See the water behind everyday life—and where AI fits
You can see the shower. You cannot see the coffee farm, the power plant behind Netflix, or the cooling behind an AI prompt.
Build an ordinary day, watch small routines compound, and compare them with disclosed AI prompt estimates without pretending every kind of water is the same.
No guilt score
No mixed-scope total
Every number sourced
Sample water receipt
At home
~146K prompt-sized volumes
37.9 L
Supply chain
~500K prompt-sized volumes
130 L
Power grid
mostly hot-water heating · ~7,404 prompt scenarios
7.7 L
Data center
20 disclosed text-prompt estimates
5.2 mL
Four water addresses
The receipt keeps each address visible, then adds only records that belong in the same accounting lane.
Build a receipt in under a minute01 · At home
Showers, sinks, leaks, washers, and appliances. Some water returns through wastewater systems.
02 · Supply chain
Agriculture and manufacturing footprints, often spread across distant places and different kinds of water.
03 · Power grid
Water consumed where electricity is generated. The result changes with the grid, device, and location.
04 · Data center
Company-reported or modeled facility water, attached to the exact workload, place, and period disclosed.
The prompt lens
The site uses three different lenses. The language changes with the evidence, so a dramatic number never masquerades as equal environmental impact.
Prompt-sized volume
Different water story
The same numeric water volume as approximately 146K disclosed 0.26 mL prompt-sized amounts.
~146K
Shared-grid model
Same electricity method
Modeled electricity-related water comparable to roughly this many prompt electricity scenarios under the same U.S.-average grid factor.
~963
Category proxy
Not TikTok-reported
A modeled marginal mobile-network scenario for TikTok, Reels, or Shorts-style viewing—not a platform water disclosure.
~3.6
The 0.26 mL and 0.24 Wh prompt baselines come from Google’s May 2025 company analysis. They are not independently verified or universal. How we know
A useful digital footnote
Always-on networks complicate per-user accounting. The label matters more than the dramatic number.
~79.7
Spreads the studied always-on digital chain across viewers. Useful for attribution—not for the next swipe.
~3.6
A marginal network-only scenario. It excludes always-on allocation, devices, and other platform infrastructure.
Small routines still compound. The receipt normalizes daily, weekly, and monthly habits, then lets you switch from one day to one year without mixing water addresses.
The national picture
These USGS categories share one consumptive-use series. The separate data-center estimate remains outside that scale because its year and boundary differ.
Comparable USGS consumption series
CONUS · average day · water years 2010–2020
Perspective is not permission
A small per-prompt amount and a meaningful local facility impact can both be true. Place, season, water source, and growth still decide what matters.
Is the facility in a water-stressed watershed?
Does it use potable, reclaimed, or another water source?
How much is withdrawn, consumed, and returned?
What happens during the hottest and driest months?
How much water is tied to the electricity supplying it?
Does the operator disclose facility-level and AI-specific demand?
Keep AI accountable. Keep the comparison honest.
The useful question is not whether AI uses water. It is how much, what kind, where, and compared with what.