Beyond Facts: How We Are Engaging with Climate Change Discussions for COP30

March 17, 2026

At COP 30 in Brazil, climate change was once again on the spotlight as one of the most pressing topics globally. From the Amazon rainforest to global policy debates, the conversation is heating up. But the key question remains: how do we move people from simply hearing about climate change knowledge to voluntarily engaging with it?

We already have scientific evidence for climate change and its impacts. Yet, public engagement often seems like an uphill battle. Is the challenge the message itself, or how we choose to deliver it?

Studying Climate Engagement in Brazil

To explore this question, we analyzed 226,775 Brazilian Portuguese Youtube videos published between 2019 and 2025 [1]. Drawing on social psychology research, we curated a list of ten persuasion strategies used in climate discourse, such as emotional appeals (e.g., evoking fear or empathy) and logical arguments (e.g., cause-and-effect reasoning). Using large language models (LLMs) with human validation, we annotated every video according to these strategies. We then measured engagement such as the video like-to-view ratio. The results were striking:

Logic- and statistics-driven content received significantly less engagement.
Moral and emotional appeals led to more user interaction.

In short, audiences responded more strongly to climate content that spoke to feelings and values rather than just raw data.

Not All Channels and Topics Are Equal

Interestingly, the effectiveness of these strategies also depend on who published the videos.

Emotional appeals boosted engagement for most channels, but not for scientific research institutes.
Logical, data-based arguments were generally less effective, but they performed better when delivered by national government channels.

This suggests that tailoring communication strategies to the messenger matters as much as the message itself.

Topic distributions for the climate youtube discourse dataset. Discussion topics also influence the effectiveness of informational appeals. For instance, when videos address climate change in the context of Africa, scarcity emerges as a strong predictor of comment activity, even though it has minimal influence overall.

Topic distributions for the climate youtube discourse dataset. Discussion topics also influence the effectiveness of informational appeals. For instance, when videos address climate change in the context of Africa, scarcity emerges as a strong predictor of comment activity, even though it has minimal influence overall.

The Role of Comments, The Risks of AI

As we navigate the complexities of climate communication, a new challenge is emerging: the ability of LLMs to generate human-like text. To test the potential influence of this, we fine-tuned models on 4,000 Brazilian Portuguese YouTube comments (just 1.5 hours using a single GPU). The resulting models could generate climate-related comments aligned with either supportive or denialist views. For example:

Supportive: “A energia solar é a energia do futuro, e o futuro já chegou!” (“Solar energy is the energy of the future, and the future is already here!”)
Denialist: “Aquecimento global é uma farsa.” (“Global warming is a hoax.”)

This illustrates a growing risk. What looks like authentic grassroots sentiment in a video’s comments section can now be automatically generated at scale. By flooding online spaces with either pro-climate or denialist messages, LLMs can blur the line between genuine opinion and engineered discourse. The consequence is more than “fake news.” It’s the possibility of mass opinion manipulation: creating a false sense of consensus, reinforcing echo chambers, and undermining informed decision-making on critical issues like climate action.

Why This Matters for COP30

The stakes at COP30 are high. Understanding how persuasion works in climate communication can help foster deeper engagement, spark more meaningful conversations, and ultimately drive collective action. Public engagement has already proven transformative, as seen with Greta Thunberg’s activism, which rippled across social media and beyond. This conversation is also deeply connected to the United Nations Sustainable Development Goals (SDGs):

SDG 13: Climate Action
SDG 7: Affordable and Clean Energy
SDG 11: Sustainable Cities and Communities
SDG 12: Responsible Consumption and Production

Climate change is not just an environmental issue. It intersects with economic development, social justice, and sustainability.

For policymakers, our findings suggest that even in formal contexts, choosing the right persuasion strategies can make scientific evidence far more impactful for the public.

Looking Ahead

This research was driven by concerns about the potential misuse of AI to shape public opinion. But it also shows the opportunity: if we better understand persuasion, we can use it responsibly to build support for climate action.

As we look towards COP30, let's commit to communicating in ways that truly resonate, turning understanding into action.

About the Authors

Wenchao Dong is a doctoral researcher at the Max Planck Institute for Security and Privacy (MPI-SP), Germany. His research lies at the intersection of social psychology and language modeling, with a particular focus on how advances in artificial intelligence can be responsibly adopted for public benefit and how they may influence long-term societal inequalities. He received his Master’s degree from the Korea Advanced Institute of Science and Technology (KAIST) and interned at the Institute for Basic Science, South Korea. His work has been partially supported by the Microsoft Accelerate Foundation Models Research (AFMR) initiative.

Marcelo Sartori Locatelli is a Master's student at Universidade Federal de Minas Gerais (UFMG), Brazil and a guest researcher at the Max Planck Institute for Security and Privacy (MPI-SP), Germany. His research interests center on computational social science and social computing, focusing on the understanding of how harmful phenomena takes shape in social media platforms.

Meeyoung (Mia) Cha is a leading AI and data science researcher dedicated to applying technology for societal benefit. She is a Scientific Director at the Max Planck Institute for Security and Privacy (MPI-SP), an Adjunct Faculty at Ruhr University Bochum (RUB) in Germany, and a Professor at the Korea Advanced Institute of Science and Technology (KAIST). Her research focuses on data science and computational social science, particularly in areas such as misinformation, poverty mapping, fraud detection, and human-machine interactions. Her work has earned widespread recognition, including Best Paper Awards, Test-of-Time Awards, the Hong Jin-Ki Creator Award, the Korean Young Information Scientist Award, and the ACM Distinguished Member status.