Sources and licences

US beef

Senorpe Asem-Hiablie, Thomas Battagliese, Kimberly R. Stackhouse-Lawson and C. Alan Rotz, A life cycle assessment of the environmental impacts of a beef system in the USA (published online 2018, journal volume 2019). © The Author(s) 2018. CC BY 4.0.

The calculator selects the global warming and absolute freshwater consumption results in Table 3. Both refer to one kilogram of boneless beef eaten in the modelled US system. The calculator scales them to the portions entered.

French foods, drinks and meals

Source ADEME, données AGRIBALYSE® v3.2 (2025). The calculator uses the simplified product indicators under the Etalab Open Licence 2.0. Product data updated 27 February 2025, metadata updated 30 June 2026, accessed 4 November 2026. See ADEME’s conditions for application reuse.

The calculator includes all 2,451 published product rows and selects nine indicators per kilogram of food, including drinks. It uses ADEME’s English product names, shortens selected labels and translates the category names. Freshwater and marine nutrient pollution are converted from kilograms to grams of phosphorus and nitrogen equivalent, respectively. Smog formation is expressed in grams of NMVOC equivalent and mineral resource depletion in milligrams of antimony equivalent. Climate, soil quality, water scarcity, acidification and fossil resource use retain their source units. Each record links its AGB identifier and carries its data-quality rating. The common food choices use a single identified product record and its data-quality rating. The extract covers the public aggregate results only. See ADEME’s data quality guidance.

The ten French meal scenarios use ADEME’s Impact CO₂ recipe estimates, which cite AGRIBALYSE v3.2 data dated 15 January 2025. We translated the meal labels and scaled the published results by meal count. See the source recipe records and the Impact CO₂ repository’s MIT notice, © 2022 Datagir. ADEME has not endorsed this calculator.

UK diets

Peter Scarborough, Michael Clark, Linda Cobiac, Keren Papier, Anika Knuppel, John Lynch, Richard Harrington, Tim Key and Marco Springmann, Vegans, vegetarians, fish-eaters and meat-eaters in the UK show discrepant environmental impacts (2023). © The Author(s) 2023. CC BY 4.0.

We selected aggregate diet-group estimates from Tables 3 and 4, shortened the category names, converted water volumes from cubic metres to litres and scaled the daily factors. The study’s estimates apply to groups standardised to 2,000 kcal per day.

Surface travel and flights

UK Department for Energy Security and Net Zero, UK greenhouse gas reporting conversion factors 2025. © Crown copyright 2025. Contains public sector information licensed under the Open Government Licence v3.0. We selected published factors, allocated the shared car across two passengers and used the international economy factor for the Paris–New York return trip.

© Commonwealth of Australia, Department of Finance, Emissions Reporting Framework 2025–26, Table 22. The selected Australian reporting factor is derived from UK DESNZ data. Adapted under CC BY 4.0, subject to the department’s exclusions.

Airport coordinates for Hobart, Melbourne, Sydney, Paris CDG and New York JFK come from OurAirports’ public-domain data, accessed 4 November 2026. We calculated great-circle distances using a spherical Earth radius of 6,371.0088 km and multiplied them by the relevant emissions factor. These estimates use airport distances rather than observed flight tracks.

Household water

© Commonwealth of Australia, Water efficiency, accessed 4 November 2026. Adapted under CC BY 4.0, subject to the website’s copyright notice. We use selected WELS band endpoints for showers and 6 kg washing machines, and illustrative full- and half-flush volumes. The general dual-flush example uses the WELS mix of four half flushes and one full flush. Results scale with minutes, loads or flushes.

Appliance electricity and water

© Commonwealth of Australia, Department of Climate Change, Energy, the Environment and Water. Energy Rating Data for household appliances – Labelled Products, GEMS extracts dated 3 November 2026, under CC BY 3.0 Australia.

We selected supplier registrations for named models and retained the registration numbers and capacities. We converted annual electricity to rated cycles or operating days using the regulator’s usage assumptions. Wash water comes from the tested program, and television viewing uses measured on-mode power. These examples cover operation.

The general 8 kg dryer estimate takes the median of 128 distinct registrations in the 3 November 2026 clothes dryer extract. For the 14-place dishwasher and 55-inch television, each measure uses the median of the two selected models of that size. The general washing machine, fridge-freezer and chest freezer choices each use one identified model. Calculation details show the reference or contributing values for each choice.

Australian grid electricity and fuels

DCCEEW 2026, Australian National Greenhouse Accounts Factors, Department of Climate Change, Energy, the Environment and Water, Canberra, August. © Commonwealth of Australia 2026. CC BY 4.0. See the department’s copyright notice.

We combined the location-based grid and upstream factors from Table 1. Tables 5–6 supply gas combustion and regional upstream factors, including the suggested proxies for Tasmania and the Northern Territory. Tables 8–9 supply fuel energy content and emissions factors for road and stationary fuel use. Each record gives the source units and conversion.

