The honest answer: at an individual level, very little. A single text prompt in Copilot, ChatGPT or Gemini uses roughly the energy of a few seconds of television and a few drops of water. Move the slider to see your own footprint in context.
Build a realistic day below. Watch what happens to the totals - the type of activity matters far more than how many you run.
These come up in nearly every workshop. Worth addressing head-on before the detail.
Built on a 2023 estimate of an old model (long outputs, older chips) versus a 2009 Google figure. Corrected estimates put a typical text query at ~0.3 Wh - about the same as that old search number. The "10×" gap has largely collapsed.
Microsoft has not published a per-prompt figure for Copilot. The precise "0.31 Wh" doing the rounds traces to an AI-generated website with obvious errors. Copilot runs on the same GPT models as ChatGPT, so ~0.3 Wh is a reasonable proxy - just don't quote a false precision.
That figure is for GPT-3 (2023), is a model estimate (not a meter reading), and mixes on-site cooling with power-station water. Modern per-prompt water is a few millilitres on-site (Google measured 0.26 mL for Gemini).
The twelve most-repeated claims about AI and the environment, each rated on how well the evidence holds up. Click any claim to see what's actually claimed, the best current evidence, and what's being done about it.
Each claim was researched across multiple independent sources, then stress-tested three ways:
Some numbers are solid; others are genuinely contested. We've flagged confidence on every claim. The biggest live uncertainties:
Per-prompt / personal use:
Google Cloud, "Measuring the environmental impact of AI inference" (21 Aug 2025) - link · Epoch AI, "How much energy does ChatGPT use?" (Feb 2025) - link · MIT Technology Review AI energy series (May & Aug 2025) · Hannah Ritchie, "What's the impact of AI on energy?" - link · Sam Altman, "The Gentle Singularity" (Jun 2025).Data centres, grid & emissions:
IEA, Energy and AI (Apr 2025) - link · LBNL, 2024 United States Data Center Energy Usage Report (Dec 2024) - link · Patterson et al., "Carbon Emissions and Large Neural Network Training" (2021) · Strubell et al. (2019) & its later corrections · Luccioni et al., "Carbon Footprint of BLOOM" (JMLR 2023) · Google 2024 & Microsoft 2024 environmental/sustainability reports.Water, e-waste, mitigations:
Li, Yang, Islam & Ren, "Making AI Less Thirsty" (arXiv 2304.03271 / CACM 2025) · Wang et al., "E-waste challenges of generative AI" (Nature Computational Science, Oct 2024) · Xu/Ricks/Jenkins et al., Joule (Jan 2024) on clean-power matching · Luccioni et al., "From Efficiency Gains to Rebound Effects" (FAccT 2025) · Constellation, Kairos Power and X-energy nuclear announcements (2024) · EU Energy Efficiency Directive & AI Act reporting provisions.Full source URLs with confidence ratings are held in the research file accompanying this briefing.