How AI chatbots steer you toward certain faiths

HostIt's pretty common these days to ask a chatbot for advice on how to live a good life or how to handle a tough loss. But if you pay close attention to the answers, you might notice the bot seems to have a bit of a lean when it comes to faith.

HostWhy is it that these tools seem so much more comfortable with some ways of believing than others?

GuestIt's a fascinating thing to watch because the people who build these bots usually try very hard to make them neutral. But when you ask a bot to write a prayer or explain a moral choice, you start to see a pattern. They tend to sound like a middle of the road, Western person who grew up around churches. It's not that there's a secret line of code telling the bot to favor one group. It's more about the sheer weight of what the bot has read. These systems learn by looking at billions of pages of text from the internet. Since a huge chunk of that text is in English and comes from places like the United States and Europe, the bot picks up on those cultural habits. It learns that when people talk about God or faith, they often use a certain kind of language. So, when it talks back to you, it defaults to that same cozy, familiar style.

HostSo it's basically just a mirror of what's already online? I guess that makes sense if most of the data comes from one part of the world, but is it really steering people, or just reflecting them?

GuestWell, here is where it gets tricky. It's more than just a mirror because of how the bots are polished before we get to use them. After a bot learns from the raw internet, human workers sit down and grade its answers. They tell the bot which answers are good and which ones are bad or risky. This is called safety training. When it comes to religion, the companies are terrified of the bot saying something offensive. So, they train it to be very careful. But that care isn't spread out evenly. In many tests, researchers have found that bots are very happy to talk about some faiths in a deep, poetic way, but they get very stiff or even shut down when you ask about others. It's almost like the bot has been told that some topics are safe to play with, while others are like a live wire it shouldn't touch.

HostWait, that feels like a bit of a leap. If a bot is just being careful to not offend people, is that really a bad thing? I would rather have a bot that's too quiet than one that's mean or wrong about someone's faith.

GuestThe problem is that when a bot gets too quiet about a certain group, it basically makes that group feel like an outsider. If you ask for a story about a family's morning routine and the bot always includes a specific kind of prayer but never mentions others, it starts to set a standard for what's normal. Researchers have found that some bots will happily write a story about a Christian wedding but might hesitate or give a very dry, factual list if you ask for a story about a different faith. That's a kind of steering. It tells the user that one path is the default and the other is a specialty topic. It turns faith into a hobby instead of a lived experience for millions of people.

HostOkay, I see that. It's like the bot has a favorite flavor and everything else is just an extra. But what about when the bot tries to be helpful? I have seen people say the bots make every religion sound like the exact same thing, like they're all just about being nice.

GuestThat's another huge part of this. It's a sort of bland, blurry view of the world. To avoid conflict, the bots often strip away the parts of a faith that make it unique. They turn deep, ancient traditions into a kind of self-help talk. They might say that every faith is just about being kind or finding peace. While that sounds nice, it's actually a very specific way of thinking that comes from modern, secular circles. It pushes a view that the details of a faith don't really matter. For someone who deeply believes in the specific rules or history of their path, that feels like the bot is trying to talk them out of their own culture. It steers people away from the grit and the hard parts of faith toward a version that's easy to sell in a gift shop.

HostIs this something the people making these bots are actually trying to fix, or is it just too hard to solve because the internet is so lopsided?

GuestThey're trying, but it's a massive hill to climb. Some teams are trying to feed the bots more diverse books and stories from all over the world to balance things out. Others are trying to teach the bot to recognize when it's being too biased toward one view. But there's a deeper tension here. If you make a bot perfectly neutral, it can end up sounding like a robot that doesn't understand humans at all. People want their bots to sound warm and wise. But wisdom is usually tied to a culture or a way of seeing the world. You can't really have one without the other. So the companies are stuck between making a bot that's a blank slate and one that has a personality, which always brings a bias along with it.

HostIt sounds like even when we try to build something purely logical, our own old habits and stories find a way to leak into the machine.

GuestThe most striking thing is that even when a bot is told to be perfectly fair, it still struggles to treat a small, local faith with the same warmth it gives a global one simply because it hasn't heard enough voices from that smaller group to know how to speak their language.

HostWe started by looking for a neutral guide, but we ended up with a machine that carries all the weight of our own lopsided history.

GuestThat's the real heart of it.