In short
Ask for energy, carbon and water per typical request for the model you use, plus the method, the system boundary and where it runs. Ask for both location-based and market-based carbon figures. Many providers cannot yet answer in full, but a specific question gets a better reply than a general one, and repeated asking builds demand.
Steps
- Name the exact product and model you use, not just the company.
- Ask for energy (Wh), carbon (g CO₂e) and water (mL) per typical request.
- Ask what the figure includes, such as cooling, idle capacity and hardware manufacture.
- Ask for location-based as well as market-based carbon figures.
- Ask whether the figure was measured or estimated, and whether anyone independent checked it.
- Keep the reply and compare it with other providers' published figures.
Ask a precise question about a specific model, and ask how the answer was worked out. A figure without a method or boundary cannot be compared with anything.
Why does it matter?
Providers publish very different numbers because they count different things. Google reports a median Gemini text prompt at 0.24 Wh when it counts accelerators, host machines, idle capacity and data-centre overhead. Mistral AI's life-cycle assessment also counts hardware manufacture and water. Neither is wrong, but you need to know which one you are looking at.
What should I ask?
Use the steps above as a short email or support ticket. The Green Web Foundation suggests asking vendors for both location-based figures (the actual grid where the work ran) and market-based figures (after renewable energy contracts), because the two can tell quite different stories.
If you are a business buyer, put the questions in your contract. In the EU, the AI Act already requires providers of general-purpose AI models to document the model's known or estimated energy consumption, so some of this information should exist.
Common mistakes
- Accepting "we are carbon neutral" as an answer. Offsets and certificates are not energy figures.
- Comparing figures with different boundaries. Check what each includes first.
- Giving up when the reply is vague. Ask again at renewal, and choose providers who answer clearly.
Organisations should read sustainable AI procurement for a full question set.
Sources
- Measuring the environmental impact of delivering AI at Google Scale, Elsworth et al., Google (2025)
- Our contribution to a global environmental standard for AI, Mistral AI (2025)
- Thinking about using AI?, Green Web Foundation (2025)
- Regulation (EU) 2024/1689 (AI Act), Annex XI, European Union (2024)
Related
How do I know if an AI tool is green?
What to look for when judging an AI tool's environmental claims: published energy figures, methods, model size options, hosting and independent ratings.
How do I check a claim about AI's energy use?
Six questions to test any claim about AI energy, water or carbon: per what, who says so, what's included, how old, compared with what, total or per use.
Sustainable AI procurement
What to ask AI suppliers about energy, carbon and water, which evidence to request, and contract clauses that make sustainability commitments enforceable.
Measurement is the bottleneck for green AI
Why AI energy figures are hard to trust or compare: self-reported numbers, estimates not meters, different boundaries, and what good looks like.