How to Use AI More Sustainably: 4 Simple Ways to Cut Energy, Water, and Carbon

1. Ask Yourself If You Actually Need AI for This

The simplest energy-saving move is also the most overlooked: don’t use AI when a simpler tool will do.

Checking the weather, finding a specific article, or looking up a quick fact doesn’t require a large language model. As one AI researcher put it, using ChatGPT for something like “What should I wear today?” is like taking a supersonic jet to the corner store.

Many search engines now layer AI-generated summaries on top of standard results by default. If you’re just hunting for a link or a quick answer, turning that off matters. In Google or Bing, you can often select “Web results only” or add -ai to your search query to skip the AI-generated response entirely.

This won’t feel dramatic, but at the scale of billions of daily queries, it adds up.

2. Use Smaller, Specialized Models When You Can

Not every task needs a trillion-parameter model. This is one of the highest-leverage changes you can make if you use AI regularly.

Smaller language models trained for specific tasks — translation, summarization, classification — consume dramatically less energy than general-purpose LLMs tackling the same jobs. Research published by UNESCO found that switching from a large general model to smaller specialized alternatives reduced energy use by up to 90 percent, while actually producing higher-quality outputs on the tasks those models were built for.

The tradeoff is convenience. Smaller models are often less polished and require more setup. But if you’re running the same task repeatedly — translating documents, summarizing reports, answering structured queries — it’s worth exploring purpose-built alternatives available through platforms like Hugging Face.

For everyday users who stick to mainstream tools, the practical version of this tip is simpler: don’t default to the most powerful model available when a lighter option is offered. Many AI platforms now let you choose between model sizes.

3. Keep Outputs Short — This Is the Big One

Here’s something most people don’t know: the length of the AI’s response drives energy use far more than the length of your prompt.

Because large language models run billions of calculations for every single word they generate, a longer output means significantly more compute. Research found that cutting a model’s output in half reduced its energy consumption by around 50 percent. Cutting the input prompt in half? Only about 5 percent savings.

The practical implication is straightforward:

• Tell your AI tool to “be brief” or set a word limit
• Ask for bullet points instead of paragraphs when that works for your use case
• Use reasoning or “problem-solving” modes only for genuinely complex tasks — these modes can produce up to ten times more words and consume significantly more energy than standard conversation modes
• Be specific in your prompts so the model doesn’t pad its response with unnecessary context

Instructions like “five bullets max” or “keep it under 100 words” aren’t just good prompt hygiene — they’re a meaningful way to reduce the compute behind every response.

4. Go Low-Res First and Batch Your Image Requests

Generating images and video is orders of magnitude more energy-intensive than generating text. Every pixel gets processed repeatedly across many iterations, making visual generation one of the heaviest workloads in consumer AI.

A few habits can reduce that footprint significantly:

• Start at low resolution. If you’re exploring a concept or testing a direction, generate at lower resolution first. Only scale up once you know the output is heading the right way.
• Edit rather than regenerate. Modifying an existing image is far less computationally intensive than generating a new one from scratch. Use inpainting or editing tools when possible.
• Batch your requests. If you need multiple images, generate them in a single session rather than across separate requests. Batching is more efficient than repeated individual generations.

These aren’t huge sacrifices. They’re small workflow adjustments that compound over time, especially for teams or creators who generate visual content regularly.

The Honest Bottom Line

None of these tips will single-handedly fix AI’s environmental footprint. The structural changes — more efficient hardware, cleaner energy sources, smarter algorithms — have to come from the companies building and running these systems.

But the researchers studying this are consistent on one point: individual choices aren’t irrelevant, especially as AI use scales into billions of daily interactions. Knowing which lever actually moves the needle — output length, model size, task appropriateness — puts you ahead of most users.

Use the right tool for the job. Keep outputs tight. Start low-res. And skip the AI overview when a plain search result will do.