AI analytics could help policymakers track climate risk perceptions

A new study published in Sustainability suggests that artificial intelligence-driven social media analytics can help policymakers track how people interpret climate risks, where misinformation spreads and which messages may support climate adaptation and sustainability action.

Using 29,576 English-language posts from X (formerly, Twitter) collected in December 2025, the study "AI-Driven Social Media Analytics for Assessing Climate Change Perceptions and Supporting Adaptation and Sustainability Policies," maps emotional and thematic patterns in climate change discourse.

Climate debate online is driven by fear, urgency and public distrust

The findings show that climate change discussion on X is overwhelmingly shaped by negative sentiment. Of the posts analyzed, 18,654 were classified as negative, compared with 6550 positive posts and 4372 neutral posts. The deep learning model used for sentiment detection achieved 92.5% accuracy, with strong precision, recall and F1-score results, indicating a reliable classification of emotional orientation across the dataset.

The dominance of negative content points to a public conversation centered on threat, uncertainty and dissatisfaction. Users frequently framed global warming as an immediate crisis rather than a distant environmental issue. Posts reflected anxiety over extreme weather, frustration with climate inaction, concern about the future of the planet and criticism of human responsibility for environmental damage. This pattern suggests that climate change is not merely being discussed as a scientific topic online, but as a lived social and political concern.

The language patterns add weight to that finding. Negative posts commonly included references to cold weather, winter, ice, fear, bad conditions, hoaxes and scams. That combination reveals two overlapping forms of discourse. One reflects genuine concern over environmental instability and perceived worsening of weather conditions. The other reflects skepticism, where users invoke cold days or snowfall to question the validity of global warming. This blend of anxiety and denial makes online climate discussion especially difficult for policymakers and communicators.

Positive posts formed a smaller but important part of the conversation. These messages emphasized awareness, science-based action, clean energy, collective responsibility and the need to stop or reduce climate harm. Rather than denying the seriousness of global warming, positive discourse tended to frame the crisis through action, adaptation and solutions. This matters for sustainability policy because constructive communication can help convert public concern into engagement.

Neutral posts largely reflected news sharing, general discussion, scientific information and non-emotional commentary. These posts show that social media functions not only as a platform for emotional reaction, but also as a channel for information exchange and public deliberation. In climate communication, this neutral space can be important because it provides room for evidence-based messaging, policy explanation and educational interventions.

The emotional structure suggests that public concern is high, but concern alone does not guarantee support for climate action. If anxiety is not matched with trusted information and practical pathways for adaptation, it can turn into fatalism, anger or resistance. At the same time, the presence of misinformation and skepticism indicates that climate communication must address false or misleading narratives directly, especially those based on short-term weather experiences.

The findings also show why AI-driven monitoring can be valuable for climate governance. Traditional surveys provide useful snapshots of public opinion, but social media analytics can detect shifts in public emotion and discourse in near real time. For governments and environmental organizations, this can help identify rising concern after extreme weather events, track misinformation patterns and design more targeted communication strategies.

Weather experiences, science debates and carbon responsibility shape the discourse

Topic modelling identified three major themes in the online conversation. The most common theme, representing 41.37% of the corpus, centered on daily weather experiences and skeptical climate discourse. Users often interpreted global warming through immediate personal observations, especially cold weather, snowfall or seasonal irregularities. In these posts, individual experience frequently became the basis for questioning scientific explanations.

This is a major challenge for climate adaptation policy. Climate change is a long-term global process, but many people judge it through local and short-term weather. When a cold day is treated as evidence against global warming, scientific communication loses ground to anecdotal reasoning. The result is a discourse in which personal perception can override scientific consensus.

The second theme, representing 31.62% of the corpus, focused on climate change, energy and scientific evidence. This cluster included discussion of fossil fuels, emissions, temperature changes, sea levels, data, evidence and energy systems. It reflects a more analytical and policy-oriented segment of the online debate, where users connect climate change to science, energy transitions and economic questions.

