EcoPrompt V1.0: an offline energy-efficient prompt optimizer for sustainable health informatics artificial intelligence tools

The rapid expansion of conversational artificial intelligence (AI) usage in healthcare, nutrition, and digital health education has made prompting a routine part of professional and personal practice. However, each AI prompt consumes electrical energy and cooling water in remote data centers, a physical footprint largely invisible to end users. Existing Green AI work has largely focused on system-level efficiency (training, infrastructure, scheduling, and data-center optimization) rather than tools that influence everyday user prompting behavior. As a result, end users in health and nutrition contexts often lack practical, point-of-use feedback that links prompt verbosity to resource implications. This study aims to increase sustainability awareness in everyday AI interactions by introducing EcoPrompt V1.0, an offline tool that estimates the approximate energy and water associated with individual prompts and encourages more efficient prompting behavior. EcoPrompt V1.0 was developed as a lightweight, privacy-preserving prompt-efficiency tool that operates entirely offline. It estimates energy (Wh) and water (L) consumption per prompt using approximations based on publicly available data for modern AI model processing. The tool analyzes a user’s original prompt, displays the estimated resource cost, and then generates a shorter, clearer version intended to reduce computational load. To evaluate the approach, a dataset of 500 manually curated nutrition- and health-related prompts was compiled, covering topics such as diet advice, food substitutions, metabolic concerns, meal planning, and patient education. Across the dataset, simplified prompts reduced input token counts by approximately 30%, producing proportional reductions in the estimated prompt-processing energy and water footprint. Because the total resource use of an AI interaction also depends on the output length generated by the model, these estimates should be interpreted as reductions associated with prompt text processing, not the complete prompt–response inference cycle. Although the absolute numerical values were small, qualitative feedback indicated a meaningful behavioral shift: users reported increased awareness of AI’s physical resource demands and adopted more concise prompting practices. EcoPrompt V1.0 demonstrates that sustainability awareness can be introduced into digital health and nutrition workflows without complex instrumentation or real-time hardware monitoring. By making the otherwise invisible resource consumption of AI prompts visible, the tool encourages more mindful and responsible AI usage. This work offers a practical, privacy-friendly mechanism for embedding sustainability thinking into modern health-informatics education and daily clinical or personal prompting practices.