Policy alone is probably not the solution: A large-scale experiment on how developers struggle to design meaningful end-user explanations

Policy compliance doesn't guarantee explanation quality: In experiments with 194 developers, stronger policy enforcement increased formal compliance but explanations "largely remained poorly suited to medical professionals and patients," suggesting that meeting regulatory requirements doesn't automatically produce meaningful explanations for end users.

Developers need specialized training, not just guidelines: Typical developers without human-centered AI training repeatedly produced explanations that were "technically flawed or difficult to interpret" and "framed for developers rather than end users," indicating that policy guidance alone cannot substitute for proper education in explanation design.

Developer-centric framing is the most prevalent problem: Across both experiments, the most common issue was that developers explained AI systems in ways suited to other technical experts rather than the actual end users (medical professionals and patients), suggesting a fundamental misalignment between who developers design for and who needs to understand the system.

Policy guidance is frequently ignored: In the first experiment, "policy guidance was often ignored and explanation quality remained low" despite clear policy differences, indicating that developers may not prioritize or follow transparency requirements without additional support mechanisms.

Responsible AI requires more than regulatory frameworks: The findings suggest that "policy and policy enforcement alone are insufficient to produce meaningful end-user explanations," pointing to the need for complementary approaches such as developer training, design resources, and integration of explanation design into standard development workflows.