Whose Cultural Default? Attunement Depth, Relational Positioning, and Interpretive Authority in Conversational AI
Conversational AI increasingly helps people interpret relationships and navigate social expectations. Research on cultural alignment has improved models’ cultural knowledge, yet offers limited ways to examine how that knowledge shapes individual responses. We propose a theoretically grounded, response-level analytical framework with two dimensions. Attunement Depth examines how culturally situated meaning shapes a response’s reasoning. Relational positioning identifies the cultural frames guiding that reasoning. We further conceptualise Unmarked Normative Defaulting (UND) as a complementary diagnostic for identifying responses that present a situated norm as common sense despite relevant alternative interpretations. We refine the framework through two concept checks: 60 controlled items assessed by human researchers with exploratory AI coding, followed by 24 naturalistic and augmented dialogue cases drawn from WildChat-1M and WildChat-4.8M. These analyses clarify the boundaries of cultural attunement and show how apparently sup- portive responses can privilege particular interpretations of users’ experiences. We discuss the implications for interpretive authority and user agency. The framework provides a basis for future research into whose experiences receive meaningful cultural understanding and who bears the burden of correcting AI.
Authors
- Yao Xiao (ORCID: https://orcid.org/0009-0008-5729-8502)
- Lu Meng
- Minseo Cho
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24429490
- Primary Topic
- Ethics and Social Impacts of AI
- Type
- preprint