Responsible LLM carbon claims in marketing (avoid greenwashing)

Customers increasingly ask for “greener AI.” Regulators and NGOs increasingly scrutinize green claims. If you publish CO₂ figures for LLM features, treat them like financial-adjacent statements: clear boundary, honest uncertainty, and no implied precision you do not have.

Say what was measured vs modeled

Distinguish metered infrastructure data from modeledfootprints using coefficients. If your product shows grams CO₂e per request, add a short line that the number comes from a methodology (with version), not from direct measurement of your user's datacenter stack unless that is true.

Use confidence language consistently

Labels like measured, benchmarked, and estimated help users interpret spread between models. Avoid ranking competitors on a single headline number unless the underlying assumptions are aligned — that is how marketing slides become litigation or reputational risk.

What to avoid

• Claiming “carbon-neutral AI” without a defined scope and credible offsets or residual transparency.
• Comparing your footprint to a generic web search or human baseline without citing sources.
• Hiding test traffic in customer-facing “sustainability” metrics.