Transparency is treated as a core principle in AI governance, yet consensus on its definition, audience, and context is lacking. Drawing on the Social Construction of Technology framework, this paper examines how various stakeholders perceive and experience transparency in generative AI.
A global survey of 400 end‑users, AI builders, and policymakers was conducted, complemented by a structured focus group with end‑users to provide qualitative depth.
- All three groups share a similar basic notion of transparency, centered on accuracy and privacy protection.
- The gap between the importance of transparency features and their perceived availability is largest for end‑users: only 28% feel current practices meet their needs well, versus 53% of builders and 49% of policymakers.
- Transparency is situational; priorities shift with task context, activating role‑specific concerns when needed and receding otherwise.
Relying solely on uniform AI disclosure or static documentation cannot satisfy diverse requirements. Pluralistic transparency infrastructures are needed to reflect the concrete needs of varied users and stakeholders.
Review: The study underscores that effective transparency design must account for power dynamics and contextual usage, moving beyond one‑size‑fits‑all solutions.