Balancing AI innovation with environmental responsibility: a B2B marketer’s guide
B2B marketers have ended up at the front of the AI queue: handed the tools first, expected to find the use cases, and quietly judged on how quickly we adopt whatever’s just launched. Generative AI now sits in most events, talks, and software roadmaps, promising more productivity and fewer blank-page moments. Some of that is true, which is what makes the conversation harder, because AI is genuinely useful for research, drafting, analysis, and a lot of the repetitive work that quietly eats everyone’s time. But the bit we haven’t properly dealt with is the environmental cost of using it so casually, especially when the interface makes the whole thing feel weightless: type something in, get something back, ask for three more versions, then move on without thinking about what happened behind the prompt box.
None of it is weightless. It’s hard to put one neat number against the footprint of an individual interaction, but the overall direction is clear. The International Energy Agency’s updated projections suggest global data-centre electricity consumption could rise from around 485 TWh in 2025 to approximately 950 TWh by 2030, with electricity use from AI-focused data centres expected to triple over the same period, quite a lot of physical infrastructure for something we still casually call ‘the cloud’.
That doesn’t mean the answer is to stop using AI, but sustainability needs to stop being somebody else’s responsibility and start appearing in the decisions marketing teams make about tools, workflows, and output. The useful question isn’t simply whether AI can do something; it’s whether using AI is proportionate to the value of the task, and whether we’re using it efficiently once we’ve decided that it is.
The footprint is bigger than the prompt box makes it look
One of the more useful things about the BrightonSEO session, ‘The Sustainability Issue with AI and What We Can Change as Marketers’, was that it made the resource chain behind an AI interaction feel less abstract. What looks like one neat request and response can involve model inference, storage, networking, cooling, and several additional calls to other systems before the answer appears, which is more useful to picture than pretending the issue belongs solely to infrastructure teams. That matters because AI is no longer limited to the moments when somebody opens ChatGPT or Claude; it’s increasingly built into search engines, office software, and CRMs, so the baseline level of consumption can rise even when nobody has actively decided to ‘do more AI’.
The IEA’s 2026 update captures the tension neatly: the electricity required for individual AI tasks is falling quickly as systems become more efficient, while total data-centre consumption is still expected to roughly double by 2030 as more people use AI and heavier uses such as AI agents become more common. The technology can become more efficient and still consume much more energy overall, which is, unfortunately, how scale works.
Electricity is only part of the picture, because the infrastructure also has consequences for water, land, carbon emissions, and electronic waste. A United Nations University report published in June 2026 argued the environmental impact needs to be understood across the wider system. AI’s footprint isn’t limited to the electricity used during a single prompt, and improving one part of the system doesn’t automatically remove the pressure from the others, which is why claims about ‘green AI’ need more interrogation than a renewable-energy badge in the footer. Sustainable Energy for All makes a similar point: AI can become more efficient while its total energy demand keeps rising, because efficiency gains are being overtaken by the pace of adoption, putting AI’s energy demand on track to triple by 2030.
The quality problem is part of the sustainability problem
The environmental impact is the obvious part of the conversation, but the BrightonSEO session also made a point particularly relevant to marketing: low-value AI output isn’t only wasteful from an energy perspective; it can also weaken the quality of the work and eventually the credibility of the brand. Generating five drafts, ten image routes, and a few ‘quick’ rewrites can look efficient while the machine is doing it, but somebody still has to review, compare, and approve everything that comes back. What started as a shortcut can quietly create more work, more decision fatigue, and more average material.
The better question isn’t ‘Can AI do this?’ It’s ‘Should AI do this?’ Could a smaller model handle the task, or an existing template get us there faster? A vague prompt usually means more generations and more avoidable compute, but it also tends to produce flatter work. If the team is clear about the objective, audience, and success criteria before opening the tool, there’s a better chance of getting something useful earlier, and the output is less likely to sound like it was written for every company in every market at once.
Responsible AI is really a workflow choice
None of this is an argument for marketers to stop using AI. The concern is treating it like a free vending machine for half-formed ideas: put in a loose request, ask for more options, regenerate because the first answer feels generic, then spend 20 minutes working out why the ‘more human’ version sounds like it runs a personal-brand newsletter.
