I'm British. Saying "please" is practically hard-coded into me. But I've stopped saying please to AI.
Not because ChatGPT has been rude to me, and not because I've suddenly forgotten my manners. I've stopped because every unnecessary word we send to an AI has to be processed somewhere.
Now, let's get something out of the way. Is removing the word "please" from your prompts going to save the planet? No. The difference made by a single word is tiny. But it raises a much bigger and more interesting question: if billions of people are going to use AI every day, should we start thinking about how efficiently we're using it?
I think we should.
Key Takeaways
- Data centres consumed around 415 terawatt-hours of electricity in 2024 (1.5% of global electricity), and the International Energy Agency expects that to more than double to around 945 TWh by 2030 (IEA, Energy and AI, 2025).
- A reasoning query can use roughly 13 times more energy than a standard query — research on production AI inference put a typical query at 0.31 Wh versus 3.91 Wh for a much longer reasoning workload (Oviedo et al., Joule, 2026).
- Google cut the energy needed for a median Gemini text prompt by 33x over 12 months (Google Cloud, 2025) — real progress, but efficiency gains tend to increase overall use, not reduce it.
- The biggest lever isn't a shorter prompt. It's a clearer one: fewer regenerations, the right model for the task, and fewer unnecessary AI tools duplicating each other's work.
AI doesn't really live in the cloud
We talk about "the cloud" as though our prompts disappear into some magical, weightless place. Of course, they don't. Behind every ChatGPT conversation, AI assistant, chatbot and AI-generated image sits physical infrastructure: servers, GPUs, networking equipment, cooling systems and increasingly enormous data centres.
And we're going to need a lot more of them.
According to the International Energy Agency's Energy and AI report, data centres consumed around 415 terawatt-hours (TWh) of electricity in 2024, approximately 1.5% of global electricity consumption. By 2030, the IEA expects that figure to more than double to around 945 TWh, slightly more electricity than Japan consumes today. AI is expected to be the most significant driver of that growth.
So while AI may feel virtual, its environmental footprint is very real.
So, should we really stop saying "please"?
Technically, removing an unnecessary word reduces the amount of information an AI system has to process. But let's put that into perspective. Removing "please" from your next ChatGPT prompt isn't going to prevent the construction of another hyperscale data centre.
And that's actually the point.
"Please" isn't really the problem. Waste is.
Multiply tiny inefficiencies across billions of AI interactions and they start to become more interesting. More importantly, there are much bigger efficiencies available to us than simply shaving a word or two from a prompt. The real opportunity is learning how to communicate with AI effectively in the first place.
Consider a very simple example. Someone asks an AI:
"Write me a proposal."
The AI produces 2,000 words. It's not quite right, so they ask it to make it shorter. Another response is generated. Then they ask it to make the proposal more professional. Another response. Then more conversational. Another response. Finally, they realise they forgot to ask for pricing and next steps, so off it goes again.
Compare that with starting with a clearer instruction:
"Write a 500-word client proposal. Use UK English and a professional but conversational tone. Structure it around the client's problem, our proposed solution, key benefits, pricing and next steps."
The second prompt actually contains more words, but there's a much better chance it produces what you need first time. That means fewer regenerations, fewer unnecessary output tokens and less computation overall.
This leads to an important point:
The greenest prompt isn't necessarily the shortest prompt. It's the prompt that gets the job done with the least wasted computation.
Welcome to "prompt pollution"
We've spent decades learning to think about physical waste. We turn lights off when we leave a room, use energy-efficient appliances, recycle, reduce unnecessary journeys and try not to leave the tap running.
But we're only just beginning to think seriously about computational waste.
As AI becomes embedded into almost everything we do, that matters. I call it prompt pollution, the unnecessary computation created when we use AI inefficiently.
Poor instructions are one source of prompt pollution, but they're certainly not the only one. Repeatedly regenerating answers, requesting unnecessarily long outputs, sending huge amounts of irrelevant context and using powerful reasoning models for trivial tasks can all increase the amount of computation required to achieve an outcome.
Then there's the business equivalent: using five different AI tools when one or two could do the job, duplicating data between systems, or automating a broken process instead of fixing the process first.
Individually, these decisions may seem insignificant. At global scale, they become much more interesting.
Not all AI requests are created equal
There's another important part of this story. Different AI tasks can require dramatically different amounts of computation.
A 2026 study of production AI inference published in Joule found that a standard frontier-model query used a median of around 0.31 Wh, while a much longer reasoning workload — roughly 15 times the length of a typical query — pushed that up to 3.91 Wh, around 13 times more energy (Oviedo et al., Joule, 2026).
