How to read water numbers
“Water use” can mean water taken from a source, delivered to a user, evaporated, returned, or counted across a supply chain. A number without that label is incomplete.
Follow the flow
01
River, lake, reservoir, aquifer, rainfall, reclaimed water, or another supply.
02
Water removed from groundwater or diverted from surface water for a purpose.
03
Water reaches a facility or household, or is applied to a field. This is not automatically consumption.
04
Cooling, irrigation, washing, manufacturing, drinking, or another end use.
Return flow
Water released back to groundwater or surface water and available for further use. Returned water can still differ in temperature or quality.
Consumption
The part evaporated, transpired, incorporated into a product or crop, or otherwise unavailable for immediate local use.
Delivered water and applied irrigation water describe movement to a user or field. Withdrawal describes movement from a source. Consumption describes what is not promptly returned. They answer different questions.
Four water addresses
The receipt keeps these flows separate. It can subtotal compatible lines within one address, but it never turns them into a single personal water score.
01
Delivered where you live
Water delivered for showers, faucets, toilets, appliances, leaks, and outdoor use. Delivery is visible, but it is not automatically the same as consumption.
02
At electricity generators
Some electricity generation withdraws and consumes water. A model may assign part of that footprint to streaming, appliances, or computing from energy use and a stated regional grid factor.
03
Behind food and products
Water associated with growing, processing, and making coffee, food, clothing, devices, and other products. This lifecycle footprint is not water delivered to your home that day.
04
At computing facilities
Operational water used for cooling or humidity control. A report must still say whether the figure is withdrawal or consumption and whether electricity-related water is included.
For food and products
Coffee, beef, and cotton figures often use water-footprint accounting. These components should stay visible instead of being collapsed into one mysterious total.
The comparison rule
Comparable water can be subtotaled. Different water stories can add perspective, but they never become one grand total.
Withdrawal stays beside withdrawal, consumption beside consumption, and lifecycle water in its own lane. A prompt-sized volume is an analogy about numeric scale—not a claim that coffee, showers, and computing have identical environmental impacts.
Three comparison lenses
The receipt chooses the most defensible lens for each line. It does not let a dramatic number outrun its boundary.
01
Streaming, appliances, and other electricity scenarios
Use electricity as the common unit, keep each system boundary visible, and apply the same regional electricity-water factor to the activity and AI prompt baseline.
“The modeled electricity-related water is the same numeric volume as about X prompt-serving electricity scenarios under the same grid-water model.”
02
Home water, food, and product footprints
Divide one published water volume by one disclosed per-prompt water volume while keeping both water addresses visible.
“This volume contains about X prompt-sized amounts. The water source, place, and accounting scope are different.”
03
Company totals, national sectors, and weak platform-specific evidence
Show the information without a prompt count when the boundaries cannot support a defensible translation.
No equivalence language and no shared quantitative axis.
Shared digital infrastructure
Networks, data centers, and devices use energy while serving many people. Assigning that energy to one viewing session is a modeling choice, not a directly metered fact.
Allocated share
An attributional model distributes always-on infrastructure across views or minutes. It is useful for footprint accounting, but it does not describe the immediate change caused by one more video.
Marginal use
A marginal model estimates the extra network or device energy created by another view. It can be much smaller because much of the system was already running.
These values are not high and low endpoints of one estimate. The receipt labels the selected model. A TikTok, Reels, or Shorts-style example remains a category proxy unless that platform publishes compatible evidence.
Evidence labels
A precise-looking result can still depend on assumptions. The evidence label tells you how the number reached the page.
A U.S.-average grid factor is not a local watershed measurement. Rainfall, water stress, power generation, cooling technology, and reuse differ by place.
Models, hardware, grids, and company disclosures change. Every receipt line keeps its source year and caveat visible, and the Sources page publishes the receipt catalog’s review date. A publication year is context—not a promise that a number is universal or permanently current.
Before comparing
01
On-site, electricity-related, embedded—or a stated combination?
02
Withdrawal, consumption, delivered, applied, or water footprint?
03
Per prompt, per kWh, per facility, per year, or per product?
04
Which watershed, electric grid, weather period, and data year?
05
Measured, company-reported, modeled, or derived from another estimate?
06
Is a range shown, and are omitted parts named?
Primary references