How Much Water Did That AI Prompt Actually Use?

by Claire L. Brady, EdD

I get some version of this question on nearly every campus I visit:

How much water am I actually using when I use AI?

Higher education professionals are paying attention to AI's environmental impact, and water is often their biggest concern. They have seen the headlines. They have heard that a single ChatGPT query consumes a bottle of water. And many are genuinely wrestling with whether their individual AI use is contributing to a much larger environmental problem.

The frustrating answer is: there isn't one reliable number.

A recent Knowable Magazine article, “How much of a problem is AI’s water use?” does an excellent job unpacking why.

One widely circulated estimate suggested that drafting a short email using an earlier version of GPT-4 could consume roughly 500 milliliters of water. But the researcher behind that estimate now says it is outdated as AI models and infrastructure have become more efficient. In 2025, Google estimated that a median Gemini query required about five drops of water for processing.

That's quite a range.

The difference isn't simply because somebody got the math wrong. The environmental cost of an AI interaction depends on much more than the prompt you type. It depends on the model, the data center processing it, how that facility is cooled, the local climate, the source of its electricity, and even where the data center is located.

And location matters enormously.

Data centers consumed an estimated 66 billion liters of water for cooling in the United States in 2023, but that represented less than one percent of total U.S. water consumption. National numbers, however, can obscure local consequences. A water-intensive data center in a water-rich region is a very different proposition from one operating in drought-strapped Arizona or New Mexico.

That distinction matters when we talk about responsible AI use.

I don't think the answer is for every faculty member, staff member, or student to feel guilty every time they open an AI tool. Nor should we dismiss the environmental impact because an individual prompt may consume relatively little.

Individual use may be small. Institutional scale is not.

A campus with thousands of students and employees increasingly embedding AI into teaching, research, advising, operations, and everyday work should absolutely be asking questions about environmental impact. Institutions can also use their purchasing power to ask vendors harder questions about energy, water, infrastructure, transparency, and sustainability commitments.

For individual users, this is also where I come back to “AI with Intention: The Leadership Guide for Higher Education”. I wrote the book because responsible AI use isn't just about learning what these tools can do. It's about making thoughtful choices about when, why, and whether to use them at all. Environmental impact belongs in that conversation.

Using AI with intention doesn't mean counting drops of water every time you write a prompt or feeling guilty every time you open ChatGPT or Gemini. It means resisting mindless AI use. Don't generate 30 versions when three will do. Don't use AI simply because it's available. And don't assume that because something is easy for us to generate, it comes without cost.

AI with intention means asking not only “Can this tool do this?” but “Is this a good use of the tool—and is the value worth the resources it requires?”

This image was created using ChatGPT