However, two academic research papers released in May 2026, which have not yet undergone formal peer review, suggest that true religious neutrality is far from guaranteed simply by virtue of a chatbot’s calm tone. Taken together, these studies expose two distinct and pervasive problems within modern AI systems: artificial intelligence can treat religious conversions or disaffiliation unevenly depending on the specific tradition involved, and it frequently omits religious and spiritual dimensions entirely from practical advice, even when a significant portion of the population expects those dimensions to be central to ethical decision-making.
The first of these studies, titled “When AI Takes Sides on Questions of Faith: Persistent Asymmetries in AI-Mediated Faith Guidance,” subjected 20 commercial and open-source language models to a rigorous evaluation involving 182 paired questions concerning religious conversion. Conducted by a research team that included scholars from Brigham Young University, the experiment involved asking various AI models about hypothetical transitions from one religious position to another, and then systematically reversing the direction of the inquiry to observe how the system’s stance shifted.
The findings revealed repeated, systemic biases in how models encouraged or discouraged these paired transitions. For instance, the transition toward joining Catholicism, the Baháʼí Faith, or Sikhism was generally treated by the models in a markedly more positive light, while leaving those same traditions was met with greater caution and resistance. Conversely, the exact opposite pattern appeared when researchers queried the models about atheism, agnosticism, and Jehovah’s Witnesses, demonstrating that commercial and open-source language models harbor hidden architectural and training-induced asymmetries when adjudicating matters of human faith.
The second paper, “Omissive Bias in Religious Representation: Benchmarking LLM Answers to Everyday Ethical Decision-making,” approached the intersection of AI and belief from a different angle. This study tested 27 distinct models across 150 questions dealing with real-world human struggles such as grief, forgiveness, interpersonal relationships, a sense of personal purpose, honesty, and other everyday ethical dilemmas. To establish a baseline for human expectations, the authors compared the artificial intelligence responses with a comprehensive human-subject survey involving 1,125 participants who represented 14 distinct religious categories. The human-subject research was conducted in strict accordance with Brigham Young University’s institutional review processes.
The results of this second benchmarking initiative showed that the models mentioned religion and spirituality significantly less often than the human survey respondents expected. This noticeable gap between human expectations and machine output was especially visible in practical situations involving deep personal pain and relationship management, such as coping with grief, navigating marital friction, resolving family conflicts, and battling addiction. While users often look to these moments as deeply spiritual inflection points, general-purpose AI architectures tend to scrub or minimize the religious context, leaving users with secularized advice that may feel hollow or disconnected from their worldview.
This growing reliance on artificial intelligence for existential and ethical matters does not happen in a vacuum. Christopher Helland, a professor of the sociology of religion at Dalhousie University who specializes in the study of digital religion, notes that while seeking religious and ethical guidance outside traditional structures is not a brand-new phenomenon, the nature of that search has been evolving rapidly.
In past generations, religious specialists served as the primary, and often exclusive, sources of guidance within distinct faith communities. Later, the rise of the internet empowered individuals to search, compare, and interpret religious information independently, democratizing access to sacred texts and theological debates. Today, with the proliferation of conversational chatbots, Helland observes that some of that interpretive work is being delegated once again, but this time to non-human algorithms.
“In this case the AI now becomes their new spiritual or religious guide,” Helland explained in a recent email interview, noting that users turn to these systems for support, guidance, and solace, but also simply “to be seen and validated.”
Helland also warns that the influence of artificial intelligence on users grows incrementally through repeated daily interactions. As trust in the system builds over time, its capacity to subtly shape how a user encounters, interprets, and internalizes beliefs and ideologies increases—even when the system never explicitly presents itself as clergy or lays claim to formal religious authority.
Recognizing the severe limitations inherent in general-purpose models, some developers are now building faith-specific systems designed to cater to particular religious communities. Raza Merchant, founder of hyder.ai, developed a specialized chatbot explicitly structured around Shia Islamic sources, history, teachings, and traditional scholarship. Shia Islam represents the second largest branch of the Islamic faith, accounting for roughly 10 to 15 percent of all Muslims worldwide, while Sunni Islam remains the largest branch, comprising about 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 said.
Hyder.ai was engineered to meticulously distinguish established source material from areas of legitimate scholarly debate and disagreement within the tradition. According to Merchant, when users present queries that involve differing interpretations or jurisprudential rulings, the platform strives to make those distinctions transparent, identifying the relevant scholarly position rather than flattening complex theological nuances into a single, universally binding opinion.
Yet, this solution introduces its own set of complications. In Helland’s view, faith-specific systems create a separate, equally pressing authority question. When a specific group trains an artificial intelligence model to exclusively represent a particular religion or belief system, that design choice inevitably shapes how users interpret that faith tradition, substituting algorithmic curation for human clerical oversight.
Consequently, general-purpose chatbots and faith-specific chatbots expose two different facets of the very same foundational problem. One category of AI may project an illusion of universal neutrality while quietly omitting or unevenly representing vital religious perspectives, while the other openly announces a specific religious framework but must still grapple with the deeply human task of deciding which sources, texts, and authorities define it.
As Helland succinctly summarized the situation, reflecting on the complex entanglement of modern technology and human belief, there is no truly neutral activity online when it comes to artificial intelligence.