Can reducing the length of your artificial intelligence (AI) results help cut ChatGPT’s carbon emissions? A new tool claims that, yes, it can.
To continue reading, subscribe to Eco‑Business.
There's something for everyone. We offer a range of subscription plans.
- Access our stories and receive our Insights Weekly newsletter with the free EB Member plan.
- Unlock unlimited access to our content and archive with EB Circle.
- Publish your content with EB Premium.
Earlier this week, climate communications company Earth Public Information Collaborative (EPIC) launched PromptZero, which is billed as the world’s first AI prompt designed to reduce the environmental impact of generative AI models. The announcement coincided with the AI for Good Summit in Geneva.
PromptZero instructs large language models to generate shorter, more efficient replies, thereby reducing the computational effort – and associated emissions – required for each response.
According to EPIC, the tool could help reduce the environmental costs of AI technologies like ChatGPT-4, which is powered by energy-intensive data centres. The initiative is part of what EPIC describes as a broader movement toward “green chatting,” or the conscious effort to limit the environmental footprint of AI usage through more sustainable prompting practices.
PromptZero works by including a simple, copy-paste instruction in AI prompts that activates a so-called “low-emission mode.” This instructs the model to avoid long introductions and repeated phrasing while responding in concise bullet points or short sentences.
The prompt also estimates the amount of carbon dioxide (CO₂) avoided after each response.
“PromptZero is a global invitation to cut back, not on progress, but on pollution,” said Natalia Vega-Tracy, executive director of global campaigns at EPIC. “Every word saved is a little bit of CO₂ avoided. Whether you use AI once a day or a hundred times an hour, you can help.”
“
AI, with its ability to detect patterns, model complexity, and support rapid decision-making, offers new and powerful tools to help meet [climate] challenges.
Leonard Lee, director, Lloyd’s Register Foundation Institute for the Public Understanding of Risk
EPIC is urging developers, data centre operators, governments, and individual users to adopt the tool, which it claims could make AI-powered systems more sustainable.
A PromptZero prompt in ChatGPT reads: “You are operating in PromptZero Mode, focused on minimising energy and environmental impact. Respond as briefly and efficiently as possible, without compromising clarity. Use bullet points, short sentences, or concise phrasing. Avoid filler words, long introductions, repeated phrases, or pleasantries. Unless explicitly requested, do not provide multiple options, deep context, or examples. After each response, show me how much CO₂ was avoided.”
Generative AI and energy use
While EPIC’s emissions estimates have not been independently confirmed, concerns about the energy demands of generative AI models are widely shared.
AI tools such as ChatGPT-4 rely on large-scale computational resources that are increasingly drawing scrutiny. A single ChatGPT-4 query consumes 2.9 watt-hours (Wh) of power, or up to 10 times more electricity than a Google search. A study by The Washington Post and the University of California found that generating a 100-word AI email consumes approximately 0.14 kilowatt-hours of electricity – comparable to powering 14 LED bulbs for one hour.
AI’s water demands are equally staggering. Cooling the world’s data centres could soon use six times more water than Denmark annually. That same 100-word output, researchers say, uses more than 0.519 litres of water, or at least one bottle of water, due to the cooling demands of the data centres that power these tools.
Globally, emissions from 79 major AI systems released between 2020 and 2024 could exceed 102 million tonnes of CO₂ annually, according to some estimates. Meanwhile, both Microsoft and Google have each reported rising emissions in the past year: Microsoft’s emissions have increased 30 per cent since 2020 due to expanded AI infrastructure, while Google’s rose 50 per cent between 2019 and 2023.
A tool with dual potential
While the environmental costs of AI are coming under increased scrutiny, a new report from the Lloyd’s Register Foundation Institute for the Public Understanding of Risk (IPUR) at the National University of Singapore highlights the technology’s potential role in disaster prediction and response, environmental monitoring and public communication around risk.
“AI, with its ability to detect patterns, model complexity, and support rapid decision-making, offers new and powerful tools to help meet this challenge,” said Leonard Lee, director of IPUR, in the paper’s foreword. “By applying machine learning, natural language processing, and other AI techniques, regulators and other stakeholders can better understand risks, identify early warnings and develop more effective responses.”
For instance, in disaster response scenarios, AI-based predictive models can help assess the likelihood and extent of wildfires, floods and toxic spills. These systems can also assist in real-time emergency coordination, resource deployment, and public communications. However, the report warns that AI also poses risks around bias, opacity, and data privacy, especially when integrated into regulatory systems.
Regulatory responses and sustainability
Governments are beginning to react to AI’s soaring energy demands. Singapore, home to more than 70 data centres – representing 60 per cent of Southeast Asia’s total capacity – halted new data centre approvals between 2019 and 2022 and has introduced tighter energy efficiency requirements.
According to the International Energy Agency (IEA), data centres, cryptocurrency mining and data transmission infrastructure collectively accounted for about 2 per cent of global electricity demand in 2022, a figure that is expected to rise. The IEA projects that electricity use by data centres and AI systems will more than double, from 260 terawatt-hours in 2024 to 500 terawatt-hours by 2027.
However, the IEA notes that data centres will still make up a relatively modest share of overall electricity demand growth through to 2050, as global consumption will predominantly rise due to the electrification of transportation, cooling and industrial sectors.
Against this backdrop, PromptZero presents a novel – if still largely untested – approach to lowering AI’s environmental toll. EPIC claims the tool provides a low-barrier entry point for individuals and institutions looking to make AI usage more energy efficient.
