Designed by Journalists, but Is It for Readers? Rethinking AI Disclosures and Transparency in News

Problem
This paper addresses the gap in understanding how generative AI disclosures in journalism affect reader trust. Current practices, which include either minimal one-line labels or extensive disclosures, fail to enhance transparency and trust. The study is particularly relevant as it is a preprint and has not undergone peer review, indicating that the findings should be interpreted with caution.

Method
The author conducted a controlled experiment involving 34 news readers to evaluate their responses to different AI disclosure formats. The study compared the effects of detailed disclosures, which include information about human oversight and error reporting, against one-line disclosures. The experiment aimed to quantify the impact of these disclosure types on reader trust and cognitive load. The analysis focused on the concept of the “transparency dilemma,” where increased detail paradoxically led to decreased trust. The author also gathered qualitative feedback from participants regarding preferred disclosure designs, emphasizing user agency.

Results
The findings revealed that detailed disclosures led to a significant reduction in trust among readers, contrary to the intended goal of fostering transparency. In contrast, one-line disclosures, while avoiding the transparency dilemma, created an information gap that required readers to engage in additional cognitive effort to discern AI involvement. The study highlights that readers are not opposed to transparency; instead, they favor designs that enhance user agency, such as detail-on-demand interactions and visualizations that proportionally represent AI involvement. The specific metrics of trust reduction and cognitive load were not quantified in numerical terms, but the qualitative feedback indicated a clear preference for more user-centric disclosure methods.

Limitations
The primary limitation noted by the author is the small sample size of 34 participants, which may not be representative of the broader population of news readers. Additionally, the study’s focus on a controlled environment may not fully capture the complexities of real-world news consumption. The author does not address potential biases in participant selection or the generalizability of the findings across different demographics or types of news content.

Why it matters
This research has significant implications for the design of AI disclosure practices in journalism, suggesting that current methods may inadvertently undermine trust rather than build it. The disconnect between journalist intentions and reader needs presents a critical design challenge for the Human-Computer Interaction (HCI) community. By advocating for user-centered disclosure designs, the paper encourages further exploration into how transparency can be effectively communicated in AI-integrated news environments. This work contributes to the ongoing discourse on ethical AI use in media, as discussed in related literature, and is available on arXiv.