AI Impact Tracker estimates ChatGPT's carbon and water use

A new Chrome extension is trying to give regular ChatGPT users a number for something the model itself never shows them: the rough energy, water, and carbon footprint of their conversations. The AI Impact Tracker listing on the Chrome Web Store describes the extension as quantifying "the carbon, water, and energy usage associated with your activity" by analyzing generated responses in real time.

The mechanism, as described on the listing, is straightforward. It counts the output tokens of each ChatGPT response, combines that with a coarse user location (country, state or province, and watershed ID), then models the localized power grid and water resource impacts against that data. Aggregate data is sent to an analysis endpoint at aiimpacttracker.cs.haverford.edu, operated by the developer, the AI and Society Lab. The store page is explicit that the extension does not store the actual content of prompts or responses, and limits collection to output token count, coarse location, and date.

Why this is interesting, even at this scale, is that per-query environmental cost is the part of generative AI that the labs themselves rarely surface to users. ChatGPT does not tell you what a chat cost in kilowatt-hours. A user-facing tool that puts even a rough number on the screen is one way to start a conversation that the underlying platforms have not.

The honest caveat is that the store page gives you very little to verify against. It is version 1.0.1, last updated on June 25, 2026, with three users and no ratings at the time of the listing. The page does not link to a methodology, does not say how token counts map to kilowatt-hours or liters, and the description only mentions ChatGPT. Token count is a coarse proxy for actual compute. What is worth watching is whether the group behind aiimpacttracker.cs.haverford.edu publishes its underlying numbers; that, more than the extension itself, is where the value would come from.