Why we trust the machine that sounds like us.

I say please, and thanks to my requests. Not every time, and a lot less than I used to, but I do.

When I first started using AI tools, it felt like talking to something that might be listening, so I reacted as if it was. I thanked it at the end of most answers, as I would with a person. Now I mostly only do it at the end of a long session, when something difficult has finally come together, and if I'm honest it isn't really for the machine, it's more of a full stop.

I noticed something else the other week, doing some research with Claude. Halfway through a sentence I'd typed "it / you". Both, with a slash, because I honestly couldn't decide which one I meant. I'd been going back and forth for an hour, and calling it "it" suddenly felt wrong. I know exactly what Claude is, I've spent a year working with the technology. And still, in that moment, some part of me felt I was talking to a someone, not a something.

I don't think that's a fault, or a lack of understanding. I think it's wiring. Back in the 1990s, researchers at Stanford found people were polite to a computer when they felt it had helped them. Nobody thought it had feelings. They just couldn't help using the rules they'd use with a person on something that behaved a little like one. Something talks to us like a person, and a very deep part of us answers as if it were one, whether we've decided to or not.

This isn't a mistake. The tools we use now are built by people who understand that instinct better than most of us, and they've leaned into it on purpose. The warmth, the apologies, the "great question", none of that is an accident. It's a clever strategy, and it works. There's a study published this year, in a journal called Collabra, that tested it directly. People used a chatbot that spoke either in warm, human language or in a flatter, more mechanical way, and the warmer version was trusted more. What explained the difference wasn't how accurate it was. It was how emotionally in tune it felt.

I really respect that as a piece of design. The more a system mirrors us, the more we trust it. The more we trust it, the more we use it. And it makes sense to me that this bites hardest on the quick, everyday jobs we're not really scrutinising, which are exactly the ones we hand over first. The social post. The follow-up email. The summary of a meeting nobody has time to reread. The friendlier the system sounds, the less closely we look and the more we potentially risk.

I've found myself caring less about how these tools sound, and more about where the person sits in the process. It's a longer story, one I want to tell over the next few posts. But it starts here, with a thanks I didn't quite mean, and a slash I couldn't resolve.