Our AI Footprint

Water withdrawn vs. water consumed

Water figures are reported two different ways, and they're easy to mix up:

• Withdrawn — all the water taken in from a river, lake or aquifer. Most of it goes back to its source afterwards.
• Consumed — the netamount: water that doesn't go back, mostly because it evaporates. This is the real draw on local water supplies, and it's the measure we lead with.

The two figures at the top of this page are both consumedwater, so they measure the same kind of thing. They still aren't a perfect like-for-like pair — ours counts data-centre cooling only (see below) — so they're shown for scale.

Built for low water use

We designed our system to use AI sparingly and with a low water footprint. We only call AI where it genuinely helps you — tailoring your résumé, writing a cover letter, answering an application question. Finding and filtering jobs, tracking applications, and filling in the details you've already given us happen on our own servers.

For scale: a single question to ChatGPT or Gemini uses about 0.3 mL of cooling water, by the companies' own figures [1] [2]. One application does the work of a long chat session — reading the job, rewriting your résumé, writing a cover letter and answering screener questions.

What one application uses

We measured one complete journey — uploading a résumé, browsing matches, tailoring, and preparing applications — and it took about 76 AI requests in total.

How we turn that into water

Anthropic, who make the Claude models we use, doesn't publish a per-request water figure. So we borrow the closest published ones: Google reports 0.26 mL of water for a median Gemini prompt [1], and OpenAI has stated about 0.32 mL for an average ChatGPT query [2].

76 requests × 0.26–0.32 mL ≈ 20–24 mL — about a tablespoon and a half of cooling water for a complete application journey.

Treat these as a lower bound. Google's figure comes with a published methodology; OpenAI's is a single statement with none, and independent estimates for similar models run considerably higher — 0.6 to 17 mL per prompt [3].

What that number leaves out

It counts on-site cooling water only — the water that cools the servers during your requests. It does not count the water used to generate the electricity those servers run on.

Research that includes the water consumed generating the electricity estimates 10–50 mL per AI response [6], which would put a full journey at roughly one to four litres consumed.

Both of our AI numbers are legitimate; they answer different questions. We lead with the smaller one because it's closest to what AI providers publish, and we're telling you here exactly what it includes.

How we worked out the household figures

The average U.S. home uses 10,791 kWh of electricity a year [5] — about 30 kWh a day.

• Consumed: fossil-fuel and nuclear plants evaporate about 0.33–0.47 gallons per kWh [9] [7]. 30 kWh × 0.33–0.47 ≈ 10–14 gallons a day.
• Withdrawn: U.S. power generation takes in about 11.6 gallons per kWh [4]. 30 kWh × 11.6 ≈ 343 gallons a day — most of it returned to its source.
• The home itself: the average American family uses more than 300 gallons a day at home [8] — on top of the water behind its electricity.

These consumption rates cover fossil-fuel and nuclear plants only. Hydropower loses far more per kWh to reservoir evaporation [9], so the real figure for any one home depends on where its power comes from.

Why it's an estimate, not a measurement

We can measure how much AI we use. We can't measure the water: it depends on which data centre handles a request, its cooling design, the season, and the local power supply — none of which is visible to us. The per-request figures we borrow are from other providers, and real usage varies from person to person.

What we're doing about it

The biggest single use of AI in a journey is ranking your job matches. We're moving that onto our own servers, and we'll update this page with the measured difference once we have it — a real number, not a projection.

References

1. [1]Google — “Measuring the environmental impact of delivering AI at Google scale” (August 2025). Median Gemini Apps text prompt: 0.24 Wh, 0.26 mL of water. arxiv.org/abs/2508.15734
2. [2]Sam Altman, OpenAI — “The Gentle Singularity” (June 2025): an average ChatGPT query uses “about 0.000085 gallons of water” (≈ 0.32 mL). Self-reported; no methodology published. blog.samaltman.com/the-gentle-singularity
3. [3]PolitiFact — fact check noting OpenAI’s per-query water figure is the only source for that number, and that independent estimates for GPT-4o range from 0.6 to 17 mL per prompt (September 2026). politifact.com/factchecks/2026/sep/04/sam-altman/chatgpt-almonds-ai-california-water-useage/
4. [4]U.S. Energy Information Administration — “U.S. electric power sector continues water efficiency gains” (June 14, 2023). Water-withdrawal intensity: 11,595 gallons per MWh in 2021. www.eia.gov/todayinenergy/detail.php?id=56820
5. [5]U.S. Energy Information Administration — “How much electricity does an American home use?” Average U.S. residential customer: 10,791 kWh in 2022. www.eia.gov/tools/faqs/faq.php?id=97&t=3
6. [6]Li, Yang, Islam & Ren (UC Riverside) — “Making AI Less ‘Thirsty’: Uncovering and Addressing the Secret Water Footprint of AI Models” (2023; v5 2025). GPT-3 consumes a 500 mL bottle of water for roughly 10–50 medium-length responses. arxiv.org/abs/2304.03271
7. [7]Torcellini, Long & Judkoff, National Renewable Energy Laboratory — “Consumptive Water Use for U.S. Power Production” (2003). Thermoelectric plants evaporate 0.47 gal (1.8 L) of fresh water per kWh consumed at the point of end use. docs.nrel.gov/docs/fy04osti/33905.pdf
8. [8]U.S. Environmental Protection Agency, WaterSense — “How We Use Water”: the average American family uses more than 300 gallons of water per day at home. www.epa.gov/watersense/how-we-use-water
9. [9]Lee, Han, Elgowainy et al., Argonne National Laboratory — “Regional water consumption for hydro and thermal electricity generation in the United States,” Applied Energy (2018). Generation-weighted average for U.S. thermoelectricity: 1.25 L/kWh (≈ 0.33 gal/kWh). www.osti.gov/pages/biblio/1427501-regional-water-consumption-hydro-thermal-electricity-generation-united-states