How to Use AI With Less Waste

You do not have to stop using AI to use it responsibly. Most of the waste comes from a few habits: using a heavy tool for a light job, asking the same question five times, or generating images and video you do not really need.

This lesson gives you simple habits that cut waste, save you time, and usually get you better answers too.

What You'll Learn

• Which AI tasks use the most energy, and which use the least
• How to pick the right size of model for a job
• How better prompts mean fewer retries
• When running a model on your own device makes sense
• How to keep your personal AI use in perspective

Start with perspective

From lesson 1, a short text prompt uses roughly 0.2 to 0.4 Wh. For one person, a few dozen of those a day is still less energy than boiling a single kettle of water.

So for most people, short text prompts are not the main part of their energy footprint. Heating, cooling, transport and food usually matter far more.

But two things are still worth your attention:

1. Heavy tasks like video, many images and long reasoning runs can use far more than a text reply.
2. Habits scale. If millions of people each waste a little, the total is large. The same good habits also save you time.

Habit 1: Match the tool to the task

Most AI apps offer a choice of models or modes. Bigger and "thinking" options are slower and use more computing for each answer. They are worth it for hard problems, but not for simple ones.

A simple rule: start light, and move up only if the answer is not good enough.

Habit 2: Write a better first prompt

Every regenerate button press runs the model again. Many retries happen because the first prompt was vague.

A short, clear prompt usually includes:

• The task: what you want ("Summarize this article").
• The audience or purpose: who it is for ("for a first-year student").
• The format: how you want it ("in five bullet points").
• Any limits: length, tone or what to leave out.

Compare these two:

• Vague: "Tell me about photosynthesis."
• Clear: "Explain photosynthesis to a 15-year-old in 5 short bullet points, with one everyday example."

The second prompt is far more likely to work the first time. If an answer is close but not right, edit or follow up ("make point 3 shorter") instead of starting again from scratch.

Habit 3: Keep inputs short

Lesson 1 showed that long inputs make the model do more work. You can help by:

• Pasting only the part you need, not a whole document, when you have a question about one section.
• Starting a new chat for a new topic, so the model does not reread a long, unrelated history.
• Asking for short answers when you only need a short answer ("Answer in two sentences").

This also tends to give you more focused answers.

Habit 4: Be thoughtful with images and video

Generated images and especially video use much more energy than text. That does not mean never use them. It means:

• Plan before you generate. Decide what you want so you need fewer attempts.
• Describe it clearly. Subject, style, setting and framing in one prompt.
• Avoid generating many versions for fun when one would do.
• Use a real photo or a free image when it fits your purpose just as well.

Habit 5: Consider a local model

You can run some smaller AI models directly on your laptop or phone. This is called a local model.

Benefits:

• No data center needed for each prompt, so no data center cooling water.
• Privacy, because your text stays on your device.
• Works offline.

Trade-offs:

• Your device still uses electricity, and how clean it is depends on your local grid.
• Small local models are usually less capable than the large cloud models.
• Big local models need powerful, expensive hardware.

Local models are a good choice for simple, private or frequent tasks. To try one, see Run AI on Your Own Computer.

Habit 6: Turn off what you do not use

Some apps run AI in the background, for example automatic summaries or AI features in search and email. If you never read them, check the settings. Turning off features you do not use saves a little computing, and it can make the app simpler too.

A quick checklist

Before you send a request, ask yourself:

1. Do I need AI for this, or would a quick search do?
2. Is this the lightest model or mode that will do the job?
3. Is my prompt clear enough to work the first time?
4. Am I sending only the text the model needs?
5. Do I really need an image or video, or would text do?

Key Takeaways

• For most people, short text prompts are a small part of their energy footprint, but heavy tasks and habits at scale matter.
• Match the tool to the task. Start with a light model and move up only if needed.
• Clear first prompts and follow-up edits mean fewer retries.
• Short inputs and fresh chats reduce work for the model and give better answers.
• Local models avoid data center use for simple tasks, with trade-offs in quality and hardware.

Sources

• Google, Measuring the environmental impact of delivering AI at Google scale (2025)
• Epoch AI, How much energy does ChatGPT use? (2025)
• Luccioni, Jernite and Strubell, Power Hungry Processing (2023)