Should You Be Worried About the Environmental Impact of Using ChatGPT?

Key takeaways: Personal AI use has minimal environmental impact compared to everyday activities. The evidence suggests your time is better spent on higher-impact climate actions.

I have a confession: I'm addicted to AI. Claude, ChatGPT, Gemini, Grok… I don't discriminate. Chatting with LLMs has become a reflex for me, in the same way that using Google or social media became a reflex in the 2000s.

But I'm also concerned about the environment. So when I started seeing headlines about AI's massive carbon footprint, I felt genuinely conflicted. Was my new habit contributing meaningfully to climate change?

After examining the available evidence—particularly thanks to excellent analysis by my colleague Andy Masley, Director of Effective Altruism DC, in his blog post on this topic—I've become quite confident that personal ChatGPT use isn't a priority environmental concern. Here's the reasoning, drawing heavily from Andy's research, plus what the evidence suggests you might consider focusing on instead.

The Scale of Personal AI Use

The most important consideration, in my view, is putting AI's energy consumption in proper context. Andy's analysis shows that whilst a ChatGPT query uses roughly 10 times more energy than a Google search, both remain tiny in absolute terms:

• A Google search uses approximately 0.3 watt-hours (Wh) of energy
• A ChatGPT query uses about 3 Wh
• The difference is just 2.7 Wh

To calibrate this: 2.7 Wh could run your laptop for three minutes. ChatGPT could generate an entire blog post using less energy than you'd consume reading it.

The comparison that convinced me: watching one hour of Netflix appears to have the same climate impact as asking ChatGPT 300 questions. I suspect announcing "I asked ChatGPT 300 questions today" might raise environmental eyebrows, whilst "I watched Netflix for an hour" wouldn't.

Confidence level: I'm reasonably confident in these energy estimates, though they're based on limited public data and may vary significantly between different AI systems and usage patterns.

The Scaling Question

A reasonable objection: "But what if everyone dramatically increases their AI use?"

Andy makes a crucial point here about how we frame comparisons. This scaling concern applies to virtually every activity we do. The evidence suggests we need to compare like with like: individual activities against other individual activities, or global trends against other global trends.

Headlines warning that "ChatGPT uses as much energy as 20,000 households" sound alarming because they compare global AI consumption to your personal life. But as Andy notes, if they said "ChatGPT uses roughly 0.05% as much energy as American cars," the relative priority becomes clearer.

Confidence level: I'm quite confident this framing issue explains much of the confusion around AI's environmental impact, though reasonable people might still prioritise differently based on their values and risk tolerance.

What the Evidence Suggests Matters More

If reducing your environmental impact is the goal, the data points toward focusing on larger emission sources:

• Transport: Living car-free where possible, reducing flights
• Home energy: Renewable heating, improved insulation
• Diet: Reducing meat consumption, particularly beef
• Consumption: Buying less, using things longer

These changes appear to reduce emissions by orders of magnitude more than adjusting AI usage.

However, I'd suggest that individual lifestyle changes, whilst valuable, likely aren't how we'll solve climate change. Research suggests that carefully targeted systemic interventions—through career choices, political engagement, and charitable giving—might be significantly more effective per unit of effort.

For instance, some analysis indicates that well-researched climate donations can reduce emissions more cost-effectively

than personal lifestyle changes, sometimes by factors of 10-100. If you're interested in exploring this approach, organisations like Doneer Effectief evaluate climate charities for effectiveness.

Confidence level: I'm moderately confident in the relative effectiveness of these approaches, though the precise multipliers depend on many assumptions and individual circumstances.

Important Limitations

This analysis has several limitations worth acknowledging:

• Energy consumption data for AI systems is often proprietary and estimates vary widely
• Future AI usage patterns could change dramatically as the technology evolves
• Environmental impact includes factors beyond energy consumption (water use, hardware manufacturing)
• The "opportunity cost" argument assumes you'd otherwise spend time on higher-impact activities

Additionally, there are legitimate concerns about AI's development that extend beyond personal environmental impact—questions about industry-wide energy use, job displacement, and ensuring these systems remain beneficial as they become more powerful.

The Bottom Line

Based on the available evidence, I believe personal AI use isn't a priority environmental concern. The data suggests your climate-focused energy is likely better spent on transport, energy, and consumption choices—or on supporting systemic changes through career and giving decisions.

That said, different people might reasonably weigh these considerations differently. If minimising your environmental footprint across all activities is important to you, reducing AI use alongside other small changes might align with your values, even if the impact is minimal.