Can you trust what AI says about climate? COP31 pushes AI firms to disclose climate footprint, publish how systems shape information online

People increasingly ask AI chatbots about climate change. At COP31 today, the countries and agencies leading global efforts on climate information integrity said no one can yet be sure how much to trust the answers, or the industry's claims about its own climate footprint.

The Global Initiative for Information Integrity on Climate Change launched its Priorities for Action on AI for Climate Information Integrity, the first global framework to address both problems. It is co-chaired by Brazil, the United Nations and UNESCO. It sets out what States, AI developers, advertisers, researchers, journalists and funders can do to ensure people can rely on climate information from AI, and about it.

Two sides of one gap

The framework's central argument is that AI has created a two-sided information gap. One side is the reliability of what AI tells people about climate science. The other is the reliability of what the public is told about AI's own environmental impact.

On the first, the Priorities lean on the Independent International Scientific Panel on AI, whose preliminary report cautions that "producing fluent text is easier than producing factual text." The Panel notes that users often mistake confident language for accuracy. It also describes "sycophancy," in which AI responses reinforce what users already believe, regardless of whether it is true. And it notes that today's models are optimised for only a small fraction of the world's languages.

The stakes rise during climate disasters, when AI-generated images and videos can spread faster than official warnings. They also rise because the tactics of climate delay have shifted from denying the science to undermining solutions and trust in them.

On the second side, the document notes that communities near AI infrastructure are often "left in the dark" about its environmental impact and lack data on its greenhouse gas emissions. Little is known about how AI is being used to boost productivity in coal, oil and gas exploration and extraction, which could enable more emissions than AI applications in renewables deliver in savings. Claims that AI will help climate modelling or energy efficiency, meanwhile, circulate with little means of testing them. Without reliable data, the Priorities warn, the information void risks being filled by greenwashing.

What is being asked of AI companies

The framework asks AI developers to:

• Assess whether and how system design, targeting and optimisation for engagement undermine climate information integrity, before release and after major updates, and publish the findings.
• Give independent researchers privacy-preserving access to data.
• Show sources and uncertainty, and not “reward agreement above accuracy.”
• Link to authoritative scientific sources on climate science questions.
• Publish how reliability varies across languages.
• During climate disasters, prioritise reliable warnings, label synthetic imagery and audio, and work with national meteorological and disaster-management services.
• Measure and publicly disclose energy use and carbon, water and land footprints under common standards that allow independent verification.
• Disclose whether and how their products are marketed or supplied for fossil fuel exploration, extraction and production.
• Ensure that claims about AI's climate benefits are specific, evidence-based and independently verifiable, and that they are never used to offset reported emissions.

States are asked to build information integrity into national AI strategies, climate plans and Action for Climate Empowerment strategies. They are also asked to require AI providers to disclose environmental impacts and assess the reliability of their climate-related outputs. Where governments use AI in climate services or early warning, they should disclose it and keep human oversight. They should also ensure fair compensation for intellectual property, including original journalism, used to train AI tools.

Journalists are asked to investigate AI's footprint and fossil fuel uses, and to verify AI-generated imagery before publication. Researchers are asked to test corporate claims against evidence and build open, multilingual benchmarks for how AI systems answer climate questions.

Follow the money

The Initiative also backed the Charter for Climate Information Integrity in Advertising, launched by the Conscious Advertising Network and Climate Action Against Disinformation. Advertising remains the dominant business model of the digital information ecosystem, but its supply chain is so opaque that brands often cannot tell what their money funds, including climate disinformation.

The charter calls for transparency across that chain, with advertisers able to see where their ads appear while user privacy is respected. It also calls for verification of all advertisers on digital platforms, action on greenwashing, and public disclosure of advertisers' relationships with fossil fuel actors.

Germany puts up the first money

The Initiative argued that this work needs resources, and Germany answered at COP31 with a EUR 2 million commitment to the Global Fund for Information Integrity on Climate Change. The fund is administered by UNESCO and pays for research, communication strategies and public awareness campaigns. The Initiative urged other countries to follow.

Momentum from Belém

The Priorities build on the Belém Declaration on Information Integrity on Climate Change. Information integrity entered a COP outcome decision for the first time a year ago at COP30, in the Mutirão Decision. The Declaration commits signatories to "equitable access to accurate, consistent, evidence-based, and understandable information on climate change for all stakeholders."

Twenty-nine countries and the European Union have now signed and are implementing it, and it remains open for endorsement. The Initiative says it will keep supporting research and projects on AI and climate information integrity. It will report progress at the Global Dialogue on AI Governance and at COP32.