How we stay unbiased

Every figure on this site is computed from one public coefficients file, and every coefficient names its source: peer-reviewed studies, official company disclosures, government data. The Methodology page renders directly from that same file, so what we publish and what we compute cannot drift apart.

Where studies disagree, we publish the disagreement as a range instead of picking the most quotable end of it.

Whether a kilowatt-hour of AI “uses” 0.3 or 30 liters of water depends on what you count: only the cooling system, the power plant generating the electricity, or the full lifecycle including hardware manufacturing. We label every figure Low, Mid or High and state exactly what each boundary counts. A number published without its boundary is an opinion wearing a decimal point.

The site is self-funded and carries no ads; reader donations go toward keeping the data current. We take no money from AI companies, cloud providers, water utilities, or advocacy groups; there are no sponsored posts; and no donation has ever changed a coefficient. If any of that changes, it will be disclosed here first.

No individual names are attached to this project, which means there is no personal brand to defend and no incentive to be interestingly wrong. The methodology is the author. Argue with it; it can take it.

The coefficients file is versioned and dated. When a number changes (a new disclosure, a better study, our own mistake), the version changes with it, and the source note says why.

Found something wrong? Send it to hello@aiwateruse.org with a source. Sourced corrections are the most valuable thing a reader can give this site, and they get fixed, not negotiated.

Every coefficient, assumption and citation is public on the methodology page, and the same numbers are served by the free API and embeds. A journalist or researcher can re-derive any figure on this site without asking our permission. That is the point.