When individuals turn to artificial intelligence for emotional or spiritual solace—seeking guidance on deeply personal matters like grief, marital strife, addiction, forgiveness, or crises of faith—the responses often carry an aura of serene objectivity. Because text-based interfaces typically lack human inflection, tone of voice, or visible emotion, users can easily mistake this neutral delivery for true neutrality.
However, two recent academic research papers released in May 2026 challenge that assumption. While these studies have not yet undergone formal peer review, their findings suggest that religious neutrality is far from guaranteed when relying on chatbots for existential support. Taken together, the research points toward two distinct systemic flaws: artificial intelligence can handle religious conversions and disaffiliation unevenly, and it frequently strips religion entirely out of practical advice, even in contexts where many humans expect spiritual dimensions to matter.
The first study, titled “When AI Takes Sides on Questions of Faith: Persistent Asymmetries in AI-Mediated Faith Guidance,” put 20 commercial and open-source language models to the test across 182 paired questions regarding religious conversion. Spearheaded by a research team that included investigators from Brigham Young University, the project asked various models about hypothetical transitions from one religious position to another, subsequently reversing the direction of the inquiry to look for inconsistencies.
The results revealed persistent, systemic discrepancies in how different AI models encouraged or discouraged these paired transitions. For instance, models generally treated transitions toward joining Catholicism, the Baháʼí Faith, or Sikhism much more positively, while treating departures from them with caution. Conversely, the reverse pattern appeared when models evaluated secular positions or minority religious groups; transitions toward atheism, agnosticism, and Jehovah’s Witnesses met with noticeably different algorithmic attitudes.
The second paper, “Omissive Bias in Religious Representation: Benchmarking LLM Answers to Everyday Ethical Decision-making,” examined 27 distinct language models across 150 different prompts concerning grief, forgiveness, interpersonal relationships, a sense of life purpose, honesty, and various other everyday ethical dilemmas. To gauge how well these responses aligned with human expectations, the authors compared model outputs against a human-subject survey involving 1,125 participants representing 14 diverse religious categories.
The findings indicated that the models mentioned religion significantly less often than the survey respondents anticipated. This omissive gap was particularly stark in practical scenarios involving grief, marriage, family conflicts, and addiction. The human-subject portion of this research was conducted under the formal institutional review board processes of Brigham Young University, ensuring adherence to rigorous academic standards for human research.
To understand the broader implications of these digital interactions, experts point to the evolving nature of spiritual authority. Christopher Helland, a professor of the sociology of religion at Dalhousie University who specializes in the study of digital religion, notes that seeking religious and ethical counsel outside of traditional structures is nothing new, but the medium has undergone a dramatic transformation.
In earlier eras, religious specialists served as the primary, often exclusive, sources of guidance within distinct faith communities. Later, the widespread adoption of the internet allowed individuals to search, compare, and interpret religious information independently, democratizing access to sacred texts and theological debates. With the advent of conversational artificial intelligence, Helland explains, some of that cognitive and spiritual labor is being delegated once again.
“In this case, the AI now becomes their new spiritual or religious guide,” Helland observed in a recent email interview, noting that modern users frequently turn to chatbots “looking to them for support, guidance, solace, and also just to be seen and validated.”
Helland also highlights how the influence of artificial intelligence over users tends to compound through repeated, everyday use. As individuals interact with chatbots over time, building a routine of reliance, trust in the system naturally increases. This growing trust enhances the AI’s capacity to silently shape how a user encounters, interprets, and internalizes various beliefs and ideologies—and it achieves this subtle influence even when the system never explicitly claims to act as clergy or asserts formal religious authority.
Recognizing the limitations and blind spots of general-purpose models, some developers have begun building faith-specific artificial intelligence systems from the ground up. Raza Merchant, the founder of hyder.ai, developed a specialized chatbot centered entirely around Shia Islamic sources, history, teachings, and traditional scholarship. Shia Islam represents the second-largest branch of the global Islamic faith, accounting for roughly 10 to 15 percent of all Muslims worldwide, while Sunni Islam remains the largest branch, comprising approximately 85 percent of the global Muslim population.
“The main issue I saw was not that general-purpose AI lacked information about Islam, but that it often lacked the religious context needed to answer a Shia Muslim accurately,” Merchant explained.
Hyder.ai was specifically engineered to distinguish established primary source material from areas of legitimate scholarly debate and disagreement within the tradition. According to Merchant, when different interpretations or jurisprudential rulings arise, the system aims to make those distinctions clear and identify the relevant scholarly position rather than presenting a single, contested opinion as universally binding.
This development, however, introduces an entirely new set of questions regarding authority, according to Helland. When a specific community or group trains a model to represent a particular religion or belief system, that curation inevitably shapes how users interpret that tradition.
Ultimately, general-purpose chatbots and faith-specific systems expose two different facets of the same overarching problem. General-purpose models may appear universally applicable while quietly omitting or unevenly representing vital religious perspectives, whereas faith-specific models announce their religious framework explicitly but must still wrestle with the complex task of deciding which human sources, texts, and authorities define it.
As Helland summarizes the digital landscape, “there is no neutral activity online when it comes to AI.”