Telling the AI Story Straight

AI helps solve environmental problems and creates new ones. Sustainability communicators should describe both with the same precision they demand of climate claims.

Artificial intelligence has become one of sustainability's most useful tools and one of its hardest communication problems. It sharpens weather forecasts and finds methane leaks from space. The data centres behind it draw more electricity and water every year.

Too often, the debate splits into two competing narratives. Technology companies emphasise breakthroughs; critics highlight environmental costs. Both accounts are partly true, and both leave audiences less informed than they should be.

Sustainability communicators have spent a decade learning to back claims with evidence, state uncertainty and resist greenwashing. AI needs the same discipline. The evidence is growing, although significant gaps remain. How organisations interpret and communicate it matters as much as the data itself.

AI as a tool

The strongest cases share three features: a defined problem, a measurable result and people checking the output.

Methane monitoring is one. The UN Environment Programme's Methane Alert and Response System combines data from more than 30 satellite instruments. AI separates real methane plumes from background noise, and analysts verify every detection before a government or company is notified. They can now process 12 to 15 times more data, and since 2024 the system has contributed to more than 40 mitigation actions at sources estimated to emit 1.2 million tonnes of methane (UNRIC). The models are light enough to train on a single GPU (AI Magazine).

Weather forecasting is another. The European Centre for Medium-Range Weather Forecasts put its machine-learning system, the AIFS, into operation on 25 February 2025. It outperforms leading physics-based models on many measures, including tropical cyclone tracks, and uses roughly 1,000 times less energy to produce a forecast once the model is trained (ECMWF).

The potential reaches further. The International Energy Agency estimates that well-documented uses of AI could save over 13 exajoules of energy by 2035, about 3% of global final energy consumption, if barriers to adoption are overcome (IEA). The condition matters. Projected savings remain hypothetical until they are delivered; the energy AI uses today is already on the meter.

AI as a cost

Two distinctions prevent most misreadings. The first is between data centres in general and AI in particular. The IEA's figures cover every data centre, from email and streaming to cloud storage. AI is the fastest-growing share, and estimates of its own footprint are modelled because companies do not report it separately.

The AI estimates span wide ranges because none of the large technology companies reviewed reports AI-specific environmental data. They rest on assumptions about AI's electricity demand, the grids that supply it and the water used for cooling and power generation.

The second distinction is between a single use and the whole system. Google reports that its median Gemini text prompt used 0.24 Wh of electricity, 0.03 g of CO₂e and 0.26 mL of water in May 2025. The figures cover serving prompts, including the supporting infrastructure, but exclude model training and some other lifecycle impacts. They come from Google's own measurements (Elsworth et al.). At an assumed 0.3 Wh per query, ten queries a day add about 0.03% to an average Briton's daily electricity use, by Hannah Ritchie's calculation (Sustainability by Numbers).

The type of task matters as much as the number of users. Simple text queries now use less electricity than running a television for the same time, the IEA notes, while video generation, reasoning and agentic tasks can use hundreds or thousands of times more energy per query. Efficiency per task is improving fast, total demand is still rising, and the strain falls on the local grids and water supplies where data centres cluster.

AI as a communications problem

Getting the numbers right is the easier part. The harder part is what organisations claim on the back of them, and three questions expose most weak AI stories.

The first is attribution. When an organisation reports lower emissions after introducing AI, how much of the improvement came from the algorithm, and how much from new equipment, changed operating practices or a cleaner energy supply? A claim that credits AI alone overstates its role and understates everything else that changed.

The second is comparability. A model's footprint depends on the task, the hardware, the efficiency of the data centre and the grid it draws on. A single number travels well and explains little. Figures measured on different boundaries can differ by an order of magnitude, as published estimates of energy per AI query already do.

The third, and the most overlooked, is materiality. Does the AI application matter to the organisation's environmental performance, or is it there because it looks innovative? The pressure to show AI adoption can push minor applications into annual reports while more effective, less fashionable measures go unmentioned. The result is a familiar problem by a new route: a sustainability story accurate in its details and misleading in its emphasis.

None of these questions is new. Offsetting claims raised the same issues of attribution, and carbon footprints the same issues of boundaries. Organisations that learned those lessons already know how to tell the AI story well.

The ethics that matter for sustainability

General AI ethics apply here too, and four issues bear directly on sustainability communication.

The most immediate is the line between evidence and illustration. AI can produce a restored wetland, a thriving reforestation site or a community benefiting from a programme in seconds. These images persuade, and they blur the line between what happened and what might. For organisations whose credibility rests on showing real outcomes, that line has to stay visible.

Accountability follows. When AI-assisted analysis feeds an environmental claim that proves wrong, the organisation that made the claim answers for it. Someone has to check the figures, sources and reasoning before publication, and everyone internally should know who that is.

Accuracy is a related risk. Language models produce fluent, confident errors, including invented figures and citations. In a field already fighting misinformation, one fabricated statistic can undo years of careful work.

Then there is the question of whose knowledge counts. Models learn from what is online, where knowledge from the Global South and from indigenous communities is thinly represented. Sustainability stories built on that material risk narrowing whose reality gets told.

Principles for communicating AI ethically

Credible AI communication rests on the standards that already govern sustainability claims. Seven principles put them into practice.

1. Start with the problem. Name what is being solved and for whom before naming the technology.
2. Quantify what you can, acknowledge what you cannot. Use credible data for both benefits and impacts, and be open about gaps, assumptions and uncertainty.
3. Say which scale you mean. Per use and in aggregate are different stories; mixing them misleads in either direction.
4. Credit AI accurately. Separate what AI contributed from the data, people and other technologies around it.
5. Show the people in the loop. Explain who checks the output and who is accountable for it.
6. Be transparent about material AI involvement. Disclose AI-generated content where it could affect authenticity, interpretation or trust, particularly in evidence-based communication.
7. Claim only what you can substantiate. “Green AI” or a “carbon-neutral model” without data and boundaries is greenwashing by another name.

The credibility test

A simple check for any AI story: could a sceptical climate scientist read it without wincing? If the benefit is specific and verified, the cost is quantified with its limits and AI's role is neither inflated nor hidden, the answer is yes. That story takes more work, and it is the one that will still hold when the next set of figures comes out.

UNUECO helps organisations communicate their sustainability and AI work with evidence and a sense of proportion. Talk to us.

Sources

International Energy Agency (2026). Key questions on energy and AI: executive summary.

International Energy Agency (2026). Data centre electricity use surged in 2025.

de Vries-Gao, A. (2026). The carbon and water footprints of data centers and what this could mean for artificial intelligence. Patterns 7(1), 101430.

Elsworth, C. et al. (2025). Measuring the environmental impact of delivering AI at Google scale. arXiv.

Ritchie, H. (2025). What's the carbon footprint of using ChatGPT or Gemini? Sustainability by Numbers.

ECMWF (2025). ECMWF's AI forecasts become operational.

United Nations Regional Information Centre (2026). AI helping UN detect methane emissions.

AI Magazine. How UNEP and IMEO use AI to track methane emissions.