That doesn't mean we shouldn't use powerful reasoning models. They are extraordinarily useful when the task requires them. It means we should think about whether we actually need that level of computational power for the job we're doing — the same thinking we'd apply to choosing the right AI model for a task instead of always reaching for the newest, most powerful one.
Do you need the most powerful model available to rewrite three sentences? Do you need deep reasoning to summarise an email? Do you need a 2,000-word answer when five bullet points would give you everything you need?
Similarly, if one paragraph of a document needs changing, asking an AI to regenerate the entire document is usually unnecessary.
As AI becomes more capable, good AI literacy won't simply mean knowing how to use AI. Increasingly, it will mean knowing how much AI to use.
The good news: AI itself is getting more efficient
The technology industry is making enormous improvements in efficiency. Google, for example, reported a 33-fold drop in energy per median Gemini text prompt over a 12-month period through improvements across models, software and hardware.
That's extraordinary progress, but efficiency improvements create another challenge. When technology becomes cheaper, faster and easier to use, we tend to use more of it.
AI is already moving from something we occasionally visit in a browser to something embedded throughout our businesses. AI assistants are appearing in customer service, sales, marketing, software development, data analysis, document processing, finance and scheduling. Voice agents can answer telephone calls, AI can analyse company data, and automated workflows can move information between systems without human intervention.
The question therefore isn't whether businesses will use more AI. They almost certainly will.
The more important question is whether we'll use it intelligently.
Better AI doesn't always mean more AI
This is something we see regularly at AI Smarter. Businesses often assume that becoming "AI-powered" means buying another AI platform. It doesn't.
In many organisations, the biggest opportunity is already sitting inside the technology they're paying for. Their CRM, Microsoft 365 or Google Workspace, accounting software, customer support platform, project management system, marketing tools, ChatGPT and existing databases may already contain much of what they need.
The problem is often that these systems aren't being used to their full potential. Workflows contain unnecessary manual steps, information gets copied from one system to another, employees perform repetitive tasks that could be automated, and teams haven't necessarily been shown how to get the best from the AI capabilities already available to them. It's the same logic behind our own AIOS approach: get the model, data, and systems you already have working together before adding another one.
Adding another tool doesn't automatically solve that problem. Sometimes it simply adds another subscription and another system for people to manage.
That's why we believe businesses should start by asking a different question. Instead of "What AI should we buy?", ask "How can we use AI and the technology we already have more effectively?"
Efficiency is good for the planet, and good for business
There's a tendency to think about sustainability and business efficiency as separate things. With AI, they're increasingly connected.
A better prompt can mean fewer generations. A better workflow can eliminate unnecessary manual steps. Choosing the appropriate model can reduce unnecessary computation. Connecting existing systems can remove duplicate work and potentially remove the need for another platform altogether. We see this play out across our own case studies — the businesses getting the most from AI are rarely the ones with the most tools.
The environmental argument matters, but there's an even more immediate benefit for businesses: people get time back.
That's ultimately what good AI implementation should achieve. It shouldn't be about using AI because it's fashionable, putting "AI-powered" on every product, or replacing every existing system with the latest shiny AI tool.
It should be about using technology intelligently to achieve better outcomes with less effort.
So, should you stop saying please to ChatGPT?
I'm going to, although being British, I admit it still feels slightly uncomfortable.
Will that single change save the planet? Of course not. But it serves as a useful reminder that AI isn't free. Every prompt requires computation, every unnecessary generation requires computation, and every inefficient workflow consumes resources somewhere.
As billions of people begin using AI every day, learning to use it efficiently is going to matter.
The future doesn't just need more powerful AI. It needs smarter use of AI.
And that's exactly where we can help.
Get more from the technology you already have
At AI Smarter, we help businesses identify where AI and automation can genuinely improve the way they work. That doesn't automatically mean adding more software.
We can examine the tools and processes you already use, identify unnecessary manual work and inefficient workflows, help your teams use AI more effectively, connect systems and automate repetitive processes.
That might involve better prompting and practical AI training. It could mean connecting systems that currently require people to manually move information between them. It might involve intelligent workflow automation, an AI assistant, a voice agent or a custom solution built around the systems you already have.
The technology comes second. The business outcome comes first.
The objective is simple: get more from the technology you're already paying for, reduce wasted effort and give your people more time to focus on the work that actually matters.
Want to find out where AI could make your business more efficient?
Book a free 30-minute discovery call with AI Smarter and let's look at the tools, processes and workflows you're already using, then identify where AI could make the biggest difference.