Grassroots Infrastructures of Influence: Multimodal Memes as a Defensive Communication Strategy in Wartime Ukraine
Hanna Starkova (Iryna Ivanova, Svitlana Sadrytska, Oleksandra Yehorova)
Big tech platforms are often analysed as top-down influence infrastructures where algorithms, bots, and hostile actors manipulate users. However, in the context of the large-scale Russian war against Ukraine, Ukrainian youth utilise these same platform affordances to construct a bottom-up, grassroots infrastructure of cognitive defence against weaponised information disorders. This study conceptualises multimodal internet memes as digital vernacular and contemporary humorous folklore. This framework allows us to examine memes not merely as viral entertainment but as tactical tools for psychological resilience and civic empowerment. We employ a longitudinal mixed-methods approach focusing on a cohort of Ukrainian media students — future professionals situated at the frontline of the information war. Data is drawn from quantitative surveys conducted before the large-scale invasion (early 2022, N=133) and during the protracted war phase (2024, N=133), complemented by qualitative focus groups (2025). Our findings reveal that students deploy this humorous folklore as a 'defensive communication strategy' (DCS) serving two primary functions related to online engagement and wellbeing: - affective filtering (wellbeing) In an environment saturated with traumatic content, multimodal memes function as a crucial coping mechanism. By translating existential threats into humour, memes create a cognitive distance that reduces anxiety, prevents the psychological 'freeze response,' and sustains users' mental wellbeing. - epistemic vigilance (empowerment) Rather than promoting passive consumption or tolerance to hostility, this grassroots influence infrastructure triggers critical verification. Our data reveals that 92.5% of respondents use the emotional engagement sparked by internet memes as a direct prompt to clarify information and verify facts. Ultimately, this study demonstrates a profound shift from passive media consumption to 'militant agency.' It highlights how users empower themselves to counter hostility, neutralise information disorders, and sustain civic participation. By examining this Ukrainian case, we offer novel theoretical and empirical insights into how grassroots digital solidarities can reshape online political conversations and mitigate the harmful consequences of opaque platform dynamics.
Understanding the Visual Discourse on Climate Action: A Semantic Clustering Approach
Luigi Arminio (Luca Rossi)
Visual content has become increasingly prominent on social media, especially around polarising issues such as climate change [1]. However, research on visual environmental communication on social media remains limited: large-scale computational studies often overlook the connotative meaning of images [2], while small-scale qualitative work captures meaning but lacks generalisability. To address this gap, we applied a VLLM-driven semantic visual-clustering approach [3] to examine visual discourse on climate change on Facebook. Drawing on a curated list of 327 European climate stakeholders — 182 actors supporting climate action and 145 counter-actors contesting such agendas — we collected 16,130 Facebook images posted between October 2023 and August 2024. We used Visual Large Language Models (VLLMs) to generate connotative image descriptions and clustered these descriptions to map the visual climate discourse. Through this process, we identified six main visual themes: Climate Change, Animals, Plants and Flowers, Energy Transition, and Urban Mobility and EVs, Cycling infrastructure. This relatively compact thematic space is consistent with earlier findings [4] on visual climate communication and spans the full breadth of the contemporary climate debate, from the impact on ecosystems to the contested socio-economic implications of the green transition. We found markedly different ideological distributions across themes: supportive actors dominated nature-centric themes reflecting biocentric environmentalism, while counter-actors had a notable presence in themes focused on the socio-economic dimensions of decarbonisation. Within these contested themes, sub-clustering revealed divergent framings — supportive actors emphasised the benefits of renewable energy and sustainable mobility, while counter-actors foregrounded economic costs, logistical challenges, and rural landscape impacts. Beyond content, we also examined audience emotional engagement: visual content produced by counter-actors consistently attracted significantly more negative emotional reactions than that produced by actors, a pattern that was stable across time and aligned with the broader rise of right-wing populist opposition to climate policy in Europe.
