JOOCGAME / little arcadeHow much water could your prompts represent? Explore a public ChatGPT claim, then see why other AI benchmarks tell different stories.
01 / YOUR INPUT
Enter the total for the period you select. A month is exactly 30 days; changing the period keeps your entered count.
Average query claim: 0.000085 US gallons. Model, method and water-system boundary were not disclosed.
≈ 0.32 mL / prompt · Read source
Different models, years and water boundaries. These are separate examples, not a range for ChatGPT or a ranking of efficiency.
0.26 mL for the median text prompt. Direct data-center cooling consumption only; excludes electricity-generation water. This is not ChatGPT.
≈ 0.26 mL / prompt · Read source
16.904 mL per modeled query: 2.200 mL direct + 14.704 mL electricity-related consumption. Historical scenario, not present-day ChatGPT; excludes training and hardware.
≈ 17 mL / prompt · Read source
No account or prompt content needed. Calculations stay in this page; shared URLs and cards contain only your count, period and source.
02 / SCENARIO RECEIPT
An estimate, with the small print20 prompts per day
ChatGPT · Altman public claim (2025)
Average query claim: 0.000085 US gallons. Model, method and water-system boundary were not disclosed.
250 mL each. A volume illustration, not an environmental-impact equivalence.
Scenario estimate only. This is not measured personal water use, a water bill, or a complete lifecycle footprint.
Open this result ↗Source snapshot 2026-10-03.1 · checked 3 Oct 2026
Water depends on the model, response length, hardware, location, weather and electricity supply. Some sources count only cooling; others also count water consumed in generating electricity. An average or median cannot measure your individual prompt.
A farm water-footprint number and a data-center cooling number describe different systems. Geography, irrigation, supply chains and whether a study counts consumption or withdrawal all matter. We do not turn them into an almond-equivalent score.
Formula: prompts × source mL per prompt ÷ 1,000 = liters. The glass illustration divides milliliters by 250. Display values are rounded to about two significant figures. No training-water allocation is added.
Sam Altman, The Gentle Singularity, 10 June 2025. 0.000085 US gallons × 3,785.411784 = 0.32176000164 mL per query.
Google, Measuring the Environmental Impact of Delivering AI at Google Scale (2025). Median Gemini Apps text prompt, May 2025: 0.26 mL, direct cooling consumption.
Li et al., Making AI Less “Thirsty” (2023 paper, v5 dated 26 March 2025), §3.3.2 and Table 1. Modeled US average: 2.200 + 14.704 = 16.904 mL/query. Assumes 4 Wh of server energy for a request with up to about 800 input words and 150–300 output words.
Sources are a dated snapshot, not live telemetry. None establishes the full footprint of a specific prompt today.
The production site play.joocgame.com uses Vercel Web Analytics for page visits and Google Analytics 4 (GA4) for anonymous Arcade gameplay events. Vercel processes page paths, referrers, visit times, devices/browsers and approximate regions. GA4 receives only game_start, level_complete and replay events with the game slug and level (and optional score on level_complete). We send no personal identifiers, wallet data or account information. Vercel page views remove query strings and fragments; preview and non-production hosts do not load these scripts. Calculator inputs and in-game progress are not sent as analytics payloads. Read how Vercel processes data
