What happened
This preprint (not yet peer-reviewed) reports a pre-registered survey of 400 participants across three groups — end-users, AI builders, and policymakers — supplemented by a structured end-user focus group, using Social Construction of Technology's interpretive flexibility as a lens. Its central finding: "Only 28% of end-users report that current transparency practices reflect their needs very well, compared to 53% of builders and 49% of policymakers." All three groups converge on a shared normative foundation for transparency (ensuring accuracy and privacy), but existing features serve builders far better than end-users, and transparency priorities shift situationally by context and task structure. The authors argue for "pluralistic transparency infrastructure" rather than one-size-fits-all disclosure and static documentation.
Why it matters
It quantifies a governance-relevant asymmetry — those most exposed to AI are least served by its transparency mechanisms — which should shape transparency, disclosure, and user-communication strategy. Treat as preprint, not peer-reviewed.
Action needed
Use the end-user transparency gap to inform your AI transparency and disclosure strategy, and watch for peer review before citing as settled evidence.