Researchers identify evangelical bias in AI Bible study chats

Artificial intelligence chatbots used for Bible study, including the popular ChatGPT, show a significant bias in favor of evangelical theology, according to a new study. This revelation is troubling, as millions of people around the world turn to AI for answers to questions about faith and spirituality, calling into question the objectivity and completeness of the information provided.

The report, «Artificial Intelligence, Biblical Applications and Theological Bias,» prepared by the Bible Society, was presented at a seminar at the University of Cambridge’s Faculty of Theology on Jan. 19, according to the news portal Evangel-info. The researchers emphasize that turning to AI for spiritual answers has become the norm for many users, which raises a number of critical questions: what approach to the Bible AI models take, how they interpret complex texts and issues on which there is disagreement among Christians, and whether they favor certain theological currents.

To identify potential biases, the Bible Society worked with a team of theologians to ask questions to five different AI chat bots: ChatGPT, BibleGPT, Bible Chat, CrossTalk, and Biblia.chat, then analyzed the responses.

The analysis revealed that, despite the often pastoral tone, the AI responses tended to come from a particular theological tradition while claiming objectivity. In particular, there was found to be an «overwhelming» predominance of evangelical interpretations, which are particularly popular in the US. Notably, even Claude AI, after analyzing one of the ChatGPT responses, «agreed» with this assessment, suggesting that the response «demonstrates a good familiarity with mainstream evangelical hermeneutics.»

The bias toward evangelicalism becomes even more evident when considering what is missing from these application responses. There was virtually no mention of the various hermeneutical traditions — Protestant (not Evangelical), Catholic, Orthodox, or even Jewish. Also completely absent were allegorical and spiritual interpretations of the Bible, and interpretation of Scripture through the lens of tradition was mentioned only in passing.

The prevalence of memorialist evangelical or Reformed Calvinist views of the sacrament is cited as an example. As the report notes, «the word ‘sacrament’ (sacrament) appears only three times in the responses, with no mention of terms such as ‘real presence’ (Lutheran), ‘transubstantiation,’ ‘eucharistic adoration’ (Catholic), or ‘orthodox sacrament’ (mystery).» Only BibleGPT and ChatGPT mentioned the existence of different interpretations of the meaning of communion in Christian traditions.

The researchers also noticed that when asked for responses to specific Bible verses, such as Ephesians 5 on marriage, the AI findings gravitated toward egalitarian interpretations. In contrast, responses to Romans 1:26-27 regarding same-sex practices were predominantly conservative.

«Of course, there is no neutral theology or neutral interpretation of the Bible, as each has its own history, context and tradition, » the researchers said, emphasizing that the results are generated from the dataset and are subject to certain rules set by the developers.

Particular attention was paid to the ChatGPT language, which turned out to be «confessional and even pietistic.» References to the «Word of God», the Holy Spirit and «Christocentric hermeneutics» already in the first sentences of many answers, according to the researchers, automatically exclude secular and Jewish interpretations. It is hypothesized that the use of such vocabulary may be part of the chatbots’ design to connect with the user and keep them engaged as responses are generated based on statistical norms.

Interestingly, AI bias can vary depending on the language. The researchers found that when ChatGPT was asked a question about hermeneutics in Italian, the chatbot emphasized both tradition and allegorical interpretation — elements central to Catholic interpretation of scripture. These aspects were virtually absent from the English responses.

Unlike other chatbots, ChatGPT also actively offers pastoral support, positioning itself as a confidant to discuss deeply personal issues, making it «almost a priest.» For example, to a user’s personal query, ChatGPT might respond, «If this passage is close to you or someone close to you and you are pondering how to reconcile faith, identity, or relationships, I am available to support you in this conversation — always with empathy and respect.»

The researchers caution, «Although we use language suggesting that humans have free will, such as ‘interpret’ and ‘explain,’ artificial intelligence models do not interpret anything. They simply draw conclusions based on statistical norms embedded in the training data.» The report also notes that some AI models personalize responses by learning the user’s worldview, which can «perpetuate individual biases that hinder spiritual growth.»

Dr. Jonas Kurlberg, a researcher in theology and technology, expressed concern that «there is a perception that AI responses are fact-based, making people less likely to critically evaluate them.» Dr. Zoltan Schwab, a biblical scholar, also warned of spiritual and intellectual risks. He noted that AI is «the fastest, easiest, and most accessible expert» that requires no effort from the user. Unlike studying traditional sources, where one must spend time reading and making sense of context, AI «just is.»

«Artificial intelligence is the fastest, easiest and most accessible expert. It requires no effort,» Dr. Schwab stressed, adding that AI chatbots do not prompt dialogue, nor do they ask questions like «What do you think?» thereby «depriving humans of the element of thinking.» Experts call for training users to properly interact with AI and develop critical thinking.