Anthropology-informed AI: modeling the spiritual-first hierarchy of illness explanation in Arabic-speaking patient communities

Introduction AI-mediated health communication lacks scalable, evidence-based models of cultural tailoring derived from authentic patient narratives. Methods We developed and validated a hybrid human-AI methodology. A large-scale analysis of 41,393 posts from Arabic-language online mental health forums was conducted. Manual annotation using Kleinman’s Explanatory Model (EM) was followed by fine-tuning a large language model (LLM) to automate this interpretive framework. Results The analysis revealed a distinct, culturally grounded hierarchy: Arabic-speaking patients consistently prioritized divine/spiritual causes first (e.g., God’s will, divine testing, evil eye, jinn), followed by moral/ethical considerations (sin, guilt, repentance), social/familial impact (honor, stigma, reputation), psychological factors, and finally biomedical symptoms. This “spiritual-first” pattern was experimentally tested by comparing spiritually-attuned AI messages versus standard secular biomedical messages. Discussion Spiritually-aligned messages significantly outperformed secular counterparts across all engagement metrics and qualitative indicators of trust, relevance, and intent to seek care, validating that communication respecting patients’ spiritual explanatory models fosters greater resonance and therapeutic alliance. A key limitation is the exclusive focus on mental health narratives; future work should investigate extension to physical illness contexts and clinical outcomes.