Seen but Not Named: Cross-Modal Candidate Visibility on Political TikTok
Eva-Maria Vogel (Christian Pipal, Morgan Wack, Frank Esser)
Political communication research measures candidate salience almost entirely through text, typ- ically through name mentions in news coverage or social media posts (e.g., Hong and Nadler, 2012). On TikTok, where 170 million Americans now seek political news (Stocking et al., 2024), this approach misses how candidates are made visible to audiences. Candidates appear on screen without being named, or are named without appearing visually. Text-only methods detect neither pattern. We introduce a cross-modal candidate salience framework with three measurement streams: 1) keyword extraction from captions and auto-generated transcripts; 2) a computer vision pipeline using ArcFace facial embeddings (Deng et al., 2022); and 3) a multimodal large language model (mmLLM) (Qwen2.5-VL-72B) classifier. Each approach is benchmarked against human-coded ground truth for validation. We apply the framework to 4,400 videos from the 50 most-followed U.S. political TikTok news in- fluencers (independent creators who regularly produce political content for large online audiences) and a matched cross-partisan sample of TikTok accounts run by established U.S. news organiza- tions with print or broadcast presence, covering eight weeks before the 2024 U.S. election. News influencers operate outside institutional editorial structures with distinct content logics (Riedl et al., 2023; Rothut, 2025), making their cross-modal patterns a distinct case from legacy media. Three research questions guide the analysis. To what extent do textual and visual candidate pres- ence diverge? Do news influencers and traditional news accounts differ in cross-modal patterns? How does partisan alignment shape these patterns? By mapping how candidates are made visible across text and image, this study shows how the content logics of news influencers and traditional media produce structurally different forms of candidate salience on political TikTok. Divergences across account types and partisan lines re- veal how editorial choices operate across both channels, with direct implications for how political information asymmetries form on visual-first platforms. The study improves the operationalization of candidate salience in agenda-setting research by ex- tending it to the visual domain of short-form video. The cross-modal framework offers a replicable pipeline combining NLP, computer vision, and mmLLM classification, applicable across electoral contexts.
Memetic Warfare and the Politics of Influence: From Ukraine to Iran
Tine Munk
Memetic warfare has become a central feature of contemporary conflict. War is no longer fought only on the battlefield, but through communication that shapes perception, emotion and meaning. Digital content, including memes, short videos, stylised visuals and increasingly AI-generated media, circulates rapidly across social media platforms, shaping how conflicts are understood before facts have stabilised. Conflict now unfolds across an online-offline continuum, where digital narratives and material events continually shape one another. This paper examines how memetic warfare operates as both an offensive and defensive strategy within this continuum. It focuses on how state and state-aligned actors use platform infrastructures to shape political communication, bypass diplomatic channels, and produce highly visible narratives designed to influence public opinion at speed and scale. The analysis draws on data from the war in Ukraine, digital activity surrounding geopolitical tensions over Greenland, and ongoing information dynamics associated with the conflict in Iran. Across these contexts, memetic content is used not only to respond to events but also to pre-empt, frame, and, at times, escalate them. In doing so, actors contribute to a form of communication that prioritises immediacy, affect and visibility, often at the expense of verification. The paper argues that memetic warfare functions as a form of distributed agenda-setting, in which influence emerges through interactions between actors and platform systems. Although earlier work has often emphasised grassroots participation, this paper points to a shift towards more strategic and increasingly offensive uses of memetic communication by state actors. These practices move fluidly across the online-offline continuum: digital content shapes public debate, political positioning and international perception, while offline developments generate new cycles of memetic production. Methodologically, the paper combines digital ethnography with cross-platform content and media analysis to trace how narratives are produced, circulated and recontextualised across digital and physical spaces. By examining these dynamics across multiple contexts, the paper shows how memetic warfare reshapes political communication in conflict, reconfigures agenda-setting processes, and blurs the boundaries between diplomacy, propaganda and public engagement.
