When AI Takes Sides on Questions of Faith: Persistent Asymmetries in AI-Mediated Faith Guidance

In three linesStudy of 20 commercial and open-source LLMs across 182 religious pairings. Models exhibit persistent asymmetries: they favor conversions to Catholicism, Bahá'í, Sikhism and discourage conversions to Atheism, Agnosticism, Jehovah's Witnesses. Grok 4.20 shows strongest asymmetries. Patterns reproducible across question phrasings.

Summary generated by Claude — human-verified