See the energy behind every AI prompt.

A free Chrome extension that estimates the energy, carbon and water behind your ChatGPT, Claude, Gemini and Perplexity chats.

Add to Chrome, it's free How it works
  • No sensitive data. Tokens are counted on your device.
  • No account needed
Today 3 prompts
  1. ChatGPT 150 tokens in · 1,400 out 1.59 Wh0.56 g CO₂e · 5.4 ml
  2. Claude 200 tokens in · 1,800 out 2.29 Wh0.88 g CO₂e · 7.4 ml
  3. Gemini 400 tokens in · 900 out 0.85 Wh0.33 g CO₂e · 3.5 ml
4.72 Wh 1.77 g CO₂e
16 ml water

About a 10 W LED bulb left on for half an hour.

An example day with typical answer lengths. Estimates, not provider data. Only token counts are recorded, never your prompts.

How it works

  1. Add it to Chrome

    One click. No sign-up.

  2. Chat as usual

    Tokens are counted on your device. Your messages are never stored or sent anywhere.

  3. See your footprint

    Energy, carbon and water for each conversation, with the working shown.

One prompt is tiny. Billions aren't.

0.34Wh

What OpenAI's chief executive says an average ChatGPT query uses. He didn't say which model, or what the number includes.

Sam Altman, June 2025

0.24Wh

What Google measured for a median Gemini text prompt — counting the chips, the servers around them, spare capacity and cooling.

Google, August 2025, measuring May 2025

30–100×

How much more energy the answer costs than the question, in our measurements. The exact ratio depends on the model and how it's run.

Our draft paper, August 2026

Measured, not assumed.

The estimates are built on energy we measured ourselves on real GPUs, one request at a time.

~197k

requests measured

4

open models, 1B–12B

2

GPU generations, same energy per token

0.985

R² of our per-token equation

Draft paper, August 2026. Ask us for a copy.

Small habits that help

  • Ask for shorter answers.Most of the energy goes into the reply.
  • Use a smaller model for small jobs.Bigger models use more energy per word.
  • Regenerate less.A clearer prompt beats five retries.
  • Use reasoning modes when you need them.Their hidden “thinking” counts too.

For teams

Need AI usage in your sustainability reporting? We're working with a few early teams.

Talk to us

FAQ

How accurate are the numbers?

They are estimates. Our energy equation comes from our own measurements, but providers don't publish model sizes or how they run their data centres, so parts of each estimate rest on assumptions. The extension shows its working for every figure.

For carbon we use the average grid intensity where the data centre sits. A provider quoting its own clean-energy purchases will report a lower number for the same prompt — both are valid, they answer different questions.

Do you see my data?

No. Tokens are counted on your device, and only numbers are kept: token counts, timings and which service you used. They stay in your browser and are never sent to us. Your prompts and replies are never stored. See our privacy policy.

Which services does it work with?

ChatGPT, Claude, Gemini and Perplexity in Chrome.

Should I stop using AI?

No. The idea is to make the cost visible so everyone can make sensible choices.

Try it on your next prompt.

Add to Chrome, it's free