Abstract

Design for sustainable behavior (DfSB) encompasses a range of intervention strategies aimed at promoting pro-environmental behaviors by making otherwise invisible environmental costs salient and actionable. The widespread integration of generative AI (GenAI) into daily workflows creates a relevant application to study how intervention modality influences behavior and attitude change, especially as AI use increasingly contributes to carbon emissions and resource consumption. This paper explores how two DfSB strategies (eco-feedback and eco-steering) can be embedded into GenAI interactions to shape user awareness of AI use and invoke reflective behavior, specifically around environmental impact. To this end, we adopt a research through design approach, treating the development of novel interactive prototypes as a form of inquiry into how augmenting sustainability information shapes user behavior. We first design a browser-based eco-feedback plugin that visualizes real-time energy consumption associated with GenAI usage, establishing information as the foundation of environmental awareness. We then introduce a dynamically adaptive keyboard that modulates typing resistance in real time based on token usage during a research task – augmenting that informational baseline with a physical layer of engagement. Through a user study (n=30), we examine how these interventions influence patterns of AI use, perceived value of AI assistance, moments of reflection, and shifts in attitudes toward the sustainability implications of GenAI. Our findings demonstrate that embedding DfSB strategies into GenAI workflows can meaningfully shift user awareness, with participants weighing environmental cost against output quality and engaging more deliberately with AI, a process shaped by slow technology principles. The varied ways participants responded to physically-augmented versus purely digital feedback highlights the importance of modality diversity in reaching a wide range of users, while a consistent gap between awareness and available tools points to open questions for future sustainable AI interface design.