Free Calculator and Whitepaper

Understand the environmental impact of your organisation’s AI use

AI use is becoming part of everyday work, but its environmental impact is rarely included in organisational reporting. This calculator provides an estimate of the electricity, water and carbon emissions associated with AI use, based on your organisation’s number of employees, adoption and daily AI queries.

Designed for organisations to estimate AI use across your workforce
Calculate electricity, water and carbon impacts at organisation and employee level
Peer-reviewed estimates of AI energy use alongside UK carbon and water factors
Download a PDF report to support your Scope 3 assessment

Calculate your AI footprint Read the methodology

Free to use. Results are generated in your browser and take around two minutes.

ESTIMATED IMPACT OF ONE MEDIUM AI QUERY

A single medium AI prompt

Electricity 4.7Wh

Water 22mL

Carbon 0.68g CO₂e

Results across 40 AI users sending 17 queries a day, 253 working days a year: 102 kg CO₂e plus 703 kWh and 3,305 litres of water, every year.

Cross provider averages from arXiv:2505.09598, applied with UK carbon factors. Illustrative only.

AI Carbon Impact Calculator | Climate Essentials

Organisation AI impact report

The environmental impact of AI use at work

Company name	-
Total employees	-
Employees using AI	-
Impact period covered	-
Date of report	-
Calculations by	Climate Essentials™

See your organisation's AI footprint

Enter your email to unlock the calculator. Takes about two minutes.

1Who uses AI

2How much they use it

3Your impact

How many of your people use AI?

Emissions from AI scale with the number of people using it. Start with your headcount, then tell us how many of them are active AI users.

Company name Used only to label your results and the PDF report. It is not part of any calculation.

Total employees Your full headcount, including people who never touch AI. This drives the sector benchmark and the per-employee figures shown at the end.

Enter a headcount of at least 1.

How should we work out your AI user count? Licence seats or survey data are far more accurate than a benchmark. Use the benchmark only when you have no internal data.

Employees actively using AI Licensed seats for tools such as ChatGPT Enterprise, Microsoft 365 Copilot or Gemini for Google Workspace, or the number of active users your staff survey identified.

Sector adoption rates come from the EY 2024 Work Reimagined Survey. They are a starting point, not a measurement of your organisation.

Your sector We multiply your headcount by the share of employees in this sector reported to use AI at work. Pick the closest match.

How much do they use it?

Step 1 gave us the reach of AI in your organisation. This step captures the intensity: how many queries get sent on a typical working day.

Simple: a short prompt and a short answer, such as a definition or a quick email. Roughly 1 to 2 sentences in.
Medium: a longer piece of work, such as drafting a report section or reviewing a document. Around 750 words in and 750 words out.
Complex: a heavy task, such as analysing a long report or a large dataset. Around 7,500 words in and 1,100 words out.

How do you want to enter query volumes? Organisation-wide totals are more accurate. If you only know your own usage, we will scale it up by the AI user count from Step 1.

Simple queries per day Short prompts with short answers. These are the cheapest queries, at roughly 3.3 Wh and 15 mL of water each.

Medium queries per day Longer drafting or review tasks, roughly 750 words in and out. About 4.7 Wh and 22 mL of water each.

Complex queries per day Long documents, large datasets or substantial generated content. About 6.5 Wh and 31 mL of water each.

Enter at least one query.

Your AI impact

These figures cover the inference phase only: the electricity and water used when your people send queries and models generate answers.

Time period Working days only: 5 per week, 20 per month, 253 per year. Weekends and public holidays are excluded.

Results

Electricity

0kWh

Water

0litres

Carbon

0kg CO₂e

Methodology and scope

This calculator estimates the environmental impact of the inference phase of AI use: the electricity and water consumed in the data centre each time one of your employees sends a query and a model generates a response.

Included

Data centre electricity used to process employee queries
Transmission and distribution losses on that electricity
Water consumed at the data centre and at the power stations supplying it
Carbon associated with supplying that water

Not included

Model training: the energy used by developers to build the models
Embodied carbon of servers, GPUs, buildings and network hardware
Employee devices, office electricity and network transport
AI embedded in software you use indirectly, such as search or recommendation engines
Wastewater treatment, where water is returned to sewer rather than evaporated

Where the numbers come from

Energy and water per query: averaged across eight major model providers, from arXiv:2505.09598. We apply the cross-provider average: about 3.3 Wh and 15 mL for a simple query, 4.7 Wh and 22 mL for a medium query, and 6.5 Wh and 31 mL for a complex query.
Electricity carbon factor: 0.14395 kg CO₂e per kWh, being UK grid generation (0.13096) plus transmission and distribution losses (0.01299).
Water carbon factor: 0.1913 kg CO₂e per cubic metre, water supply only. Treatment is excluded because evaporative cooling consumes water rather than discharging it to sewer.
Sector adoption rates: EY 2024 Work Reimagined Survey.
Working days: 253 per year, 20 per month, 5 per week.

What this means for your carbon accounting

Under the GHG Protocol these emissions are not a separate category. Electricity consumed in a third party data centre you do not own or control sits in Scope 3, Category 1 (purchased goods and services) as part of the AI service you buy, or Category 8 if you lease the infrastructure. If you already account for AI subscriptions on a spend basis, adding these figures on top would double count them.

How accurate is this?

Treat the output as an order of magnitude, not a reportable figure. The main sources of uncertainty are:

Per-query energy varies by more than five times between providers and models. A footprint calculated on the cross-provider average could be substantially wrong for an organisation standardised on one tool.
Where a query lands between simple, medium and complex is a judgement, and users are poor at estimating their own volumes.
UK grid factors are applied throughout, but inference may run in any region. A query served from a coal-heavy grid carries several times the carbon of one served from a low carbon grid.
Data centre water use depends on the cooling system and local climate, both of which vary widely.
Because training and hardware are excluded, the true lifecycle footprint of your AI use is higher than the figure shown.

Full method: The Carbon Impact of Artificial Intelligence: Implications for Business Footprints, Climate Essentials™ (2026).

Cover of the Climate Essentials whitepaper on the carbon impact of artificial intelligence

Free whitepaper

The full methodology, free to download

This paper sets out the methodology, assumptions, and carbon factors used by Climate Essentials to develop this calculator. The report also has recommendations to help your business reduce the environmental impact of AI usage in the workplace.

The methodology

How per query energy and water figures are derived, and which carbon factors are applied to each.

The assumptions

What sits inside the boundary, what we have excluded, and where the numbers are least certain.

Where AI sits in Scope 3

How to treat AI use under the GHG Protocol without double counting it against your software spend.

Recommendations

Practical steps to cut AI related emissions, from model choice to query design and procurement terms.

Download the whitepaper

PDF, free. We will email you the link and occasional updates on carbon reporting.