In the age of A.I. many of us are going to have to use A.I. tools out of necessity for our job and for our own work.

However these new technologies and new systems come with a cost that is difficult to quantify yet not impossible to approximate. We don’t want to turn a blind eye because we want to live in reality and be thankful as we mindfully count those high costs that the Earth and natural world daily pay to keep humanity alive, entertained and satisfied.

We want to also keep track of what we can do to balance out our own actions. Planting trees, creating rain gardens and exploring other means of sequestering carbon, encouraging wild-life and stewarding water are things that we should already be doing but if we want to balance out our A.I. usage, the A.I usage cost Calculator below can help.

AI Usage Impact Calculator

Estimate the energy, carbon, and water footprint of a single AI query.

How many of each did you do?
Enter a count for any activity (0 to 1,000,000). Leave the rest at zero. Totals add up across all of them.
Quick questions or fact-checks
Back-and-forth conversations
Writing help (email, essay, post)
Document reading or analysis
Image generations
Hard reasoning, coding, or research tasks
Each activity uses a typical token estimate. Switch to Advanced for exact token control of a single query.
Not sure? ChatGPT's default is GPT-5.5; Google's is Gemini Flash; Claude's is Sonnet.

Estimated impact per query

Energy
—
Wh
Carbon
—
g CO₂e
Water
—
mL

To redeem it

What it would take to undo this query's footprint through ecological tending. One query is tiny — these are shown as honest fractions of a year's work by one tree or one garden. What is a rain garden?

Trees to plant
—
each growing for a year to absorb it
Rain gardens
—
100 sq ft, each recharging for a year
This very conversation What it cost to build this calculator — and why long threads add up fast

The calculator you're reading was built through a long back-and-forth with an AI. Here's what that cost — and why long threads cost far more than people expect.

Why a conversation isn't just "a lot of queries":
Every turn resends the entire conversation so far as input. Turn 40 re-reads turns 1–39. So the cost grows roughly with the square of the thread length, not linearly. This thread's final context is about — tokens — but cumulatively the model read closer to —.
Energy
—
Wh
Carbon
—
g CO₂e
Water
—
mL
Assumes prompt caching is on, which is what really happens — providers save their work on the earlier part of a thread instead of redoing it every turn, so re-reading old messages costs roughly 15% of fresh text. Without caching this thread would come to about 1,044 Wh instead, roughly four times higher.
Trees to plant
—
each growing for a year to absorb it
Rain gardens
—
100 sq ft, each recharging for a year
Sources & caveats: Energy is interpolated between measured anchor points from the Jegham et al. "How Hungry is AI?" benchmark (2025), OpenAI/Google official disclosures, and University of Rhode Island GPT-5 estimates. Points between anchors are linear interpolations, not measurements. Water uses region-specific intensity (cooling + optional power-generation water). Carbon = energy × regional grid intensity (IEA 2026). Anthropic and xAI publish no official per-query data — those are third-party estimates. Offset figures assume ~10 kg CO₂/tree/year (young tree) and ~10,000 gallons/year infiltrated by a 100 sq ft rain garden; both vary widely by species, climate, and soil. Real impact varies widely by datacenter, batching, and infrastructure.