Clear need for updated guidance

• With generative AI rapidly expanding across digital media, participants agreed on the urgency of developing more robust, transparent approaches to understanding energy use and carbon emissions.

Focused scope agreed

• The group aligned on prioritising energy consumption and related emissions in the near term, while recognising water use and other impacts as important open questions for future work.

Framework before figures

• Strong consensus emerged that establishing clear boundaries, functional units and assumptions is more important than producing headline numbers, particularly given limited data and fast‑moving technology.

Training vs. inference complexity

• Discussion highlighted challenges in separating model training impacts from ongoing inference, with agreement that limitations should be clearly documented where precise allocation is not possible.

Importance of counterfactuals

• Participants emphasised that AI impacts must be assessed relative to what AI replaces, noting that credible counterfactuals are essential for meaningful interpretation of results.

Media‑specific use cases prioritised

• Interest centred on use cases within the digital media value chain where AI may reduce, or unintentionally increase, overall emissions, rather than generic enterprise AI applications.

Transparency and data gaps acknowledged

• Limited visibility into closed‑source models and supplier infrastructure was identified as a key barrier, reinforcing the need for flexible methodologies that can work with partial data.

Alignment with responsible AI efforts

• The group discussed how environmental impacts could sit alongside existing responsible AI initiatives, helping embed sustainability as a core consideration rather than an afterthought.

Collaborative approach confirmed

• Participants expressed willingness to contribute anonymised data, use cases and expertise, and supported exploring external partnerships to strengthen credibility and relevance.

The session reinforced a shared ambition to develop practical, evidence‑based guidance that helps organisations understand and manage the real environmental impacts of AI.