Ditch the Niceties in AI Prompts to Save Energy Use, Say Researchers

Why It Matters

Even modest prompt trimming offers a low‑cost lever to curb AI’s rapidly expanding carbon footprint, a critical step as AI workloads threaten to dominate global data‑center power consumption.

Key Takeaways

• Removing “please” can cut ChatGPT energy use 25%
• Concise prompts could save 87‑98 GWh annually, powering 760k homes
• AI already consumes 20% of data‑center electricity, may reach 40%
• Image generation uses 60× more energy than text queries
• Researchers urge energy caps and transparent reporting for AI firms

Pulse Analysis

The United Nations University Institute for Water, Environment and Health released a landmark assessment that quantifies AI’s hidden environmental toll. By dissecting token‑level processing, the researchers demonstrated that every unnecessary word forces large‑language models to handle extra tokens, inflating power draw. In the case of ChatGPT, trimming polite filler could reduce energy consumption by a quarter, translating into 87‑98 GWh saved each year—an amount comparable to the residential electricity needs of hundreds of thousands of households in sub‑Saharan Africa. This insight reframes a seemingly trivial user habit into a measurable sustainability lever.

Beyond individual prompts, the report paints a broader picture of AI’s escalating demand on data‑center resources. Today, AI workloads represent about 20% of data‑center electricity use, a figure projected to climb to 40% within a few years as generative models proliferate across enterprises. Coupled with a forecasted 9.3 trillion litres of water required by 2030, the sector’s footprint rivals that of entire regions. The authors therefore advocate for regulatory interventions: mandatory energy‑consumption disclosures, caps on both corporate and personal AI usage, and public education campaigns that promote efficient interaction patterns.

For practitioners, the path forward is straightforward. Users should craft prompts that are concise, avoid unnecessary back‑and‑forth, and select smaller models when possible. Companies can embed energy‑aware defaults into APIs, provide real‑time consumption metrics, and prioritize model architectures that balance performance with power efficiency. By treating AI as a tool—like a knife that can heal or harm—organizations can harness its benefits while mitigating its environmental impact, turning a cultural shift in prompt etiquette into a tangible climate‑action strategy.