Anger, Fear and Images: What Makes Political Posts Go Viral Online

Anger travels faster than joy on social media, and new research suggests that the combination of negative emotion and rich multimedia content may be the single most potent recipe for viral political content. A study published in Information Systems Frontiers by Venu Bhaskar Puthineedi of NEOMA Business School and Ashish Kumar Jha of Trinity Business School at Trinity College Dublin dissects the anatomy of online virality with unusual precision, analyzing 17,548 posts and roughly 1.5 million comments drawn from Reddit’s largest political communities. Their central finding is stark: negative emotions, particularly anger, disgust and fear, are strongly associated with both engagement and within-platform virality, while positive emotions exert considerably weaker effects. In an era when generative artificial intelligence can manufacture emotionally charged text, images and videos at industrial scale, understanding these organic amplification dynamics has become an urgent scientific and societal priority.

The researchers set out to answer a deceptively simple question: what actually drives a political post to spread within a platform? Rather than relying on coarse positive-versus-negative sentiment labels, the team deployed state-of-the-art deep learning language models, including BERT and DistilRoBERTa, to capture fine-grained affective expressions across the massive comment corpus. These transformer-based models, trained on vast text corpora, can distinguish subtle emotional states such as anger, disgust, fear, sadness, surprise and joy, allowing the researchers to quantify the emotional texture of political discourse at a scale no human coding team could match. The approach reflects a broader shift in information systems research, where natural language processing has become a standard instrument for measuring the psychological signals embedded in millions of online interactions.

The results confirm and extend a long-standing suspicion among communication scholars. Posts saturated with negative emotions, especially anger, disgust and fear, attracted significantly more engagement and were far more likely to rank among the most viral content in their communities. Positive emotions, by contrast, showed only weak associations with spread. This asymmetry echoes earlier findings, notably the influential 2012 study by Jonah Berger and Katherine Milkman on what makes online content viral, but the new work goes further by examining within-platform virality in dedicated political communities, where audiences are already politically engaged and emotionally primed. In such environments, the researchers suggest, outrage functions almost as a currency of attention, rewarding content that provokes indignation or alarm.

Equally consequential is the study’s second major pillar: multimedia presence. Posts accompanied by images or video did not merely reach larger audiences; they also carried measurably higher emotional intensity. The authors interpret this through the lens of peripheral cues in information processing theory. When users scroll rapidly through feeds, they often engage in heuristic rather than systematic processing, and multimedia elements act as powerful peripheral signals that trigger affective responses before any careful evaluation of the message occurs. A striking image can amplify the emotional charge of a headline, and that amplified emotion, in turn, drives clicks, comments and shares. The finding aligns with prior research on visual cueing and image-based engagement in marketing and information systems, but its implications for political communication are considerably more troubling.

It is precisely this emotion-plus-multimedia mechanism that generative AI has supercharged. Tools capable of producing photorealistic images, synthetic video and fluent persuasive text have collapsed the cost of manufacturing emotionally resonant content to nearly zero. The researchers frame their analysis explicitly against this backdrop: the organic amplification dynamics they document become especially consequential when malicious actors can deliberately engineer the very emotional and multimedia cues that the algorithms and audiences reward. Deepfakes and AI-generated imagery depicting fabricated events can exploit the same peripheral-route processing that makes authentic multimedia so engaging, meaning that the virality engine documented on Reddit operates identically on false content as on true content. Emotion does not discriminate between accurate and fabricated claims; it simply accelerates both.

Beyond emotion and media format, the study uncovered a third, less expected class of associations involving platform governance characteristics. Features related to moderator status and account-level authentication were linked to lower emotional amplification, suggesting that governance-related design choices may dampen the emotional escalation that fuels virality. This is a notable contribution because most virality research focuses exclusively on content attributes while ignoring the institutional context in which content circulates. The finding hints that interventions at the level of community structure, such as identifiable, accountable participants and active moderation, could serve as friction against the outrage economy, though the authors are careful to describe these as associations rather than proven causal mechanisms.

Methodologically, the study demonstrates the value of large-scale computational text analysis paired with rigorous validation. The researchers supplemented their model-based emotion scores with external validation exercises comparing machine-generated labels against human reviewer codings, and they tested the robustness of their virality findings using multiple operationalizations, including thresholds based on the top five percent of posts and alternative rankings by upvotes. Regression models controlling for a battery of content and creator characteristics isolated the incremental contribution of emotional and multimedia variables. The authors also report descriptive and correlation analyses across their extensive variable set, and the underlying data are available from the authors on request, reflecting growing norms of transparency in computational social science.

The theoretical stakes extend into classic information systems frameworks. The study engages the elaboration likelihood model, which distinguishes central from peripheral routes of persuasion, and media synchronicity theory, which describes how different media affordances support different communication processes. By showing that multimedia cues operate largely through peripheral emotional pathways, the research offers a bridge between these established theories and the empirical realities of contemporary platform ecosystems. It also connects to a rich literature on social presence, online disinhibition and toxic behavior, which has long argued that anonymity and reduced social cues reshape how users express and absorb emotion online. What the new study adds is a quantified account of how those expressions translate into differential spread within real political communities.

The practical implications are equally significant. For platform designers, the findings suggest that ranking systems which reward engagement may be inadvertently optimizing for negative affect, and that multimedia-aware moderation could target the specific combination of imagery and outrage that spreads fastest. For policymakers grappling with misinformation, the study underscores that fact-checking alone, which operates on the slow central route of processing, may be structurally mismatched against content that persuades through fast peripheral cues. Interventions that introduce friction, prompt deliberation or leverage System 2 thinking, as prior fake-news research has proposed, gain new urgency when the emotional triggers of virality can now be synthetically mass-produced. The researchers position their work as a foundation for governance in the generative AI era, arguing that effective regulation must be grounded in an accurate empirical picture of how content actually spreads.

Limitations remain, as the authors acknowledge. Reddit’s community structure, voting mechanics and moderation norms differ from those of other platforms, and within-platform virality is not identical to cross-platform diffusion into the broader information environment. The observational design cannot fully rule out confounding, and emotional expression measured in text may not capture the full affective experience of users. Yet the scale of the evidence, the granularity of the emotion measurement and the convergence with prior literature make a compelling case that the study’s core conclusions generalize widely. As generative AI continues to lower the barriers to producing emotionally charged multimedia content, the mechanisms documented here will only grow in importance. The study’s message to researchers, platforms and regulators alike is clear: to understand misinformation, one must first understand emotion, and to understand emotion online, one must understand the multimedia vessels in which it travels.

Subject of Research: Emotional and multimedia drivers of virality and misinformation risk in political social media during the generative AI era

Article Title: Emotional Resonance and Multimedia Presence: Unpacking Virality and Misinformation Risks in the Generative AI Era

Article References: Puthineedi, V. B., & Jha, A. K. (2026). Emotional Resonance and Multimedia Presence: Unpacking Virality and Misinformation Risks in the Generative AI Era. Information Systems Frontiers. https://doi.org/10.1007/s10796-026-10816-1

Image Credits: AI Generated

DOI: 10.1007/s10796-026-10816-1

Keywords: emotional resonance, multimedia presence, online virality, misinformation, generative AI, Reddit, social media governance, negative emotions, deep learning, BERT, engagement, platform moderation