Skip the 'Please': How Shorter AI Prompts Could Save Enough Energy to Power 760,000 Homes

UN Report Links Prompt Length to AI Energy Consumption

A comprehensive United Nations report has exposed a surprising connection between how people write AI prompts and the environmental impact of artificial intelligence. Researchers from the United Nations University Institute for Water, Environment and Health (UNU-INWEH) found that eliminating unnecessary words like "please" and "thank you" from interactions with ChatGPT and similar platforms could reduce energy consumption by up to 25 percent 1

1

. The study estimates that adopting concise AI prompts across typical usage patterns could save between 87 and 98 gigawatt-hours of electricity annually—enough to meet the residential power needs of up to 760,000 people in sub-Saharan Africa 1

1

.

How Large Language Models Process Your Words

The energy use of AI chatbots stems from how large language models process text. These systems break down prompts into small units called tokens, and every additional word increases both the computational load required to understand the request and the energy needed to generate responses 1

1

. Kaveh Madani at UNU-INWEH explains that shorter prompts may also simplify tasks, further reducing power requirements. "We are not saying be rude to your AI. But don't fall into the interaction trap and don't go falling in love with it either," Madani noted 1

1

. The report warns against extended conversational loops with chatbots, as continuous back-and-forth exchanges multiply total compute usage through repeated model inference cycles 3

3

.

Scale of Daily Usage Amplifies Environmental Concerns

The scale of AI adoption makes even small inefficiencies significant. ChatGPT alone now processes approximately 2.5 billion queries every day, while Google handles 16 billion searches, most incorporating AI-generated summaries 1

1

. At a conservative estimate of 0.42 Wh per prompt, ChatGPT's daily operations consume roughly 383 GWh of electricity annually—sufficient to power nearly 3 million people in sub-Saharan Africa 2

2

. The environmental concern of AI usage extends beyond electricity to water consumption, with data centers projected to require 9.3 trillion liters of water by 2030—enough to meet minimum annual domestic water needs for all 1.3 billion people in sub-Saharan Africa 1

1

.

Data Centers Face Mounting Energy Demands

AI currently accounts for approximately 20 percent of energy consumed by data centers, but this share is projected to double to around 40 percent within the next few years 1

1

. By 2030, AI energy consumption alone could reach 378 terawatt-hours annually, while data centers collectively could use 945 TWh—nearly 3 percent of projected global electricity use 1

1

. Miriam Aczel at UNU-INWEH emphasizes the unprecedented pace: "You're looking at something on a global scale that is being adopted faster than any other technology in the history of technology, so the energy use is increasing very rapidly" 1

1

.

Image and Video Generation Multiply Energy Costs

The type of AI task selected dramatically affects power consumption. Generating a single AI image requires 2.9 Wh—60 times more energy than a standard text query and enough to power a 10-watt LED bulb for approximately 17 minutes 1

1

2

2

. Complex video generation consumes over 415 Wh per clip—up to 8,000 times more than text queries and sufficient to power the same bulb for about 1.7 days 1

1

2

2

. A typical ChatGPT-style text query uses about 200 times more energy than basic spam filtering 2

2

.

Practical Steps to Reduce AI Energy Footprint

Researchers recommend several strategies to save energy with AI prompts. Users should write direct, concise instructions without unnecessary pleasantries, avoid extended conversational exchanges, and select less powerful models when tasks don't require advanced capabilities 1

1

2

2

. The study also calls for systemic changes, including mandatory disclosure of AI energy consumption by tech companies, government-imposed energy caps, and public education on efficient AI usage 1

1

. Madani stresses balanced usage: "We are not saying AI is bad. We are just saying let's use it in a proper way. It's like a knife: you can save a patient's life in the operating theatre, but you can also kill someone with it" 1

1

. The key issue isn't AI usage itself but the inefficient AI interaction patterns that create unnecessary computational load 3

3

.