DRAIN Extension Track your AI water footprint in real time.
Install Free
drAIn
We can't prompt a new planet
scroll

IS YOUR AI
THIRSTY AF?

Move the sliders to check your own AI usage. It's calculated based on the midpoint between what tech companies report and what independent researchers have found.

💬 Text Prompts
ChatGPT, Claude, Gemini, Perplexity — any text conversation
0
0ml water
🖼️ Images Generated
DALL-E, Midjourney, ImageFX, Firefly, Canva AI
0
0ml water
🎬 Videos Generated
Sora, Runway, Pika, Veo — any AI video creation
0
0ml water
🎙️ Voice Generations
ElevenLabs, TTS — any AI audio synthesis
0
0ml water
Today
0ml
Start moving the sliders
Per Month
0ml
×30 days
Per Year
0ml
×365 days
Daily usage in 500ml bottles
Move a slider to see your impact
🚿
0
seconds of
a shower
🌱
0
days watering
a houseplant
☕
0
cups of
coffee brewed

THE COST OF
A PROMPT

Companies only count water used directly on-site for cooling. Independent researchers also include the water consumed by power stations generating the electricity that runs the servers. The number we show is the midpoint between both.

Action Company says We show Research says
💬
Text Prompt
ChatGPT, Claude, Gemini, Perplexity
~3ml
21ml
~39ml
🖼️
Image Generation
DALL-E, Midjourney, ImageFX, Firefly
~30ml
165ml
~300ml
🎬
Video Generation
Sora, Runway, Pika, Veo
~200ml
700ml
~1,200ml
🎙️
Voice Generation
ElevenLabs, TTS synthesis
~4ml
25ml
~45ml
10 – 30×
The gap between what companies self-report and what researchers find when accounting for indirect water from power generation.
700ml
One AI video generation uses roughly the same water as a large bottle plus a cup of tea.
2023
The year the first peer-reviewed research on AI water consumption was published. This was years after mass AI deployment had already begun.
↑
AI water demand is growing faster than efficiency gains. As models scale, the total footprint keeps rising despite hardware improvements.
0
Mandatory disclosure requirements for AI water consumption exist in most countries. These figures are voluntary and inconsistently reported.
// a note from us
WE AREN'T
HAITERS

Despite what it might look like, we're not trying to stop people from using AI. In fact, we used it to help us on this project, and it's really our own daily use of AI and our questions around its sustainability that inspired us to act. We hope this website and the DRAIN web extension will help everyone become a little more aware of the impact of their prompts, because if the tech companies that own the servers won't take responsibility, then it's down to us.

DRAIN Extension Track your AI water footprint in real time.
Install Free
// sources & methodology
Li, P. et al. (2023). "Making AI Less Thirsty: Uncovering and Addressing the Secret Water Footprint of AI Models." UC Riverside / UT Arlington. arXiv:2304.03271
Google Environmental Report (2025). Data centre water consumption disclosures. ~0.26ml per Gemini query (direct water only).
OpenAI (2025). Sam Altman stated ChatGPT uses approximately 0.32ml of water per prompt (direct water only, company figure).
IEA (2024). "Electricity 2024: Data Centres and the Energy World." International Energy Agency. Water intensity of power generation methodology.
Figures shown are the midpoint between company self-reported direct water and independent full-accounting estimates. Text: 3–39ml → 21ml. Image: 30–300ml → 165ml. Video: 200–1,200ml → 700ml. Voice: 4–45ml → 25ml.