The AI Footprint All tools

What does your AI use add up to?

Pick the AI you use and roughly how much. See a year of it in electricity, carbon, and things you already have a feel for.

A back-and-forth chat counts each message you send.
How many are long, complex jobs? A quick question gets a short answer. Writing a report, working through code or "research" modes produce answers around 20 times longer, and cost about that much more.
Where the data centre is, not where you are. AI companies do not say which one answers you, so world average is the honest default.

How far to trust this

Every figure starts from the same calculation as the rest of the site: the energy one answer takes, worked out from how big the model is and how much text it writes. Those figures are good to roughly a factor of two to three, which is why results here are rounded to two digits, why models whose size is secret show a range, and why we only suggest a switch when the saving is at least 5 times.

Carbon depends on the electricity grid the data centre runs on. No AI company says which data centre answered you, so we start from the world average. The grid picker shows what the same use would release on a particular country's electricity; it is not a claim about where your answers are actually made.

The comparisons

A kettle is one litre boiled from cold (100 Wh: the physics says 93, plus a little lost to the air). A phone charge is flat to full on a typical phone (20 Wh from the socket). A light bulb hour is a 10 W LED, the usual replacement for an old 60 W bulb. Driving uses the US Environmental Protection Agency's figure for an average car, which is heavier than most, so the distances are on the short side.

Data built 2026-10-17 · how these numbers are made · the full comparison of 277 models · all tools