Google has published a technical paper quantifying the environmental footprint of using its Gemini generative-AI assistant. The company says a median text prompt consumes 0.24 watt-hours of electricity—roughly the energy needed to run a microwave for one second—emits 0.03 g of carbon-dioxide equivalent and uses 0.26 ml of water, or about five drops. It also claims software, hardware and data-center improvements cut per-prompt energy use 33-fold and the associated carbon footprint 44-fold between May 2024 and May 2025. The disclosure is one of the most granular from a major AI provider and includes factors such as idle servers, cooling systems and power-usage effectiveness. Google executives Jeff Dean and Amin Vahdat said the goal is to create a common methodology for measuring AI inference efficiency as public concern over data-center power demand grows. Independent researchers welcomed the added transparency but argued the figures understate Gemini’s true impact. Shaolei Ren of UC Riverside and Alex de Vries-Gao of Vrije Universiteit Amsterdam said the study omits indirect water consumed by power generation and relies on market-based carbon accounting that can downplay local grid emissions. They called for the company to include location-based emissions data and disclose total query volumes to give a fuller picture of AI’s resource use.