Abstract

Users frequently interact with products in ways that diverge from designer intent, yet this gap between intended and perceived affordances is rarely captured at scale. This paper introduces a data-driven framework leveraging large language models to extract and compare structured affordance representations from product documentation and online reviews. By embedding these representations in a shared semantic space, we quantify alignment between intended and perceived use. We demonstrate the approach in a case study related to sustainable design, analyzing two Kindle Paperwhite generations to identify how real-world user behaviors impact sustainability. The framework surfaced 2,756 perceived affordances, identifying critical mismatches such as high demand for physical page-turn buttons and unintended behaviors that drain battery life. These findings demonstrate that computationally comparing intended and perceived affordances at scale can provide designers with empirical, use-phase evidence to bridge the design intent - user perception gap.