AI-generated content and misinformation are increasingly prevalent on social networks. While prior research primarily examined textual misinformation, fewer studies have focused on visual content's role in virality. In this work, we present the first large-scale analysis of how misinformation and AI-generated images propagate through repost cascades across five ideologically diverse Reddit communities. By integrating textual sentiment, visual attributes, and diffusion metrics (e.g., time-to-first repost, community reach), our framework accurately predicts both immediate post-level virality (AUC=0.83) and long-term cascade-level spread (AUC=0.998). These findings offer essential insights for moderating synthetic and misleading visual content online.
The rise of social media platforms like Reddit, Instagram, and X has transformed information dissemination. Unlike traditional media, which rely on editorial oversight to ensure accuracy, these platforms prioritize user-driven engagement signals, such as upvotes and shares, over credibility. Consequently, physiologically arousing or false content often spreads faster than neutral or truthful information [1], [2].
Compounding this issue is the rise of generative AI (GenAI) and deepfake image manipulation tools. Visual content is processed more quickly than text and often transcends language barriers. Past research has also found that images are inherently more memorable than text [3], [4]. Regardless, distinguishing real from synthetic visuals remains difficult, even with recent advances in detection [5], [6]. While recent studies examine visual misinformation, such as COVID-19 images on Twitter [7] and meme virality on Reddit [8], research on virality in ideologically driven communities remains sparse.
To address this, we analyze how misinformation and GenAI influence the virality of visual and textual posts across diverse Reddit communities. We specifically study virality at two distinct levels: immediate individual post popularity (“postlevel virality”) and long-term content propagation through sharing or reposting (“cascade-level virality”). A diffusion cascade refers to the complete pathway through which a post spreads across a network via repeated reposting. At the cascade level, we examine features of entire diffusion cascades, such as overall size, structural complexity, and community reach. Specifically, we explore the following questions:
- How do visual and textual features affect virality at the individual post and cascade levels? 2) When predicting how widely content spreads (cascadelevel virality), how do content-based predictors such as text and visual attributes compare against early sharing signals and community interaction patterns (diffusion context-based features)? 3) What post-specific features characterize pure misinformation, pure AI-generated, and misinformation and AIgenerated imagery (“mixed-flag” content) classifications at the post and cascade level?
The first question builds upon findings that emotionally engaging or misleading content tends to be more viral [1], [2]. Prior studies have focused on general imagery, typically without distinguishing misinformation or synthetic media [9], [10]. By explicitly examining misinformation and AIgenerated visuals, we offer targeted insights into how these specific content types shape virality uniquely within ideologically driven communities.
The second question addresses the gap in understanding Reddit’s diffusion dynamics due to its lack of an explicit social graph-unlike Twitter, where propagation is more straightforward to analyze. Our study reconstructs repost cascades by leveraging shared image URLs and crossposting data, allowing us to compare how intrinsic features of posts (content-based) versus early sharing behaviors and diffusion metrics (contextbased) predict virality. Additionally, we introduce novel visual markers specific to generative AI, such as digital artifacts and image noise, identifying previously unexplored connections between these visual elements and content diffusion.
The third question investigates the distinct and combined effects of misinformation and AI-generated imagery (“mixedflag posts”). Mixed-flag posts-those identified as containing both misinformation and synthetic or manipulated imagery-may drive unique engagement patterns that neither misinformation nor synthetic visuals alone can explain. Prior research highlights misinformation’s power to engage via novelty or emotional provocation, and AI-generated content’s effectiveness through visually striking imagery and positive framing despite questionable credibility [2], [11]. However, how these features interact in combination remains underexplored. We address this gap by systematically studying pure misinformation, pure AI-generated content, and mixed-flag posts, providing nuanced insights into their distinct engagement patterns and diffusion trajectories.
We present the first large-scale cascade-level study of visual content diffusion on Reddit, focusing on how misinformation and GenAI content spread across user communities. Our dataset includes 1,800+ repost cascades clustered via shared image URLs and crossposts, encompassing 5,660 posts from five ideologically diverse subreddits. Each cascade is enriched with multimodal features: text sentiment, visual descriptors (e.g., digital manipulation indicators, image sharpness, and GenAI-specific markers), and user interaction metrics (e.g., upvote ratio, comment counts). Diffusion-based attributes such as cascade depth, Wiener index (structural virality), timeto-first repost, and cross-subreddit reach are also computed. Misinformation posts receive unusually high engagement (mean score of 36,000+, 2,
This content is AI-processed based on open access ArXiv data.
