What do your purchases really cost the planet?

CarbonEye uses AI to identify objects — but the AI never invents footprint numbers. All impacts are computed from a fixed dataset of published emission factors.

CO₂e means "CO₂ equivalent" — all greenhouse gases rolled into one number by their warming effect.

1. 1

   Identify & decompose

   A single AI vision call (Gemini) recognizes the objects and estimates their physical makeup: which materials and how many kilograms of each, where the product was likely manufactured, its typical yearly electricity and water use, and its service life.

2. 2

   Compute deterministically

   A local calculation engine multiplies each material's mass by emission factors drawn from public life-cycle-assessment literature (~40 materials, each with kg CO₂e, liters of water, and embodied energy per kg). On top of that it adds:

   • Manufacturing — a category-specific overhead for forming, assembly and factory energy (e.g. +30% for electronics, +50% for apparel).
   • Transport — freight emissions by route (sea, air, regional, local), including packaging mass.
   • Lifetime use — yearly electricity × lifespan × world-average grid intensity (0.4 kg CO₂e/kWh), plus water and consumables used over its life (pet food for a dog, fuel for a car, coffee for a coffee maker).
   • End of life — average collection and disposal processing.
3. 3

   Validate against published data

   The goal is to land in the right ballpark: within an order of magnitude (a factor of 10) of reality. To check that, the system is tested against 45 products with published footprints — manufacturer environmental reports, Environmental Product Declarations, and peer-reviewed LCA studies. Across all 45, every estimate lands within 10× of its published value, 93% within a factor of 3, and most within 2×.

The everyday comparisons — and where they come from

• 0.17 kg CO₂e per km — average petrol car including fuel production; UK DEFRA greenhouse-gas conversion factors.
• 21 kg CO₂ per tree-year — a mature tree absorbs ~48 lb/year; USDA Forest Service / Arbor Day Foundation estimate.
• 65 L per shower — ~8 minutes at ~8 L/min with a standard showerhead; US EPA WaterSense.
• 10 kWh per day — typical European household electricity (~3,600 kWh/year); Eurostat / IEA household averages.
• AI text prompt: 0.1–2 Wh and 0.0003–0.01 L of water (central estimate ~0.3 Wh / ~0.0015 L) — triangulated from three kinds of sources, which disagree: Google's self-reported median Gemini prompt (0.24 Wh; 0.00026 L counting on-site cooling only), OpenAI's stated ~0.34 Wh average, and the independent "How Hungry is AI?" benchmark (2025), which measured GPT-4o at 0.42–1.8 Wh depending on prompt length and counts total water including electricity generation (~3.4 L/kWh). Reasoning models on long prompts can exceed 30 Wh — far above this band. The on-page comparisons use the full range.
• 0.4 kg CO₂e per kWh — world-average grid intensity used for lifetime electricity; IEA (~0.44 kg/kWh and falling).
• Material factors — compiled from the University of Bath ICE embodied-carbon database, Water Footprint Network studies, and published manufacturer LCA reports/EPDs; freight factors from DEFRA per-tonne-km values.

What to keep in mind

• These are order-of-magnitude estimates using global-average factors. A specific product can differ several-fold with recycled content, factory location, brand and shipping route — which is why headline figures are shown as ranges (roughly "half to double" per lifecycle stage; wider for water).
• Use-phase electricity assumes a world-average grid — a coal-heavy or hydro-rich grid shifts it strongly either way.
• Packaging materials, repairs and retail overheads aren't modeled (packaging's shipping weight is). Food uses a handful of coarse categories.
• When an item's materials fall outside our dataset — or it isn't a manufactured product at all (a pet, a service, a building) — the result carries a visible data-coverage warning and a widened range.
• CarbonEye is built for everyday curiosity, not professional reporting. The goal is to make the hidden cost of ordinary things visible — and get people thinking about it. It aims to be right within an order of magnitude, not to the decimal. For real decisions or formal reporting, a proper LCA or Environmental Product Declaration is the right tool.