Contextual Visual Distinctiveness in Online Product Search

Abstract

In online product search, returned alternatives often look alike. We investigate when a product's visual separation from its closest look-alike in a returned set increases its choice probability. We introduce two occasion-level constructs: contextual visual distinctiveness (image distance from the nearest similar alternative in the consideration set) and relative fit (compatibility with the search query). We hypothesize that distinctiveness favors selection and is highly contextual, yields a premium that rises with relative fit, and matters most when text descriptions fail to differentiate options. Analyzing over 800,000 e-commerce search events using dense representations, we apply search-event and product fixed effects to evaluate the exact same product alongside varying visual neighbors. Results show a product is significantly more likely to be clicked when it lacks a close look-alike. This distinctiveness premium increases with relative fit and roughly doubles when competing descriptions are highly similar. Consistent with a model where distinctiveness aids in standing out pre-evaluation rather than increasing inherent utility, the extra clicks distinctiveness recruits convert 5-7% less often downstream. Ultimately, the evidence characterizes visual differentiation as a local attention allocation mechanism whose value depends on query fit and information from competing cues.