Still, Google presents the study as a comprehensive look at AI energy use in real-world conditions. The key stats: 0.24 watt-hours, 0.03 grams of CO₂, and about five drops of water (0.26 milliliters) per text prompt.
All the things Google didn't count (it's a lot)
While Google's move toward transparency is notable, the study leaves out major details. The biggest omission: it only counts a "typical" Gemini App text prompt. But what does "typical" mean?
There's no information on average prompt length or how more intensive tasks are factored in. The real energy drains—document analysis, media uploads, image generation, audio transcription, live video, advanced web searches, and code generation—don't factor in. By highlighting only simple, basic text queries, Google sidesteps the higher energy costs of complex workloads and leaves out the likelihood that real averages are much higher once edge cases are included.
Google's AI also powers products beyond the Gemini app. It runs behind the scenes in services like Google Search, multiplying energy use far past what the study measures.
