Reducing SMiShing Vulnerability: Glanceable Recommendations Versus Attributive XAI

Artificial Intelligence is increasingly integrated into high-stakes decision-making, yet its underlying reasoning often remains opaque to end users. Explainable AI (XAI) tries to bridge this gap by communicating the rationale behind system recommendations in ways users can understand and act upon. SMiShing, or SMS-based phishing, is a compelling application area for XAI given the rapid, high-pressure nature of the task and potentially serious consequences of a wrong decision. This study investigated the effectiveness of an attributive XAI intervention on message classification accuracy under three conditions: No AI, Recommendation, and Attributive explanation. Both AI-assisted conditions surpassed unaided judgment; however, overall accuracy remained alarmingly low across the board. Adding feature-based explanations provided no measurable benefit beyond a simple binary recommendation, suggesting that users are not effectively leveraging explanatory content in this context. Findings highlight the difficulty of SMiShing detection and the need for continued development of effective AI-assisted interventions.