Paper Type
Complete
Paper Number
PACIS2025-1686
Description
LLMs have attracted widespread attentions due to remarkable problem-solving capabilities. However, in some scenarios, such as open-ended tasks, users still show low adoption intentions and satisfaction with LLMs. Compared to the sustained efforts to improve information quality, the impacts of users’ cognitive bias have been largely ignored. Drawing on confirmation bias theory, this study investigates the interaction effects of expectation (in)consistency and information quality on perceived trust, cognitive load, and subsequent users’ acceptance of LLMs. A 2 (expectation consistency / inconsistency) ×2 (high-quality / low-quality information) between-subject experimental design was conducted to verify theoretical model. Our findings reveal that expectation consistency leads to lower cognitive load and information quality weakens the negative impact of expectation consistency on cognitive load. More interestingly, expectation consistency improves perceived trust when information quality is low, whereas it decreases perceived trust when information quality is high. Finally, the theoretical contribution and practical implications were discussed.
Recommended Citation
Li, Yingying; Wang, Ying; and Sun, Yongqiang, "Improving Quality or Catering to Users? Understanding Confirmation Bias in Large Language Model Interactions" (2025). PACIS 2025 Proceedings. 11.
https://aisel.aisnet.org/pacis2025/is_adoption/is_adoption/11
Improving Quality or Catering to Users? Understanding Confirmation Bias in Large Language Model Interactions
LLMs have attracted widespread attentions due to remarkable problem-solving capabilities. However, in some scenarios, such as open-ended tasks, users still show low adoption intentions and satisfaction with LLMs. Compared to the sustained efforts to improve information quality, the impacts of users’ cognitive bias have been largely ignored. Drawing on confirmation bias theory, this study investigates the interaction effects of expectation (in)consistency and information quality on perceived trust, cognitive load, and subsequent users’ acceptance of LLMs. A 2 (expectation consistency / inconsistency) ×2 (high-quality / low-quality information) between-subject experimental design was conducted to verify theoretical model. Our findings reveal that expectation consistency leads to lower cognitive load and information quality weakens the negative impact of expectation consistency on cognitive load. More interestingly, expectation consistency improves perceived trust when information quality is low, whereas it decreases perceived trust when information quality is high. Finally, the theoretical contribution and practical implications were discussed.
Comments
Innovation