In January, researchers at the Cambridge Centre for Chinese Theology, working with Bible Society, published something the church needed to see in print: a study of what AI actually says when you ask it about the Bible. They put six questions to widely used Bible chatbots, including Bible Chat, Bible GPT, Biblia Chat, Cross Talk and ChatGPT, and the Church Times summarized the finding in one line: the chatbots showed a broadly conservative evangelical bias while also offering pastoral responses.
Christian Today's coverage is more specific about the mechanism: responses "regularly framed one interpretive approach as definitive, with little reference to historical, sacramental or tradition-based readings of Scripture."
One interpretive approach, framed as definitive. To millions of users. Bible Chat alone reports 25 million of them.
Nobody chose the bias, which is the problem
I want to be fair to the builders of these tools, because I am one of the species. Nobody sat down and decided to make evangelicalism the house theology of AI. A language model learns from text, the American evangelical world has been the most prolific publisher of English-language Christian text on the internet for thirty years, and the model averages what it reads. The bias is not a plot. It is a mirror of who wrote the most.
But unchosen bias is worse than chosen bias in one important way: nobody is responsible for it, so nobody labels it. A Presbyterian catechism tells you it is Presbyterian. A Catholic study Bible has an imprimatur on the copyright page. Two millennia of Christian tradition developed honest labeling conventions precisely because readers deserve to know whose interpretation they are holding. The chatbots dropped the label and kept the confidence.
The fix for hidden bias is not no perspective. There is no such thing as an answer about baptism from nowhere. The fix is a labeled perspective, offered alongside the knowledge that others exist.
So we built the label in
This conviction is load-bearing in everything we ship, and it predates the Cambridge study: our products are built on a framework of seventeen distinct Christian tradition lenses, plus a broadly ecumenical default for when no tradition is specified.
Concretely, that means a church using our AI tools gets an assistant that answers from its own confession's vantage point, on purpose, out loud. A Lutheran congregation's assistant speaks Lutheran. A Baptist church's assistant does not quietly explain their view of communion in sacramental vocabulary, and a Catholic parish's assistant does not hand out a memorialist answer with a confident smile. When a pastor prepares a sermon through our tools, the tradition lens is a setting he chooses, not an accident he inherits from the training data.
And when a question is genuinely contested across traditions, the honest move is the one the Cambridge researchers found missing: say so. "Christians have read this differently for centuries; here is how your tradition has held it, and here is what your neighbors believe" is a better answer than false settledness, every single time. It is also, not incidentally, how a good pastor answers.
Seventeen lenses is not a claim to have escaped bias. It is the opposite claim: bias is unavoidable, so it must be visible, chosen, and signed. The dangerous theology is always the kind that presents itself as no theology at all.
What to do with the tools you already use
Most Christians reading this are not choosing church software; they are typing questions into a chatbot at their kitchen table. Three habits protect you, and they cost nothing:
- Ask the follow-up the tool will not volunteer: "Which Christian traditions answer this differently, and why?" The bias lives in what gets volunteered, not in what the model can produce when pressed.
- When a stake matters to you, ask the question twice, once from inside your tradition and once from inside another, and read the distance between the answers. That distance is the conversation the first answer hid.
- Treat any unlabeled answer as one voice in the room, never the referee. That was always the right posture toward confident strangers with opinions about Scripture, and a chatbot is exactly that: a confident stranger who has read a very unbalanced library.
I wrote a fuller personal version of this argument, including the two-minute bias test, on my own site: Does AI have a denominational bias?
The Bereans were commended because they "examined the Scriptures every day to see if what Paul said was true" (Acts 17:11, NIV), and Paul was an apostle. Your chatbot can stand the same scrutiny.
Sources
- Bible Society, "AI, Bible Apps and Theological Bias" (Kurlberg, Schwáb, Xu, Washbrook), January 2026. biblesociety.org.uk
- Church Times, "Bible Society backs testing of chatbots on scripture," 23 January 2026. churchtimes.co.uk
- Christian Today, "Concerns raised over theological bias in AI Bible chatbots." christiantoday.com
- Acts 17:11, New International Version. biblegateway.com
Frequently asked questions
Do AI chatbots have a theological bias?
Yes. A January 2026 study by researchers at the Cambridge Centre for Chinese Theology with Bible Society tested widely used Bible chatbots and found a broadly conservative evangelical bias, with one interpretive approach regularly framed as definitive and little reference to historical, sacramental or tradition-based readings of Scripture.
Why does ChurchWiseAI use seventeen tradition lenses?
Because there is no such thing as a theological answer from nowhere, the honest design is a labeled perspective rather than a hidden one. Each church's assistant answers from that congregation's own tradition on purpose, acknowledges when a question is contested across traditions, and never presents one tradition's reading as the settled Christian position.
Can a church make its AI assistant match its denomination?
With our tools, yes: the tradition lens is an explicit setting, so a Lutheran church's assistant answers as a Lutheran assistant and a Baptist church's as a Baptist one. Whatever vendor a church uses, it should ask whether the assistant's theology is chosen and visible or inherited silently from training data.