Google has released a technical paper detailing its methodology for measuring the energy, emissions, and water impact of Gemini AI prompts.

According to Google, the median Gemini Apps text prompt uses 0.24 watt-hours (Wh) of energy, emits 0.03 grams of carbon dioxide equivalent (gCO2e), and consumes 0.26 milliliters (or about five drops) of water. The company claimed that this is much lower than many public estimates, with the per-prompt energy impact equivalent to watching TV for less than nine seconds.

As a result, per 1 million queries, Gemini would consume 240,000Wh (240kWh), emit 30,000 gCO2e (30 kgCO2e), and consume 260,000 milliliters (260 liters) of water.

According to Google, it is making strides in reducing the energy and carbon footprint of the median text prompt. The company said that over the last 12 months, it has seen energy and total carbon footprint drop by 33x and 44x, respectively. The results are built on its latest reductions in data center emissions and work to decarbonize its energy consumption.

“We hope this study contributes to ongoing efforts to develop efficient AI at this critical time for energy, sustainability, and scientific discovery — to benefit everyone,” said Ben Gomes, chief technologist, learning and sustainability at Google.

The methodology is based on relevant energy sources in its inference-serving stack. These include active AI accelerators (TPUs), host CPUs and DRAM, idle machine provisioning, and data center overhead. Excluded from the methodology are external network energy, end-user device energy, model training, and data storage.

However, according to some experts, the data is misleading. “They’re just hiding the critical information,” Shaolei Ren, an associate professor of electrical and computer engineering at the University of California, Riverside, told The Verge. “This really spreads the wrong message to the world.”

A significant issue that was omitted, according to experts, was indirect water use, with the study only including water that data centers use in cooling, and not the water needed for large generation projects that power the data centers themselves.

In addition, experts pointed out that Google only included a “market-based” measure of carbon emissions and not “location-based” which considers the impact that a data center has wherever it operates by taking into account the current mix of clean and dirty energy of the local power grid.

Google has yet to submit the paper for peer review, however, it has indicated that it is open to doing so in the future.

Google competitor OpenAI has also been in the news lately regarding its energy use. A recent report revealed that its latest model, Chat GPT-5 can require, on average, more than 18Wh of electricity, with some responses reaching 40Wh.

Earlier this year, OpenAI CEO Sam Altman provided a glimpse into the individual power and water usage of ChatGPT’s previous models.

"The average query uses about 0.34 watt-hours, about what an oven would use in a little over one second, or a high-efficiency lightbulb would use in a couple of minutes," Altman said in a blog post. "It also uses about 0.000085 gallons of water; roughly one-fifteenth of a teaspoon."

0.000085 gallons is around 0.386418ml, slightly higher than Google's claimed Gemini water use.