Sustainable Prompting with Copilot: What Sustainable AI Usage Really Looks Like

AI usage is increasing rapidly, and with it the consumption of energy and resources. Companies, media and policymakers worldwide are discussing how we can leverage the benefits of AI without losing sight of environmental and climate impacts. Especially now, with AI models like Copilot being used daily by millions of people and organisations, sustainable usage matters more than ever.

AI consumes electricity, water and generates CO₂ – and not in insignificant amounts. A recent study by Jegham et al. (2025) quantified the environmental footprint of AI queries for the first time. The results:

• A single Copilot request using GPT-4o consumes around 0.42 Wh of electricity – roughly 40% more than a Google search.

• Scaled to 700 million requests per day, this corresponds to the electricity usage of 35,000 US households, the water consumption of 1.2 million people, and CO₂ emissions that would require a forest the size of Chicago to compensate.

Even more interesting: not only the model itself but also the hardware and data center infrastructure determine the environmental impact. Modern hardware and renewable energy make Copilot significantly more sustainable.

Many people focus on the training of large models when discussing AI sustainability. But the study shows: up to 90% of energy consumption occurs during usage — in other words, with every single query (inference). The more we use AI, the more important sustainable prompting becomes.

Sustainable prompting means using Copilot as efficiently as possible — achieving the best possible result with minimal computational effort. This conserves resources and often leads to better answers.

Three tips for sustainable prompts:

1. Clear, targeted prompts: The more precise the question, the less computation is needed.

2. Provide context and sources: Copilot needs to “think” less and can deliver the right result faster.

3. Avoid unnecessary repetitions: Every prompt counts — better to ask once clearly than refine multiple times.

Copilot’s resource consumption is not only a sustainability issue but also an economic one. These energy and infrastructure costs are borne primarily by cloud providers, but they indirectly affect licensing and operational expenses. For companies and end users, this means: the more efficient the prompt, the fewer resources are consumed — and the more stable the pricing remains over time. Sustainable prompting is therefore not only beneficial for the environment and infrastructure but also economically sensible in the long run.

The study shows that the infrastructure behind Copilot is becoming more efficient, but overall consumption continues to rise with usage. Microsoft is investing heavily in green data centers and renewable energy — yet we, as users, can also make a difference with sustainable prompting.