Seb Royce is creative and communications partner at Eden Lab, a green growth consultancy and venture studio. Having spent years as a creative director and agency founder, Seb shifted his focus toward sustainability after what he describes as a “road to Damascus moment” about the environmental and societal weight of his work footprint. He’s now one of the more thoughtful voices at the intersection of marketing, creativity, and environmental impact.
In this session, Seb covered the real environmental cost of AI adoption, why that doesn’t mean you should stop using it, and the practical steps any marketer can take to use it more intentionally.
Why AI’s environmental cost is worth talking about
Every prompt, generated image, or AI analysis requires energy-intensive processing on remote servers. Seb shared a striking frame for the scale of it: if global data centers were a country, they’d sit somewhere between Japan and Russia in terms of energy consumption, and that’s before accounting for the projected growth in AI adoption.
Water is part of the story too, used to cool the systems running everything. And then there are the raw materials behind the hardware itself.
But Seb is clear that this isn’t a reason to stop. “One of us writing a brief on Claude isn’t really the issue,” he said. “It’s the cumulative effect.” A team generating tens of thousands of AI images a week, or using large frontier models for tasks that don’t need them, starts to look very different in terms of footprint.
Right-sizing your AI models
One of Seb’s most memorable points: using a large language model for a simple task is a bit like driving a Formula 1 car to the shops. It’s powerful, but you’re using far more than you need.
There are smaller, lighter models available, often as subsets of the bigger ones, that handle simpler tasks well without the same energy overhead. If you’re inside a larger organisation, this is worth raising as a conversation at policy level, because the cumulative impact across a team can be significant.
Seb’s own current stack includes Claude (for strategy, briefs, and scripts, partly because of Anthropic’s focus on efficiency), Adobe Firefly over DALL-E 3 (for its science-based climate targets and renewable energy commitments), and DaVinci Resolve over CapCut, since it runs locally rather than in the cloud.
Better prompts are a sustainability intervention
This one landed well in the room, and it’s probably the most immediately actionable thing here.
Every time you hit regenerate on a vague prompt, Seb points out, you’re spending carbon as well as time. Most of us are still experimenting with prompting, going in underprepared, getting it wrong, and starting over. Taking 30 seconds to prep a prompt before sending it, being specific about the output format, tone, and context, and giving AI the “why” behind the task all make a real difference to how quickly you get something usable.
Seb also flagged something people often overlook: LLMs are conversational. Rather than scrapping a prompt and starting fresh, you can iterate within the same thread. That takes less compute, and often gets you somewhere better faster.
“Better prompts are really a sustainability intervention,” he said. “It’s probably the easiest thing all of us can do.”
Thinking about your AI supply chain
This is the more advanced consideration, but worth being aware of. Most people think their AI relationship is with the app on their screen, whether that’s Claude, ChatGPT, or Jasper. But there’s a three-layer supply chain behind it: the app you use, the foundational model powering it, and the infrastructure (like AWS) running the compute.
As an individual, you can’t change what’s happening at the infrastructure level. But being aware of it, asking questions about it, and choosing tools whose sustainability commitments you can actually read about is a reasonable place to start.
Tracking progress rather than perfection
The honest answer to “how do I measure my AI footprint?” is that the standardised tools to do that properly don’t exist yet. But Seb has a practical workaround: track the number of prompts it takes you to complete a piece of work, and see if that number goes down over time.
He gave the example of recutting a Guinness ad with new environmental footage. The first time, it took a certain number of prompts. A recent film took half as many. That’s progress, and progress is measurable even when specific energy stats aren’t.
“What you can track is progress,” Seb said. “And if you’re getting better at prompting, getting more efficient, that is a good thing.”
You don’t have to have all the answers
There’s a strand running through this whole conversation which feels worth naming directly: you don’t need to be an expert, you don’t need to have the perfect setup, and you definitely shouldn’t feel guilty about using AI.
What Seb is really describing is a mindset shift rather than a checklist. Be curious. Ask questions. Check in on whether the tools you’re using align with your values. And if they don’t, know that you can switch things out relatively quickly as the landscape keeps changing.
“Just do what you can,” as someone told Seb during his sustainability training. It took the pressure off, and it’s probably the right frame here too.
The marketers who will navigate this well aren’t the ones who abandon AI or the ones who use it without thinking. They’re the ones who stay curious, stay informed, and treat intentionality as an ongoing practice rather than a one-time audit.
Want to catch the next one live?