"We have more AI agents in our firm than humans, which forced us to evaluate our own environmental impact," explains Scott Lane. "While we can't control how major tech giants run their infrastructure, we can control two key choices: who we host with and how we write our prompts."

Because large language models require more processing power the harder they work, phrasing directly determines a query's carbon footprint. In a 2025 technical paper on its Gemini models, Google revealed that a median text prompt consumes 0.24 watt-hours of energy — less than nine seconds of watching TV — emits 0.03g of carbon dioxide and uses five drops of water.

A study titled Green Prompt Engineering for Sustainable Generative AI, from researchers at Boston University, the University of California and the University of Nottingham, found that optimised prompting can reduce LLM inference energy and carbon dioxide emissions by up to 48%, with model-agnostic techniques cutting power consumption across systems including OpenAI's GPT-4o, Cohere and Mistral AI's Mistral 7B.

Beyond individual prompt habits, major providers including IBM, Microsoft, SAP, Salesforce and ServiceNow are embedding sustainability directly into core technology stacks.

"If you're an IT leader, you definitely need a sustainability platform, but it has to be connected to your core tech stack," Scott says. "Sustainability data lives across every part of a company, from HR to finance to logistics. The only sensible way to manage it is to use a module embedded directly within your existing enterprise system."

Having spent over a decade as a lawyer for tech firms and 15 years building global compliance programmes for Fortune 2000 companies, Scott has watched climate data shift from scattered spreadsheets and physical storage rooms to something AI can now automatically consolidate into a clear, actionable picture.

For CEOs overwhelmed by ESG compliance, Scott says the starting point isn't technology at all, but board-level risk assessment. "A board's job is strategy, governance, finance and risk. But the risk landscape has grown massively to include climate, human rights and AI governance."

He pushes back on the instinct to adopt AI for its own sake: "You do not need a use case for AI — you need a problem to solve." Leaders, he says, should anchor every technology investment to a concrete operational bottleneck rather than buying software first and searching for a purpose later.

Key Takeaways

• →Dropping unnecessary pleasantries like "please" and "thank you" from AI prompts is a simple, immediate way to lower energy consumption, says Speeki CEO Scott Lane.
• →A median Gemini text prompt uses 0.24 watt-hours of energy, emits 0.03g of CO2 and consumes five drops of water, per Google's 2025 technical paper.
• →Optimised prompt engineering can cut LLM inference energy and emissions by up to 48%, according to a Boston University, UC and Nottingham study.
• →Major providers like IBM, Microsoft, SAP, Salesforce and ServiceNow are embedding sustainability modules directly into core enterprise tech stacks.
• →Only around 3% of consumers will pay a premium for sustainability alone, meaning green products must also be cheaper and functionally better.
• →Scott Lane advises business leaders to start with a concrete operational problem rather than chasing AI use cases for their own sake.