BRIGHT CHECK / WATER & AI
One prompt. One bottle of water?
What does the 500 mL estimate actually cover? Compare the study’s modeled scenario with a separate email estimate, then take the source-backed receipt with you.
BEFORE THE RECEIPT
What did the original paper count?
Make a guess to jump to the evidence. Every choice opens the same answer. Nothing is scored, counted or saved.
THE SOURCE RECEIPT
The bottle covered more than one response.
Misleading as stated. The original paper estimated 500 milliliters for roughly 10–50 medium-length GPT-3-style responses, depending on location. It assumed 0.004 kilowatt-hours per request and combined onsite cooling with electricity-generation water. It did not meter today’s ChatGPT, a named facility, or every prompt.
- Denominator
- Roughly 10–50 medium responses for 500 mL, depending on place and time.
- Model & basis
- GPT-3 operational water consumption, estimated from assumptions; not a meter reading of today’s ChatGPT.
- Boundary
- Onsite cooling plus water associated with electricity generation.
The original paper’s section 3.3 and Table 1 vary by location. The journal edition presents this work. Bright’s full water guide explains withdrawal, consumption, seasons and cooling systems.
A FAIR COMPLICATION
The email example is a different calculation.
The Washington Post and UC Riverside modeled about 519 mL for GPT-4 to write a 100-word email in 2024. That is a different model and task, not a correction of the GPT-3 paper or a universal current average.
Read the Washington Post/UC Riverside analysis ↗ . The two examples cannot be collapsed into a current per-prompt average.
Local water source, season, consumption versus withdrawal, cooling design and electricity mix still matter. This check does not show that AI water use is harmless.
TAKE THE RECEIPT
Share the denominator with the claim.
The copied text keeps the GPT-3 model, 10–50 response range, modeled basis, cooling and electricity boundary, separate email example and plain canonical link together.
Permanent link: https://brightaifuture.com/bright-check/water-per-prompt
WHAT WOULD CHANGE THIS?
Follow the evidence, not a universal number.
Model-, hardware-, token-, site-, hour- and cooling-specific metered energy and water data, with the electricity source and accounting boundary disclosed.
Revisit the full water guide for the wider evidence and source history. You can follow Bright’s broader water question on this device; that is not a claim-specific alert or email subscription.
Following and a reading bookmark stay on this device.
Prefer email? Bright Weekly is one general edition, not updates filtered to this check.