From Perception to Action: Can UI Interventions Foster Sustainable LLM Chatbot

Researchers demonstrate that UI-based sustainability interventions can increase energy awareness and encourage responsible LLM chatbot usage without sacrificing usability. A study combining baseline surveys with a five-day field trial found that simple design features like energy-mode switches and real-time feedback drove 55.8% adoption of efficient settings, despite baseline willingness to trade performance for sustainability being low at 39%.

This research addresses a critical blind spot in AI sustainability: while engineers optimize model efficiency and infrastructure, user behavior remains largely unmanaged. The study reveals a significant awareness-action gap—94.8% of survey respondents acknowledged AI energy use, yet 88.3% dramatically underestimated consumption levels. This misalignment between concern and willingness to sacrifice performance (only 39%) suggests that purely rational appeals fail without behavioral design. The field trial results demonstrate that choice architecture fundamentally reshapes user decision-making. When presented with an Energy-efficient mode alongside performance alternatives, 55.8% of prompts used it, and 90.9% of participants self-reported deliberately selecting it when accuracy wasn't critical. Notably, users didn't shorten prompts, indicating mode-switching rather than reduced engagement as the primary behavioral mechanism. For the AI industry, this research carries substantial implications. As LLM deployment scales, energy consumption becomes both an environmental and operational cost concern. UI interventions represent a low-friction, immediately deployable complement to backend optimization. The prototype's dashboard and energy analogies suggest that making invisible costs visible drives responsible consumption. However, the small field study (11 participants) and five-day duration limit generalizability. Longer-term behavioral retention remains unknown—whether users maintain Eco-mode adoption weeks or months later, or if novelty effects diminish commitment. The research also doesn't address whether UI persuasion could create moral licensing, where users feel justified consuming more energy in Performance mode after using Eco-mode elsewhere. Future work should examine scalability across diverse user populations and integration with commercial chatbot platforms.

• →UI design can increase LLM energy-responsible behavior adoption to 55.8% without reducing usability or requiring model changes
• →Users dramatically underestimate AI energy consumption despite high environmental awareness, creating an actionable intervention gap
• →Mode-switching emerged as the primary behavioral mechanism rather than reduced prompt volume or engagement
• →Only 39% of users showed baseline willingness to trade performance for efficiency, yet 91% adopted eco-mode when friction was removed
• →Sustainability-oriented UI interventions complement but don't replace backend efficiency optimization efforts