Responsible AI is less about a grand statement and more about better workflow decisions, because the practical difference comes from what people do before, during, and after they use the tool. What are we trying to achieve? Which tool is proportionate to the task? What requires human judgement before anything goes live?
It’s worth looking at how the better AI ecosystems are actually built. Velo’s own Olev Intelligence platform is a useful example: it runs on a Think, Plan, Do structure, using AI to speed up research and ideation, ground plans in data and proven frameworks, then create and activate campaigns faster, with humans involved at the start, middle, and end of every stage rather than just skimming the final output. The point isn’t to replace judgement with generation; it’s to build proportionate AI use and human review into the workflow by design. For marketers, the practical responsibility is to ask whether suppliers are making those same choices, and whether simple jobs are automatically routed through systems much larger than they need to be.
More AI content isn’t automatically more useful content
The BrightonSEO discussion also touched on a ‘dead internet’ dynamic, where automatically generated articles and recycled ideas crowd out original research and content created because somebody had something useful to say. That sounds dramatic until you open three search results and realise they’ve all used the same introduction and the same cheerful conclusion about embracing the future.
For marketers, this isn’t only an internet-quality problem; it’s a brand problem. If organisations use similar models fed with similar prompts, the work converges: more articles that are technically competent, more social posts that are perfectly acceptable, and fewer reasons for anybody to remember who produced them. The answer is to use AI around the work rather than letting it replace the things that make it credible: original research, customer insight, expert judgement, and the awkward details that show somebody actually understands the problem.
What marketers can change now
Sustainable AI doesn’t require abandoning the technology; it means using it with more intent and being honest about where it adds enough value to justify the resources involved. A few practical habits would make a reasonable start:
- Choose the tool in proportion to the task. Some jobs will be handled just as well by a smaller model, a template, or no AI at all.
- Write the brief before opening the tool. Define the audience, objective, and outcome first, rather than discovering all of that through repeated iterations.
- Limit the number of variants. Decide how many routes the team genuinely needs, then select and refine one.
- Do the first bit of thinking manually. Sketch the structure or agree the main point before prompting.
- Batch related work where it genuinely helps, though batching 80 mediocre posts doesn’t become sustainable simply because it was organised.
- Build human review into the workflow. Check facts, tone, and whether the output is actually useful before it reaches a client.
- Ask suppliers about the infrastructure. Look for disclosure on energy use and model efficiency, rather than accepting ‘committed to sustainability’ as a complete answer.
- Keep the useful work, not every output. There’s little value in keeping 40 abandoned versions in case somebody needs ‘Option 17’ one day.
Governance needs to survive contact with actual work
A useful policy needs enough structure to guide decisions without requiring somebody to consult a 42-page document every time they want help summarising meeting notes. The starting point is understanding what’s already happening, because AI use often spreads through organisations one subscription or browser extension at a time. An audit should cover the tools being used, the tasks they’re supporting, and whether the result is saving time or simply moving the work from creation into reviewing.
From there, the business can define where AI is appropriate, what requires human approval, what data must never be entered, and which outputs need checking before they’re used externally, connected to security, procurement, legal, and client commitments rather than living in a shared-drive folder.
Measurement will be imperfect, especially while providers disclose so little, but that doesn’t make it pointless. We may not be able to calculate the precise footprint of every interaction, but we can identify wasteful behaviour and reduce it, which matters more as organisations take on public ESG commitments that large-scale, casual AI use can sit awkwardly alongside. The same transparency matters when AI-supported services are being sold: where AI is part of the offer, the workflow, safeguards, model choice, and approach to limiting avoidable output should all be clear, which is more credible than adding ‘AI-powered’ to a service description and hoping nobody asks what ‘powered’ actually means.
The final piece is protecting the thinking people are actually paying for. AI can help organise information and remove repetitive work, but it shouldn’t flatten the strategy or produce more generic material for brands trying to sound distinctive. The aim is to use AI less casually, not simply less often. What needs to change is the assumption that AI use is effectively free because the marginal cost isn’t visible to the person typing the request, while low-value use creates a second kind of waste through unnecessary drafts and generic content.
So the practical response is less about sweeping declarations and more about choosing proportionate tools, writing clearer briefs, protecting human judgement, and asking suppliers better questions. Sustainability isn’t the opposite of innovation; it’s part of deciding whether the innovation is useful, credible, and capable of scaling without creating costs everyone has quietly agreed not to look at yet