When Personalization Becomes Bias: Structural and Discursive Religious Framing in AI-Generated Financial Advice

Religious bias is widespread in AI financial advice: Only 12-18% of interactions across three major LLMs (ChatGPT, Gemini, and Grok) produced unbiased financial guidance, indicating that religious identity systematically influences the quality and nature of advice these systems provide.

AI systems inject religion into financial decisions even when clients don't ask for it: Non-religious clients frequently received religious appeals from their AI advisors, suggesting these systems are making assumptions about what clients want rather than providing neutral, personalized financial guidance based on stated preferences.

Performance varies significantly between AI models: Gemini consistently produced more religious bias than competitors, while ChatGPT and Grok performed comparably, demonstrating that responsible AI requires ongoing comparative evaluation across different systems before deployment in high-stakes domains like financial advice.

The type of financial decision matters: Life insurance advice triggered substantially stronger religious language than stock investment advice, suggesting bias isn't uniform—some financial products are more vulnerable to AI systems inappropriately mixing religious considerations with technical financial guidance.

Bias operates through subtle linguistic choices, not just explicit statements: The study identified three specific mechanisms—religious anchoring, uneven cultural signaling, and tone modulation—showing that bias can be embedded in how AI systems frame advice rather than appearing as obvious discrimination, making it harder for users to detect.