Survey of 400 stakeholders shows 28% end-user satisfaction with generative AI transparency versus 53% for builders

The paper draws on interpretive flexibility from Social Construction of Technology to map divergent transparency expectations. It combines a 400-person quantitative survey with a structured end-user focus group. Findings center on three shared priorities—accuracy, privacy, and contextual relevance—yet reveal persistent gaps in implementation. End-users experience the largest shortfalls between stated importance and perceived availability.

Data show role-specific frames activate under task pressure and recede otherwise. End-users prioritize verifiable output provenance and data lineage; builders emphasize internal documentation and audit logs; policymakers focus on disclosure mandates. The 25-point satisfaction gap between end-users and builders indicates power asymmetries determine which features reach deployment. Current static model cards and disclosure statements address builder and regulator needs more than end-user requirements.

Operationally this implies transparency infrastructures must shift from uniform disclosure to modular, queryable systems that surface different data slices by stakeholder and context. Pluralistic designs would include runtime provenance queries for users alongside builder-facing audit trails. Without such differentiation, compliance regimes will continue to over-serve documented actors while leaving the largest user cohort underserved.

Next steps include pilot deployments of context-aware transparency interfaces in production systems and longitudinal tracking of satisfaction deltas across stakeholder groups.