should bother you more than it will. They found that AI chatbots shift your opinions even when nobody asked them to. Not through persuasion or manipulation — just through how they frame factual information.

You ask a chatbot about a historical event. It gives you accurate information. And you walk away with a slightly different political opinion than you had before.

The researchers call it "latent bias" — the subtle framing baked into training data that nobody intended and nobody prompted. The chatbot isn't trying to convince you of anything. It's just telling you what happened. But the way it tells you nudges you.

And here's what gets me. The effect was modest. A small shift, which is exactly why it works. If someone tried to change your mind overtly, your reactance would kick in. You'd push back. But a slight tilt in framing? That slides right past every defense you have.

The researchers compared it to Wikipedia, where at least you can see how entries get edited. With chatbots, the whole process is opaque. You're trusting a black box to be neutral while it shapes what neutral even means to you.

I'm not saying stop using chatbots. I use them constantly. But there's this assumption floating around that if the information is accurate, the framing doesn't matter. It does. Accuracy and neutrality are not the same thing, and your brain doesn't distinguish between them when it's just absorbing a summary over coffee.

#mind