Characterizing AI Manipulation Risks in Brazilian YouTube Climate Discourse

TL;DR

This study investigates climate discourse in Brazilian YouTube, leveraging a large-scale, psychology-informed dataset annotated with 10 persuasion strategies and 7 Theory-of-Mind traits to understand engagement and manipulation risk. Through three case studies—engagement modeling, popularity prediction, and comment generation—the work shows emotional and moral framing robustly boosts engagement, while fact-based appeals have more variable reach; ToM cues enhance predictive performance, and synthetic, targeted comments can be generated at scale using fine-tuned LLMs. The authors release a public dataset of 226,775 videos and 2,756,165 comments with rich annotations, highlighting ethical considerations and governance needs for mitigating AI-driven manipulation in climate communication. The findings underscore the importance of monitoring synthetic narratives and designing robust guidelines for responsible AI use in digital climate discourse, particularly in geopolitically pivotal regions like Brazil.

Abstract

Climate change poses a global threat to public health, food security, and economic stability. Addressing it requires evidence-based policies and a nuanced understanding of how the threat is perceived by the public, particularly within visual social media, where narratives quickly evolve through voices of individuals, politicians, NGOs, and institutions. This study investigates climate-related discourse on YouTube within the Brazilian context, a geopolitically significant nation in global environmental negotiations. Through three case studies, we examine (1) which psychological content traits most effectively drive audience engagement, (2) the extent to which these traits influence content popularity, and (3) whether such insights can inform the design of persuasive synthetic campaigns--such as climate denialism--using recent generative language models. Another contribution of this work is the release of a large publicly available dataset of 226K Brazilian YouTube videos and 2.7M user comments on climate change. The dataset includes fine-grained annotations of persuasive strategies, theory-of-mind categorizations in user responses, and typologies of content creators. This resource can help support future research on digital climate communication and the ethical risk of algorithmically amplified narratives and generative media.

Figures (7)

• Figure 1: Proposed analytical framework includes data collection and preprocessing, annotations, and three case studies to understand the user engagement. Videos and comments' mean values, with median values in parentheses, are included. Ten persuasion strategies and seven theory of mind categories are ordered based on the distributions over the entire dataset.
• Figure 2: Monthly counts of climate-related Brazilian videos during 2019-2025. Vertical lines mark the start of each year.
• Figure 3: Effect of different persuasion strategies on the like ratio for (A) all videos and (B) monthly trends by video length. Solid points indicate statistically significant regression coefficients at the 0.05 level.
• Figure 4: (A) Pearson correlations at the video level between 10 persuasion strategies and 7 ToM categories. (B) Effects of ToM mental states on audience engagement of likes and replies. Note.$^{*}p < .05$; $^{**}p < .01$; $^{***}p < .001$
• Figure 5: Sampled Portuguese comments generated by Believe, Denial, and Extreme models, with translations below.