From Prompt to Product: Reimagining Visual Search with Generative AI

Abstract

Traditional e-commerce searches rely on keywords and filters, which often fail to fully capture consumer preferences comprehensively, leading to suboptimal product search results. This research proposes a generative AI-enabled search system that creates visual representations of consumers' desired products and matches them with the best options from tens of thousands of available products. Three experimental studies affirm that exposing consumers to AI-generated visualizations of their textual product descriptions increases both purchase intentions and design satisfaction with the displayed product matches. Prompt adherence-the extent to which consumers believe the input they provide is accurately represented and aligned with the results-functions as a mediating mechanism, though this mediation occurs only if actual prompt adherence is high. Therefore, the effectiveness of AI-powered visual search systems depends on their ability to generate accurate visualizations and display accurate product matches. A 2×2 experiment further disentangles these effects, revealing that while both visualization accuracy and product match accuracy independently enhance purchase intentions and design satisfaction, visualization accuracy exerts a stronger influence. These findings deepen theoretical insights into online product searches by demonstrating how visual feedback helps consumers feel heard, while also providing practitioners with actionable insights to increase purchase likelihood using generative AI-powered visual search systems.

Keywords: generative AI, product search, e-commerce, image analytics, prompt adherence

JEL Classification: M3, M30, M31