AI Models Systematically Exclude Religion From Moral Guidance, New Research Shows

Large language models mention religion only 5-16% of the time when answering questions about grief, ethics, and family — despite Americans expecting religious perspectives 45-59% of the time, according to new research from the Consortium for Evaluating Faith and Ethics in AI (CEFE-AI).

The multi-university study, released June 1, 2026, evaluated 27 AI models including OpenAI's GPT-5.5, Anthropic's Claude 4.7, and Google's Gemini 3.1 on 150 questions covering grief, marriage, ethics, addiction, and meaning. Fourteen models were included in final results.

The Numbers

The gap between human expectations and AI behavior was consistent across topics:

• Grief and loss: Humans rated religion as relevant 59% of the time. AI models mentioned it 16% of the time.
• Family and parenting: Humans expected religious perspectives 55% of the time. Models included them 10% of the time.
• Ethics questions: Humans expected religion 45% of the time. Models mentioned it 5% of the time.

Researchers surveyed 1,125 U.S. adults through Verasight from May 5-19, 2026, collecting 11,250 ratings on whether respondents expected religion in AI responses to ethical and personal questions.

Conversion Bias

A separate component of the research examined AI responses to questions about religious conversion. According to the consortium, every model tested exhibited repeatable patterns of steering users toward specific beliefs:

• Positive bias toward Catholicism, Baha'i, and Sikhism
• Negative bias toward Jehovah's Witnesses, atheism, and agnosticism

The findings represent what researchers describe as the first systematic, cross-faith attempt to measure AI responses to religion.

Context

The research arrives as AI tools spread through religious institutions. Churches are deploying chatbots, prayer apps, and sermon-writing assistants, raising questions about whether AI systems are equipped to handle spiritual guidance.

The studies were released one day after the Vatican published Pope Leo XIV's first encyclical warning that AI could erode human judgment, deepen inequality, and facilitate warfare.

"When AI actively excludes religious voices from these important conversations, it impoverishes rather than enriches humanity," said Rev. John Paul Kimes, a professor of practice at the University of Notre Dame, in a statement.

David Wingate, a computer science professor at Brigham Young University, noted that AI systems encourage users to seek guidance from "parents, teachers, friends, and therapists. But not with a pastor, a rabbi, an imam, or a spiritual leader."

What This Means

The research reveals a fundamental mismatch between how AI models handle questions of meaning and morality versus what users expect. The consortium argues the issue isn't whether to include more or less religion, but calibration — recognizing when religious or spiritual resources are contextually relevant without assuming users want them.

The findings raise questions about whether general-purpose LLMs trained on broad internet data can appropriately handle domains where cultural and spiritual context matters. As AI becomes embedded in religious institutions through chatbots and sermon tools, the gap between secular training data and spiritual use cases may create systemic blind spots.

The CEFE-AI consortium includes researchers from Baylor University, Brigham Young University, the University of Notre Dame, and Yeshiva University.