How CO2Scan estimates are calculated, what data sources we use, and what our results mean.
CO2Scan is an AI-powered carbon footprint estimation tool that provides indicative lifecycle carbon estimates for food, drink, and beauty and personal care products. Estimates are expressed in kilograms of CO₂ equivalent (kg CO₂e) per retail unit.
CO2Scan estimates are based on lifecycle assessment (LCA) methodology — a scientific approach that accounts for environmental impacts across the full chain of a product's life, from raw material extraction through to disposal.
CO2Scan draws on three primary sources to generate estimates:
The primary source of product data for food and drink. A free, open and collaborative database of food products contributed by volunteers worldwide. Available under the Open Database Licence (ODbL). Fields used include product name, brand, quantity, categories, ingredients, packaging, country of origin, manufacturing location, labels and Eco-Score data.
The primary source of product data for beauty and personal care products. A sister database to Open Food Facts covering cosmetics, toiletries and personal care items. The same data fields are retrieved where available.
Where product data is available, it is passed to Claude (claude-sonnet) along with a detailed system prompt incorporating lifecycle assessment methodology and published carbon intensity reference data. Where no database match is found, Claude estimates from product name or image alone at lower confidence.
CO2Scan estimates follow a cradle-to-grave lifecycle assessment approach, covering all major stages of a product's carbon footprint:
Each stage is expressed as a percentage of the total estimated footprint and as an absolute figure in kg CO₂e.
The AI model is calibrated against published lifecycle assessment data from peer-reviewed sources including the University of Oxford's Food and Climate Research Network, Our World in Data, and WRAP. The following reference anchors are used:
| Product | Reference range (kg CO₂e per retail unit) |
|---|---|
| Single avocado (imported, air freight) | 1.5–2.5 kg |
| 250g ground coffee (imported) | 3.5–5.0 kg |
| 1 litre whole milk | 2.5–3.5 kg |
| 500g chicken breast | 3.0–4.0 kg |
| 500g beef mince | 12.0–15.0 kg |
| 400g baked beans (canned) | 0.5–0.8 kg |
| 500g white bread | 0.8–1.2 kg |
| 500g cheddar cheese | 5.0–6.0 kg |
| 330ml beer (local) | 0.3–0.5 kg |
| 75cl wine (imported) | 1.5–2.0 kg |
CO2Scan applies a proprietary A–E grading system to each estimate, based on the total estimated carbon footprint per retail unit. This grading system was developed by CO2Scan and is not an officially certified or regulated carbon label.
| Grade | Range (kg CO₂e) | Impact level | Examples |
|---|---|---|---|
A |
Under 0.5 kg | Very low | Most fruits, vegetables, soap bars |
B |
0.5–2.0 kg | Low | Grains, plant-based foods, small toiletries |
C |
2.0–5.0 kg | Moderate | Chicken, eggs, shampoo, moisturisers |
D |
5.0–12.0 kg | High | Pork, hard cheeses, fragrances |
E |
Over 12.0 kg | Very high | Beef, lamb |
Each estimate is assigned a confidence level based on the quality and completeness of available product data:
Users and clients should be aware of the following limitations when interpreting CO2Scan estimates:
Estimates are generated using AI modelling and published reference data. They are not the result of certified lifecycle assessment audits.
Product data in Open Food Facts and Open Beauty Facts is crowd-sourced and may be incomplete, outdated or inaccurate for some products.
Where database data is unavailable, estimates are based on AI modelling from product name or image alone and carry greater uncertainty.
Carbon footprints vary by season, supplier, geography and production method. CO2Scan estimates reflect typical averages and may not capture product-specific variations.
Beauty and personal care product carbon data is less established than food product data. Estimates for these categories carry higher uncertainty.
CO2Scan estimates should not be used as the sole basis for regulated environmental claims, carbon offsetting or compliance reporting without independent verification.
CO2Scan uses data from the following open sources under their respective licences:
Carbon estimation is performed using Anthropic's Claude AI model. CO2Scan is not affiliated with Open Food Facts, Open Beauty Facts or Anthropic.
This methodology document is published under version 1.0. Significant changes to the estimation approach will result in a new version number being issued.
Get in touch — we welcome feedback from researchers, businesses and users.