# How do I write a green AI policy?

> What a short green AI policy should say: when to use AI, which defaults to set, what to ask suppliers, how to measure, and who owns it.

Source: https://greenai.guide/how-do-i/write-a-green-ai-policy/  
Updated: 2026-11-19

## In short

Keep it short and practical: say when AI is worth using, set efficient defaults (smallest adequate model, output limits, features off unless needed), require energy questions in procurement, and commit to measuring per task. Give it an owner, a review date and a few measurable commitments, and attach it to processes you already run.

## Steps

1. State the aim in one sentence covering both efficiency and using AI only where it adds value.
2. Set defaults for model size, output length and which AI features are switched on.
3. Add energy and location questions to procurement and contract renewals.
4. Commit to measuring AI use per useful task with a stated method.
5. Name an owner, set a review date and list three to five measurable commitments.

**A good green AI policy is one page that changes defaults, not a long statement of values.** People follow defaults
far more reliably than reminders, so the policy should decide what happens when nobody thinks about it.

## What should it cover?

| Section | What to say |
|---|---|
| Purpose | Use AI where it adds value, and use it efficiently. |
| When to use AI | Check first whether search, a template or existing software does the job. The W3C Web Sustainability Guidelines likewise ask teams to evaluate the need for AI before implementing it. |
| Defaults | Smallest model that meets the quality bar; short outputs; AI features off unless needed. UNESCO and UCL found shorter prompts and responses could cut energy by more than half. |
| Images and video | Generate what you need; refine rather than regenerate. |
| Buying | Ask for energy per request, location and reporting; see [AI procurement](/how-do-i/put-sustainability-into-ai-procurement/). |
| Measuring | Per useful task, with a written method; see [track AI energy per task](/how-do-i/track-ai-energy-per-task/). |
| Claims | Only what you can evidence; see [honest claims](/how-do-i/make-honest-claims-about-green-ai/). |
| Ownership | A named owner, a review date, three to five commitments. |

## Where should it live?

Attach it to processes you already have: procurement, data protection, sustainability reporting and staff training.
If you run an AI management system under ISO/IEC 42001, add environmental impact to it rather than starting a
separate document.

## Common mistakes

- Writing rules nobody can check. "Use AI responsibly" is not measurable; "default to the small model in our
  assistant" is.
- Banning AI outright. That tends to push use into unmanaged personal tools.
- No review date. Models and tools change every few months.

Our [green AI policy guide](/practice/green-ai-policy/) has a template and worked examples.

## Sources

- [ISO/IEC 42001:2023 Information technology – Artificial intelligence – Management system](https://www.iso.org/standard/81230.html), ISO (2023)
- [Web Sustainability Guidelines](https://www.w3.org/TR/web-sustainability-guidelines/), W3C (2025)
- [Smarter, smaller, stronger: resource-efficient generative AI and the future of digital transformation](https://www.unesco.org/en/articles/ai-large-language-models-new-report-shows-small-changes-can-reduce-energy-use-90), UNESCO and University College London (2025)
