Measuring Consumer Visual Interest Using Millions of Text and Image Search

Abstract

Despite the importance of visual information for consumer decision-making, quantifying consumer visual interest for brands and products across markets and over time remains arduous. This paper introduces Excess Image Search (EIS) as a new behavioral, high-frequency, time- and space-varying, and easy-to-implement measure of visual interest. Analyzing about 134.8 million records of online searches in the U.S. car market (55 brands, 659 car models), we probe EIS’s internal validity by showing that a brand’s or product’s EIS strongly correlates to measures of visual salience at the brand level (e.g., consumer ratings of brand visibility) or product level (e.g., style, logo). Turning to EIS’s external validity, we examine how the local presence or relevance of visible consumption affects a brand’s or product’s EIS. First, shocks weakening visible consumption locally, namely (1) COVID-19 lockdown-induced social distancing and (2) adverse weather, trigger a drop in EIS in high- (not low-) density states. Second, we estimate the sensitivity of EIS to changes in price signals in the used car market. A rise in product price increases the product’s EIS the following week, an effect arising in unequal states, where visible consumption is more sought after. Our main results replicate with a sample of apparel brands.

Keywords: visual interest, big data, online search, competitive analysis