STRAIGHT ANSWER
Does every AI prompt use a bottle of water?
No. The claim traces to a 2023 modeled estimate that GPT-3 consumed about one 500 mL bottle across roughly 10 to 50 medium-length responses (the first April 2023 draft said 20 to 50 questions and answers). That is per dozens of responses, modeled, and not a current meter reading.
The context
The Li et al. preprint 'Making AI Less Thirsty' modeled on-site and electricity-related water for GPT-3 in 2023. Later measured figures from Google (0.26 mL per median prompt, May 2025) are roughly 40 to 200 times smaller, partly because of a narrower boundary and partly because hardware, serving, and cooling changed.
The published figures, side by side
Different boundaries, dates, and metrics. This is a reading aid, not a ranking. Do not add the rows.
| Figure | Value | Boundary | Period | Evidence | Source |
|---|---|---|---|---|---|
| Google, median Gemini Apps text prompt | 0.26 mL and 0.24 Wh | On-site cooling water only; excludes electricity-related water | May 2025 point-in-time median | Reported · primary | Google inference study |
| OpenAI CEO, average ChatGPT query | about 0.32 mL and 0.34 Wh | Boundary, sample, and method not published | June 2025 blog post | Reported · primary | Altman, The Gentle Singularity |
| Li et al., GPT-3 scenario | 10 to 50 mL per medium-length response | Direct plus electricity-related water; varies by location and time | 2023 preprint | Modeled | Making AI Less Thirsty |
What this answer does not say
- That the study was wrong. It was a modeled scenario with a stated boundary.
- That today's services use zero water per prompt.
- That the bottle figure applies to any model other than the one modeled.