Inspiration

"What if people could actually see the environmental cost of AI — in real time?"

  • A single 100-word ChatGPT prompt consumes 519ml of fresh water just to cool servers
  • AI data centers projected to use 12% of all US electricity by 2028
  • AI could emit 44 million metric tons of CO₂ annually by 2030
  • No platform talks about this at the personal level — we wanted to change that
  • Framework: UN SDGs 6 (Clean Water) · 7 (Clean Energy) · 13 (Climate Action)

What it does

A Chrome extension that silently tracks your AI environmental footprint across every major platform — without changing a single habit.

Tracks across: ChatGPT · Claude · Gemini · Midjourney · Perplexity · GitHub Copilot

For every prompt it calculates:

  • 💧 Water consumed for server cooling
  • ⚡ Electricity used
  • 🌫️ CO₂ emitted

Converts numbers into human comparisons: water bottles · lightbulb minutes · meters driven

Dashboard features:

  • Today / Session / All-time breakdown
  • Platform-by-platform cost ranking
  • Personalized Eco Score (A+ to D)
  • Claude API-powered green suggestions based on your actual patterns
  • Prompt optimizer — rewrites prompts to be shorter before you hit send

How we built it

Three-layer architecture:

Layer What it does
🔍 Detection Content script detects prompt submissions via DOM listeners across each platform
⚙️ Calculation Background service worker maps interactions to research-backed impact constants
🧠 Intelligence Claude API (claude-sonnet-4-20250514) powers tips, narratives, and prompt optimization

Stack: Vanilla JS · Chrome Manifest V3 · chrome.storage.local · Claude API · Space Mono + DM Sans


Challenges we ran into

  • DOM detection — every AI site uses different button selectors, content-editable structures, and shadow DOM patterns that change with every update
  • Water vs. energy trade-off — evaporative cooling saves energy but uses more water; air cooling does the opposite — hard to represent accurately without oversimplifying
  • UX tone — early versions felt preachy; took several rewrites to land on data-forward and empowering rather than guilt-inducing
  • Manifest V3 restrictions — no persistent background pages; all Claude API calls had to be routed through the service worker via chrome.runtime.sendMessage

Accomplishments that we're proud of

  • ✅ Works entirely passively — zero behavior change required from users
  • Prompt optimizer is the only tool in this space that intervenes before the resource is consumed
  • ✅ Every number backed by peer-reviewed research — UC Riverside · Nature Sustainability · MIT · IEA · Cornell
  • ✅ Fully demoable product built in under 24 hours addressing 3 UN SDGs simultaneously

What we learned

  • The environmental cost of AI is invisible by design — no platform publishes per-query resource data; we stitched together 6 research papers just for baseline numbers
  • Behavior change tools work best when ambient, not interruptive — the quiet toast notification after each prompt is more effective than the dashboard
  • Using the Claude API as a generative backend (not just a chatbot) unlocked patterns we hadn't considered before

What's next for EcoTrace AI

Feature Description
📍 Regional carbon intensity Adjust CO₂ in real time based on local grid energy mix
🏢 Corporate dashboards Team-level ESG reporting aligned with CSRD & SEC requirements
🤖 Greener model recommendations Flag when a smaller model (Haiku, GPT-4o mini) would do the same job
📱 Mobile companion app Extend tracking to Siri, Google Assistant, and AI-embedded apps
🔌 Open API REST endpoint so any app can get environmental cost per interaction

Built With

Share this project:

Updates