UK materials, motorcycles, taxis and ferries

UK Department for Energy Security and Net Zero, Greenhouse gas reporting: conversion factors 2026. © Crown copyright 2026. Contains public sector information licensed under the Open Government Licence v3.0.

We selected published factors from the June full workbook. Motorcycle estimates use one rider per vehicle. Taxis and mixed passenger/freight ferries use passenger-kilometres. Material purchases use the primary, reused or closed-loop recycled column in the Material use worksheet, converted from tonnes to kilograms. The material factors cover procurement through the point of sale. See the 2026 methodology for the accounting boundaries.

Gemini text prompts

Cooper Elsworth, Keguo Huang, David Patterson, Ian Schneider, Robert Sedivy, Savannah Goodman, Ben Townsend, Parthasarathy Ranganathan, Jeff Dean, Amin Vahdat, Ben Gomes and James Manyika (2025), Measuring the environmental impact of delivering AI at Google Scale. CC BY 4.0.

We converted the May 2025 Gemini Apps median text-prompt estimates to kWh, kg CO₂e and litres of on-site cooling water, then scaled them to the request count. The paper’s scope is inference on the measured system.

ChatGPT and Le Chat

Sam Altman reported two approximate ChatGPT averages in The Gentle Singularity (June 2025). We converted Wh to kWh and assumed US liquid gallons to convert water to litres. The article gives no measurement method or water boundary.

Mistral AI’s July 2025 life-cycle study, conducted with Carbone 4 and ADEME, reports three marginal impacts for a 400-token Le Chat response. We converted grams of CO₂e to kilograms and millilitres of water to litres. Mineral depletion retains the published unit. The original articles and illustrations remain with their publishers.

AI text, image and video tasks

Alexandra Sasha Luccioni, Yacine Jernite and Emma Strubell, Power Hungry Processing, corrected version, Table 2. The adapted task subset is available under CC BY-SA 4.0. We selected task means and standard deviations and divided the means by 1,000 for one inference.

Julien Delavande, Regis Pierrard and Sasha Luccioni, Video Killed the Energy Budget (2025), Tables 2 and 4. The video subset is adapted under CC BY 4.0. We selected four clip configurations, summed the three reported component means for each and converted Wh to kWh.

Felipe Oviedo, Fiodar Kazhamiaka, Esha Choukse, Allen Kim, Amy Luers, Melanie Nakagawa, Ricardo Bianchini and Juan M. Lavista Ferres, Energy Use of AI Inference, Efficiency Pathways, and Test-Time Scaling (2026), Table 1. We selected the simulated DeepSeek-R1 long-query median and interquartile range, and converted the median to kWh. The adapted reasoning subset is available under CC BY-SA 4.0.

Download the adapted task, video and reasoning records. Each subset retains its source and licence.

AI coding and web-app development

Jae-Won Chung and coauthors, The ML.ENERGY Benchmark (2025), v3 dataset, Apache License 2.0. We selected two Qwen3-Coder configurations from the September 2026 leaderboard snapshot and converted joules per generated token to kWh.

The web-app scenarios apply these GPU measurements to AI-authored budgets for planning, implementation, tests and revisions. The benchmark measured code completion. The budgets and their allocation to development stages are assumptions made for this calculator.

Arihant Tripathy, Ch Pavan Harshit and Karthik Vaidhyanathan, SWEnergy, Table 1. The bug-fix attempt uses one reported AutoCodeRover result converted from kJ to kWh, with the study’s corresponding success rate. The paper and its tables remain with their rights holders.

Everyday electricity comparisons

The calculator supplies the assumed input watts and run times for the bulb, kettle, microwave and laptop examples. Energy is input power multiplied by time. The US Energy Information Administration’s guide to measuring electricity explains the units.

Cotton T-shirt

Gerald Feichtinger, Andrea Katharina Scherer and Wolfgang Posch (2026), Reducing the environmental footprint of cotton T-shirt production through automation and reshoring, scenario ESM1 in the supplementary results. © The Author(s) 2026. CC BY 4.0. We use the published climate result for producing and distributing a 200 g T-shirt and summarise the scenario in our own words.

Electronic devices

Contains information from Boavizta’s Environmental Footprint Data, accessed 4 November 2026. The adapted electronics database is made available under the Open Database License 1.0. Its individual contents are available under the Database Contents License 1.0.

Download the complete adapted electronics database (JavaScript). We selected eight Apple models, retained their reported carbon totals and added labels, scope notes and first-owner use periods. Each record preserves the original lifetime field and links the manufacturer’s report. The general device choices reference individual models in this download. The database licences apply to this subset. The reports remain with their publishers.

Labour evidence

Good On You’s Uniqlo rating is available directly from its publisher. The catalogue links to the rating under the publisher’s website terms.

The Fast Retailing entries summarise selected factual findings in our own words from KnowTheChain’s 2023 apparel and footwear benchmark and the company’s production-partner disclosure. These are company-level references. The reports, supplier list, artwork and logos remain with their respective rights holders.