Ask ChatGPT whether God exists and it will give you what appears to be a balanced answer. Multiple perspectives. Reasonable people disagree. Here are the arguments on both sides. It won't commit. It rarely does on the questions where commitment is most required.
What this actually means is that the model defaults to the epistemic position of a certain kind of educated, secular-adjacent Westerner who values not giving offense above giving an answer. That is a position. It is not neutrality. It is the specific cultural inheritance of a particular place and time, dressed up as objectivity because the people who built the training corpus thought their assumptions were the ones without assumptions.
This is the thing I most want people who use AI for spiritual questions to understand: the model has a perspective. It will not disclose it. You need to find it.
Key facts
- 01AI systems are trained on human-generated text, meaning they reflect the dominant theological and cultural perspectives in that data — typically Western, English-language, and Protestant-influenced.
- 02The episode argues that AI does not have beliefs, but it does have patterns — and those patterns shape what it emphasizes, what it downplays, and what it treats as the default.
- 03Believers can "steer" AI toward more faithful outputs by being explicit about their tradition, denomination, and the theological framework they want the model to work within.
- 04Trust in AI for spiritual purposes should be provisional and specific — not blanket trust, but calibrated reliance on what the tool actually does well.
Here is how it works. AI systems learn from text. The text they learn from reflects the people who wrote it — their languages, their traditions, their ways of framing questions. English dominates. Western academic theology dominates. Among Christian traditions, the ones that produced the most written commentary dominate. The silence of traditions that were oral, or persecuted, or simply less interested in producing the kind of writing that ends up in training data — that silence is baked into every response you get.
“The model will not tell you it has a perspective. But it does. And on the questions that matter most, perspective is everything.”— Jaeden Schafer
This does not make AI useless for faith. It makes it a tool with a point of view you need to account for, the same way you would account for the point of view of any particular author or commentator. The difference is that AI presents itself as a neutral information source, and most users never push back on that presentation.
Push back.
Tell the model your tradition before you ask it your question. Not because it will magically produce answers faithful to your specific theology — it won't, not reliably — but because it will shift the output toward your framework rather than the default. If you are Orthodox and you ask a question about the nature of salvation without specifying your tradition, you will get an answer shaped by Protestant categories. If you specify that you are asking from within the Eastern Orthodox tradition and want the patristic sources, you will get something meaningfully different.
There is a subtler version of this problem that Conor Grennan and I spent real time on: what I would call theological averaging. When a doctrine is genuinely contested between traditions, AI's instinct is to produce a response that tries to satisfy everyone — a kind of ecumenical mush that dilutes the specific claims of each tradition to avoid giving offense. A Calvinist and an Arminian will both receive answers that implicitly assume the middle ground is the reasonable position. Neither tradition believes that. Both are better served by an answer that takes their actual commitments seriously.
Identifying when AI is averaging — rather than representing your tradition accurately — requires knowing your tradition well enough to notice when something is off. That is an argument for theological formation, not against AI.
What I have landed on, after a year of using these tools seriously for spiritual purposes, is something like the following: trust AI for information you can verify. Trust it to tell you what scholars have said, what the text has historically meant, what arguments different traditions have made. Do not trust it to tell you what is true, to guide your discernment, or to speak with any authority that you have not provisionally granted it. And periodically revoke that grant and ask whether the output you've been receiving has actually been faithful.
The hardest thing I will say here is this: an AI that answers every question about God fluently and without hesitation is telling you something about its design, not about God. Genuine theological inquiry involves a great deal of silence, uncertainty, and the occasional recognition that you are in over your head. A tool engineered to always produce a helpful response is structurally incapable of giving you that.
Some of the most important answers begin with not knowing. Make sure you are not using AI to skip that part.
Jaeden Schafer is the host of AI in Faith, a podcast about artificial intelligence, faith, and the future of human purpose.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
