A calm desk by a sunny window with a laptop, a mug of tea, a small potted plant, a green notebook with a pencil and a stack of blank cards.

The short answer: by an estimate from our engineering team, one EASI report, from a 15 to 20 minute English language sample to the finished draft, most likely uses 40 to 60 watt hours of electricity and makes about 15 to 20 grams of CO2. The full range is 15 to 130 watt hours and about 5 to 46 grams of CO2, depending on the sample and the report.

That is about 30 to 45 minutes of streaming video. Below, I explain what EASI is, where that number comes from, how it compares with everyday things, and what it does not include.

A kind question that deserves a real answer

Many of you have asked me a thoughtful question: what does AI cost the planet? I hear it from working clinicians, and I hear it even more from college students and graduate SLP students who are deciding how they want to use AI in their careers. It is a fair thing to wonder about, and you deserve a real answer with real sources, not a slogan. So we asked our engineering team to work it out for EASI, and I am sharing the result here, including the parts that are uncertain.

First, what is EASI?

EASI is a clinician controlled AI platform for speech language pathologists. An SLP records an English language sample, EASI transcribes it, and the clinician reviews and attests each transcript line. Then, with EASI report drafting (in beta), EASI drafts a report for the clinician to review, edit and sign. EASI drafts; the clinician decides. EASI transcribes English only.

So "one EASI report" is more than a single question to a chatbot. It includes transcribing the audio, reading the full transcript, and a series of drafting steps that each work from that long transcript.

How much energy does one AI request use?

Published figures for a single short AI request are small:

One major provider measured 0.24 watt hours and 0.03 grams of CO2e for a median text prompt to its AI app (Google Cloud, August 2025).
An independent research group estimates about 0.3 watt hours for a typical chatbot query (Epoch AI, February 2025).

Long inputs use much more. The same independent analysis estimates about 2.5 watt hours for one query with a long input, and about 40 watt hours for a very long one. That matters for EASI, because every drafting step reads a whole transcript. This is why one EASI report uses far more than one chatbot question, and why we would rather tell you that plainly.

Our estimate for one EASI report

Here is the estimate from our engineering team, for one report from a 15 to 20 minute language sample to the finished draft:

	Most likely	Full range
Electricity	40 to 60 Wh	15 to 130 Wh
Carbon	about 15 to 20 g CO2	about 5 to 46 g CO2

How we got the carbon number: we multiplied the electricity by the US average grid factor from the US EPA's eGRID2023 data, about 0.350 kg of CO2e (CO2e, or carbon dioxide equivalent, counts all greenhouse gases as CO2) per kilowatt hour (EPA eGRID2023 summary data). 40 Wh × 0.350 = about 14 g, and 60 Wh × 0.350 = about 21 g, which we round to about 15 to 20 g. For the full range, 15 Wh × 0.350 = about 5 g, and 130 Wh × 0.350 = about 46 g. California's grid is cleaner, about 0.195 kg of CO2e per kilowatt hour in the same EPA data, so the same report made in California would come to less, about 8 to 12 g at the most likely estimate.

A few honest notes about this number:

It is an estimate, not a measurement. Our engineers built it by reading how EASI's code works and applying published energy rates. We are now logging usage so we can narrow the range with real runs, and we will update this post when we do.
It covers using EASI, meaning the electricity for transcription, drafting and running the app. Like the published AI figures above, it does not include building the chips and data centers or training the AI models.
Longer samples and extra drafting features use more. A longer recording, or asking EASI to draft additional documents, moves a report toward the high end of the range.
We changed our earlier number. An older estimate on our site was much lower. When our team looked again more carefully, it was too low, so we are sharing the newer, larger range instead.

How does one EASI report compare with everyday things?

These comparisons are context, not judgment. We all stream shows, make tea, drink lattes and buy shoes, and none of that makes anyone a bad person. The goal is simply to give the number a sense of scale.

Everyday thing	Its footprint	One EASI report (most likely estimate)	Source
Energy (electricity in use)
Streaming video	About 77 Wh per hour	About 30 to 45 minutes of streaming	IEA, 2020
Boiling water in an electric kettle	About 108 to 115 Wh per liter	About a third to a half of a liter	Fraunhofer ISI and Swedish Energy Agency, 2022
Carbon
A café latte	About 550 g CO2e across its whole life	About one thirtieth of a latte	Nab and Maslin, 2020
A short drive	About 400 g CO2 per mile from the tailpipe	About 200 to 260 feet, a city block or less	US EPA, 2023
A pair of running shoes	About 14 kg CO2e across its life	One pair equals 700 or more EASI reports	MIT, 2013

Fair notes on these comparisons

Whole life versus electricity in use. The latte and the shoes are counted across their whole life, from farm or factory onward. The EASI estimate, the AI request figures, streaming and the kettle are mostly the electricity used at the moment you use them. That makes the carbon comparisons a helpful sense of scale, not an exact match.
Grids vary. The same electricity makes different amounts of carbon in different places. In the US EPA's 2023 data, California's grid releases about 0.195 kg of CO2e per kilowatt hour, while the US average is about 0.350. On a cleaner grid, the same report would make less carbon; on a grid that burns more coal, it would make more.
Numbers change. Hardware and software keep getting more efficient, and published AI estimates vary widely. We will keep checking our own numbers and share updates.

Our commitment

We will keep measuring, keep our method open, and update this post as real usage data comes in. Please check our sources, question our assumptions and ask good questions about this post. Every number above has a source listed below.

If you want the full method, the drivers behind the range and how AI energy per request has changed, read our white paper, The Energy Footprint of an AI Drafted SLP Report.

See how EASI drafts reports for clinicians to review: AI report drafting for SLPs.

Learn more about EASI at NorthernSpeech.com/EASI.

Frequently asked questions