{"slug": "i-built-an-ai-that-tells-you-how-much-water-it-just-drank", "title": "I Built an AI That Tells You How Much Water It Just Drank", "summary": "An engineer built JSTJO, an AI interface that displays a visible water counter that depletes with each query, to make the hidden environmental cost of AI tangible. The project highlights that a single AI query uses 10-25 milliliters of water when indirect usage is included, and that major companies like Google and Amazon have disclosed massive water footprints, with the UN launching an initiative to standardize such disclosures.", "body_md": "In June the UN asked every major AI company to publish its water footprint. Most still haven’t. So I built the version that does.\n\nAsk ChatGPT a question and something happens that nobody puts on the receipt.\n\nSomewhere, a server heats up. Somewhere else, a cooling tower evaporates water to carry that heat away. And somewhere further back — usually much further back — a power plant burns through more water to make the electricity that ran the whole thing.\n\nYou will never see any of it. There is no number on the screen. There is no line item. The interface is a text box and a blinking cursor, and the cost is somebody else’s problem in somebody else’s county.\n\nI got stuck on that. Not on whether AI is good or bad — on the fact that a product this widely used has a physical footprint that is essentially invisible to the person using it.\n\nSo I built JSTJO, an AI that shows you the cost while you use it. Every answer spends a drop from a visible counter. The send key is a keycap that reads WSTWTR. The counter empties as you talk. You cannot use it without being reminded what it takes.\n\nThis is what I learned putting it together.\n\nThe numbers are worse than the vibes\n\nThe debate about AI and water usually runs on vibes — one side says it’s destroying the planet, the other says a query uses less water than a hamburger. Both sides mostly argue without numbers. Here are the ones that made me actually build something.\n\nGoogle used 10.9 billion gallons of water in 2025. That’s the company’s own disclosure — up 34% in a single year, and roughly double what it used in 2021. The curve is not flattening. The curve is doing the opposite of flattening.\n\nAmazon disclosed its water footprint for the first time in June 2026: 2.5 billion gallons. It took until this year for one of the largest data-center operators on Earth to publish that number at all.\n\nMicrosoft says it will be water-positive five years early. Which is genuinely good — and also tells you the pressure is working. Companies do not accelerate sustainability targets out of boredom.\n\nAnd in June 2026, the UN launched the AI Environmental Transparency Initiative, formally asking AI companies to publish standardized water and energy disclosures. It’s voluntary. Most haven’t. But the direction is set: this information is becoming something users are entitled to, not something companies graciously reveal.\n\nThe honest per-query estimate, when you include the indirect water — the power generation, not just the data-center cooling — lands somewhere between 10 and 25 milliliters per answer. The often-quoted 0.3ml figure counts only the water the data center touches directly. The International Energy Agency estimates about two-thirds of AI’s real water footprint is indirect. Counting only the direct share is like measuring a road trip by the fuel you burned in the driveway.\n\nTen to twenty-five milliliters is not much water. A billion queries a day is not “not much water” anymore. Scale is the entire story.\n\nTransparency changes behavior — mine did\n\nHere’s the part I didn’t expect.\n\nBecome a Medium member\n\nI built the drop counter as an honesty feature — a way to keep the product from lying by omission. But once it was in front of me, it changed how I used my own product.\n\nI stopped sending throwaway messages. I stopped asking the model to redo things I could fix myself in ten seconds. Not because anyone told me to. Because the counter went down when I did, and I could see it.\n\nThere’s a name for this in behavioral economics — feedback effects. Smart electricity meters reduce household consumption not by rationing anything but by making the invisible visible. Nobody’s electricity got more expensive. They just finally saw it.\n\nSoftware interfaces make the same choice constantly, silently, in the other direction. Infinite scroll hides time. One-click checkout hides money. A chat box hides infrastructure. Every interface decides what you get to see, and most decide you should see nothing.\n\nI wanted to build one that decided the opposite.\n\nWhat JSTJO actually does\n\nJSTJO is an AI assistant — ask it anything, it answers, searches the web, reads your files. The difference is the meter.\n\nYou start with 50 free drops a month. Every answer spends one. When you run low, you can wait for the refill, play the built-in water game to earn more, or pay. The paid tiers remove the ceiling — but the counter never goes away, because the counter is the point.\n\nAnd 5% of everything the product earns funds clean water projects. Not as an offset gimmick — as a line item you can check. The site header shows the running total raised. The ledger of what’s actually been sent is public. If those two numbers ever diverge for long, you’d see it, which is precisely the kind of accountability I’m arguing every AI product owes its users.\n\nIs my counter perfectly accurate? No — and I don’t claim it is. It’s an honest estimate built from published research, in an industry where the alternative on offer is no number at all. I’ll take a good-faith estimate over a blank screen every time.\n\nThe part where I tell you what I want\n\nI want the big labs to put the number on the screen.\n\nNot in a sustainability PDF published once a year. Not in a footnote with a methodology nobody can audit. On the screen, next to the answer, where the person who caused the water use can see it.\n\nThe UN has now asked for basically this at the corporate level. Users can ask for it at the product level. It is not technically hard — I am one person and I built a working version. It is a choice about what the interface admits.\n\nEvery AI answer drinks water. The only question is whether you get to know.\n\nYou can try it at jstjo.ai — the counter’s on screen, emptying as you use it, so you can see the water going.\n\nJSTJO is free to use at [jstjo.ai](https://dev.tourl) — 50 answers a month, no card. Every one shows you what it cost.\n\n*Sources: Google 2026 Environmental Report; Amazon sustainability report, June 2026; IEA data centre water estimates; Lawrence Berkeley National Laboratory, December 2024; UC Riverside research on AI water consumption; UN University AI Environmental Transparency Initiative, June 2026.*", "url": "https://wpnews.pro/news/i-built-an-ai-that-tells-you-how-much-water-it-just-drank", "canonical_source": "https://dev.to/joannamarrie/i-built-an-ai-that-tells-you-how-much-water-it-just-drank-48b1", "published_at": "2026-08-10 20:22:11+00:00", "updated_at": "2026-08-10 20:47:14.305064+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-ethics", "ai-policy", "ai-products"], "entities": ["Google", "Amazon", "Microsoft", "United Nations", "International Energy Agency", "JSTJO"], "alternates": {"html": "https://wpnews.pro/news/i-built-an-ai-that-tells-you-how-much-water-it-just-drank", "markdown": "https://wpnews.pro/news/i-built-an-ai-that-tells-you-how-much-water-it-just-drank.md", "text": "https://wpnews.pro/news/i-built-an-ai-that-tells-you-how-much-water-it-just-drank.txt", "jsonld": "https://wpnews.pro/news/i-built-an-ai-that-tells-you-how-much-water-it-just-drank.jsonld"}}