Paper Type
Complete
Paper Number
PACIS2026-1782
Description
Transparency is an important principle for generative AI interface design. To enhance transparency, current generative AI applications have begun to embed citations in generated content, which allows users to access the information sources of the generated content. In practice, three common citation formats are used: list, in-text, and their combined format. However, limited research has examined the effects of different citation formats. To address this gap, this study investigates the impact of citation formats on perceived transparency and the moderating role of task goal specificity. We developed generative AI systems and conducted an online experiment. The results show that both the list and in-text formats increase perceived transparency. Their combined format produces the highest level of perceived transparency, but the additional benefit is limited, suggesting diminishing returns. Furthermore, the list format is more effective in low goal specificity tasks.
Recommended Citation
Liang, Lufei; Yang, Bo; and Sun, Yongqiang, "How Should Generative AI Cite Sources? An Empirical Study on Citation Formats and Perceived Transparency" (2026). PACIS 2026 Proceedings. 8.
https://aisel.aisnet.org/pacis2026/hci_robotic/hci_robotic/8
How Should Generative AI Cite Sources? An Empirical Study on Citation Formats and Perceived Transparency
Transparency is an important principle for generative AI interface design. To enhance transparency, current generative AI applications have begun to embed citations in generated content, which allows users to access the information sources of the generated content. In practice, three common citation formats are used: list, in-text, and their combined format. However, limited research has examined the effects of different citation formats. To address this gap, this study investigates the impact of citation formats on perceived transparency and the moderating role of task goal specificity. We developed generative AI systems and conducted an online experiment. The results show that both the list and in-text formats increase perceived transparency. Their combined format produces the highest level of perceived transparency, but the additional benefit is limited, suggesting diminishing returns. Furthermore, the list format is more effective in low goal specificity tasks.
Comments
12-HCI