This part of the conversation creates an opening for evidence-based climate policy. Users in this cluster are not simply reacting emotionally. They are engaging with causes, data and proposed solutions. For policymakers, this audience may be more receptive to detailed communication about renewable energy, carbon reduction, infrastructure resilience and adaptation planning.

The third theme, representing 27.01% of the corpus, focused on carbon emissions, human impact and cause-effect debates. Users discussed CO2, carbon, causation, human responsibility, cooling, ice and whether global warming is natural or human-induced. This theme shows that responsibility remains a central fault line in climate discourse. People are not only debating whether climate change is happening, but who or what is causing it and what should be done.

Responsibility debates are critical for sustainability governance. Climate policy often depends on public acceptance of emission reductions, energy reforms, transport changes, industrial regulation and lifestyle shifts. If people dispute the human role in climate change, they may resist policies that require collective action or behavioral change. If they accept human responsibility, they may be more open to adaptation and mitigation measures.

The three themes together show that social media climate discourse is not a simple split between believers and deniers. It is a layered conversation where everyday experience, scientific reasoning, skepticism, policy concern and responsibility attribution coexist. A user may express fear about climate impacts while also questioning specific policies. Another may accept climate science but criticize government action. This complexity makes AI-driven discourse analysis useful because it can capture patterns that would be difficult to detect manually at scale.

The keyword relationships also show that climate discourse is organized around interconnected concepts rather than isolated terms. Discussions of global warming are linked to weather, people, science, carbon, energy, evidence, money, politics and planetary risk. This confirms that climate change is understood online as a broad social issue, not only an environmental one.

For sustainability communication, it's clear that climate campaigns cannot rely only on technical explanations. They must connect climate science to daily experiences, address misinformation, explain responsibility and offer practical solutions. Public messaging must be emotionally aware, scientifically grounded and locally relevant.

AI analytics can help policymakers build stronger climate adaptation strategies

The research further highlights a growing role for AI in climate policy. AI-driven social media analytics can help governments and institutions understand how people perceive climate risk, where confusion emerges and which narratives dominate public discussion. That information can support more adaptive and responsive environmental governance.

Climate adaptation policies often require public cooperation, including changes in water use, urban planning, disaster preparedness, agriculture, housing, insurance and emergency response. Public perception affects whether those policies are accepted, ignored or resisted. If social media shows widespread concern about extreme heat, flooding or storms, officials can respond with targeted information and preparedness campaigns. If misinformation is spreading, they can intervene earlier with corrective communication.

The analysis also shows that negative sentiment can be both a challenge and an opportunity. High concern may signal readiness for policy action, but it may also reflect fear, distrust or anger. Effective communication should avoid amplifying panic. It should pair risk information with credible solutions, showing how adaptation, clean energy and sustainable behavior can reduce harm.

The positive discourse found in the dataset offers a foundation for that approach. Posts emphasizing science, action, energy transformation and collective responsibility show that social media can support public mobilization. Environmental organizations can use such narratives to promote renewable energy, emissions reduction, conservation and local adaptation initiatives.

Skepticism requires a different strategy. The weather-based skeptical theme indicates that many users rely on direct observation to interpret climate change. Communication efforts should therefore explain the difference between weather and climate in accessible terms, while connecting long-term warming trends to familiar local impacts. Trust-based messaging from scientists, community leaders and public agencies can help bridge the gap between technical knowledge and public experience.

The findings also point to the need for better climate literacy. Misinformation and contested claims about carbon, causation and policy responsibility show that knowledge gaps remain significant. Education campaigns should not only present scientific facts, but also explain how climate evidence is produced, why short-term weather does not disprove long-term warming, and how human activity contributes to emissions.

AI tools can make this work more precise. By tracking sentiment and topics over time, institutions can identify which messages resonate, which claims are spreading and which communities may need different forms of outreach. Social media monitoring can also reveal emerging risks after disasters, policy announcements or international climate events.

The study's limitations are equally important for understanding what the results can and cannot show. The dataset came only from X, focused on English-language posts and covered one month. Social media users are not fully representative of the global population. Automated accounts were filtered, but some bot influence may remain. Sarcasm, informal language and symbolic expressions can also complicate sentiment classification.