The Impact of Multimodal Search Results in Generative AI-powered Search Engines: Evidence from a Field Experiment

2026年7月6日（周一）10:00 – 11:30

信息管理与工程学院308会议室

Qinglai He （贺庆来）is an assistant professor and John and Anne Oros Term Professor at the Wisconsin School of Business, University of Wisconsin–Madison. She earned her PhD in information systems from the W. P. Carey School of Business at Arizona State University. Her research interests include human-AI interaction, incentives and creativity in user-generated content, content moderation, and online platform policy. Her research work has been published in premier journals such as Management Science, MIS Quarterly, and Information Systems Research. She has received numerous awards and honors, including the Management Science Best Paper Award (IS department) in Years 2022-2025, the Best Associate Editor at the 2024 International Conference on Information Systems, runner-up award for the 2021 ACM SIGMIS Doctoral Dissertation granted by AIS, and the junior faculty teaching and research awards from the Wisconsin School of Business.

The Impact of Multimodal Search Results in Generative AI-powered Search Engines: Evidence from a Field Experiment

The integration of generative AI into traditional search engines has given rise to GenAI-powered search engines in recent years and enabled them to deliver multimodal outputs that combine text and images. Despite the rapid growth in user adoption and investment interests, empirical research on user behavior in real-world GenAI-powered search engines remains limited. Leveraging a field experiment conducted by a leading GenAI-powered search engine in Asia, we examine how multimodal outputs (i.e., search results incorporating both text and images) affect users’ subsequent search volume and query diversity. Our findings indicate that the introduction of multimodal outputs does not lead to significant changes in overall search volume. However, the effects vary across users with different usage intents. Specifically, users primarily engaged in creation-oriented tasks exhibit an increase in search volume following the introduction of multimodal results. In addition, our results also show that multimodal outputs reshape user behavior by encouraging more creation-oriented searches. We further conduct preliminary analyses of factors that potentially drive the observed effects, including the stimulation of content exploration, image aesthetics, and informational explicitness. Finally, we discuss the practical implications of our findings for the design and operations of GenAI-powered search platforms.