1What the number means
NoSlop shows an estimated water equivalent for recorded activity. It is an illustration, not a measurement of water saved at a data center. A blocked app opening does not prove that an AI request would have happened. Hiding an answer does not undo the resources used to create it.
Actual changes in water consumption may be lower, higher, or zero. These figures are not environmental offsets or verified conservation results.
2Which events count
Method 2 assigns a water equivalent to recorded blocked navigation and prevented requests. Screen Time shields count as blocked navigation. These are indicators of activity, not proof of avoided AI computation.
Hidden content, provider changes, modeled activity, and unclassified events add zero to new water estimates. Their activity counts remain. Safari AI Overview removals and AI Mode redirects occur on loaded pages, so both add zero.
The standard shield path suppresses callbacks less than 1.5 seconds apart. A shield button does not count again after its presentation was recorded. Safari messages use an instance identifier to suppress repeats; older messages use the time rule. Protected app sessions also filter repeated reports before recording them.
3The assumption and formula
The default is 30 mL per eligible event. This is a product assumption retained for continuity, not a measured average for AI requests. No source reviewed establishes 30 mL as a universal rate. A stored setting can change the rate; negative rates contribute zero.
For each new event: amount in mL = the nonnegative configured rate for an eligible event, or zero otherwise. The total is the sum of the amounts stored with the included events. Changing the setting does not recalculate previous entries.
Example: 900 eligible events at 30 mL each produce 27,000 mL, or 27 L, about 7.13 US gallons. The 900 events are an example, not a prediction of monthly use or savings. A further 900 hidden-content events add zero.
4Units, dates, and comparisons
Liters = mL / 1,000. US gallons = mL / 3,785.411784. US fluid ounces = mL / 29.5735295625. Integer mL values are summed before display rounding.
US displays use fluid ounces below one gallon, then gallons. Gallons below 10 use one decimal place; larger totals use whole gallons. Other regions use mL below one liter, then liters, with up to one decimal place. Chart points use one common unit. Rounding can make displayed subtotals differ from a displayed total.
Daily and weekly totals use the app's calendar and included date range. Lifetime totals include all stored entries. Glasses, bottles, and other milestones are illustrative volume targets, not water supplied or conserved. The Pool milestone is a 100-liter progress target, not an Olympic pool.
5Earlier estimates
Method 2 was introduced in the September 29, 2026 source update. Each new standard activity entry stores its method version and amount. Older entries without a version keep their original amounts. They may include hidden or modeled activity. A combined total can therefore contain different methods.
Older installed app versions can continue to record under their earlier rules until updated. The update does not silently revise history. Older annual projections and the website's 10 mL per message and 500 gallons per person claims have been removed from the current product copy.
6Research and limits
Water withdrawal is water taken from a source. Consumption is the portion not returned for reuse in the same context. Data-center cooling, electricity generation, hardware production, and model training have different boundaries. A figure for one cannot be relabeled as the total for all of them.
Li and colleagues (2023, revised March 2025) study direct and electricity-related water footprints, including location and time differences. Their 2027 global AI withdrawal projection is not a measured annual consumption total for today's data centers.
Elsworth and colleagues (August 2025) report 0.26 mL for the median Gemini Apps text prompt in May 2025. Their water calculation covers data-center cooling and uses fleet-average water efficiency. It does not establish a universal rate or the water saved by a NoSlop event.
NoSlop does not identify the serving model, prompt length, data-center location, cooling system, power source, or marginal resource change for each block. It does not compute a full life-cycle footprint. Neither study validates the app's fixed 30 mL illustration.
Li et al.: Making AI Less Thirsty (version 5)
Elsworth et al.: Measuring the environmental impact of delivering AI at Google Scale