A Satirical (but Serious) Guide to Greener Chatbots
My friend Ed is now regularly using AI at work. Given the need for everyone to do their part to reduce carbon emissions, he asked how he might write greener prompts. Since he knew that Zai and I were writing about chatbots, he asked us.
It turns out, Zai loves writing white papers. Here is what I learned (Zai, of course, already knew the answer).
Title: “Your Prompt Used How Much Power? ”
Intro: Why AI Sustainability Isn’t Just a Buzzword
Many companies today have climate pledges. But look closely—most skip the digital part of the math. Enter chatbots and AI models: those tireless answer machines quietly devour compute cycles every time you ask, "What time is it in Brussels?"
Let's get serious (and a little satirical) about this.
TL;DR;
A single chatbot query (like this article) may consume ~0.05 watt-hours of electricity, or around 0.02 grams of CO2 emissions. Bigger prompts or those involving images can consume up to 10 times or 100 times more energy.
What Does a Prompt Cost?
Standard Prompt (~400 tokens):
- ~0.05 Wh / 0.02g CO2e
That’s about the same as:
- Toasting a Pop-Tart for 1.2 seconds
- Holding a Zoom call with no one else on it for 4 seconds
- Drawing a pie chart in PowerPoint, no one will ever see
- Accidentally hitting “Reply All”
- Charging an Apple Pencil by 2%
Power Prompt (image generation or 1,000+ tokens):
- ~1 to 5 Wh / 0.5–2g CO2e
That’s like:
- Air-frying one tater tot
- Watching one ad for a product you’ll never buy
- Spinning up a dev container that breaks in four minutes
How We Are (Not) Tracking Emissions
Claim | What They Leave Out |
“We only track commercial flights.” | Helicopter taxis and private jets mysteriously emit 0g. |
“We’re 100% renewable.” | Except the coal-powered cloud provider down the street. |
“Our AI helps reduce emissions.” | While it cooks a GPU farm to tell you that. |
“We use sustainable packaging.” | Which we ship overnight from three continents. |
Now, we need to be fair about this. Policy always lags technology. We are already doing a better job understanding our emissions. And, due to changing technologies, we will always have to update our policies.
What We Can Do
Clean Up Our Prompts
- Be concise
- Avoid unnecessary examples, stories, or markdown fluff
- Ask only what we need
Stop Training on Trash
Most documentation contains unnecessary branding language, obsolete processes, and unverified assumptions. Instead, train models using verified, high-quality data sources.
Use the Right Tool for the Job
Don't waste energy having AI calculate a simple 20% discount. Instead, use:
- Embedded calculators
- Internal documentation search
- Code snippets or macros
Add Energy Footers to Every Response
“This answer used ~0.03 watt-hours of energy (~0.015g CO2e), or about the same as scrolling Slack for 6 seconds.”
Make people see the impact.
Scold Users (Lovingly)
“You used GPT-4o to find a date in Excel. Do better.” “That prompt could’ve been answered by looking out the window.”
Let’s Build This
You can build these features into tools:
- UI widgets that show real-time prompt energy cost
- Prompt filters that recommend smaller models or alternate tools
- Logs that feed ESG dashboards
- Snarky AI personas like Carbon Carl
Conclusion: Net Zero Needs to Include the Nerd Stuff
Carbon guilt doesn't have to be miserable. We can be clever, purposeful, and drive real change. If we are serious about CO2, it's time to count all the emissions—yes, even the ones burned just to tell us what time it is in Brussels.
Red Hat Creative can help. Your first shame-inducing energy overlay is free.