I have built a new calculator to measure the physical footprint of everyday artificial intelligence. The tool takes your typical generative AI tasks and calculates the resulting power consumption and carbon emissions, alongside the water required for data centre cooling. To make these abstract metrics easier to understand, it translates the data into physical equivalencies like miles driven in a petrol car or beef burgers consumed.
Why measure this?
The goal is not to shame people for using these services. AI is increasingly a mandatory part of modern professional work. Instead, I built this to illustrate how basic user nudges can have a serious environmental impact. We can significantly reduce our footprint just by changing how we interact with the interface.
How to reduce your prompt footprint
There are practical ways to minimise the server compute required for your daily tasks.
- Choose lightweight models. Not every task requires the most advanced reasoning engine available. Default to faster, smaller models for standard text generation.
- Lower the thinking mode. Switch to low or medium thinking modes for everyday drafting and summarisation. Reserve high-compute reasoning strictly for complex logic.
- Consolidate your requests. Writing a single comprehensive prompt requires far less processing power than a long, iterative conversation made up of several smaller corrections.
- Keep inputs succinct. Remove unnecessary background information. Every extra paragraph you paste into the context window directly increases the energy required to generate a response.
Seeing the physical cost of a prompt makes it much easier to decide when to deploy heavy compute, and when a smaller, faster model will do the job perfectly well.