Gendered Pathways in AI Companionship: Cross-Community Behavior and Toxicity Patterns on Reddit

Gendered Pathways in AI Companionship: Cross-Community Behavior and Toxicity Patterns on Reddit Erica Coppolillo1, 2, 3 and Emilio Ferrara1 1University of Southern California, Los Angeles, California 2University of Calabria, Rende, Italy 3ICAR-CNR, Rende, Italy Abstract AI-companionship platforms are rapidly reshaping how peo- ple form emotional, romantic, and parasocial bonds with non- human agents, raising new questions about how these rela- tionships intersect with gendered online behavior and expo- sure to harmful content. Focusing on the MyBoyfriendIsAI (MBIA) subreddit, we reconstruct the Reddit activity histo- ries of more than 3,000 highly engaged users over two years, yielding over 67,000 historical submissions. We then situ- ate MBIA within a broader ecosystem by building a his- torical interaction network spanning more than 2,000 sub- reddits, which enables us to trace cross-community path- ways and measure how toxicity and emotional expression vary across these trajectories. We find that MBIA users pri- marily traverse four surrounding community spheres (AI- companionship, porn-related, forum-like, and gaming) and that participation across the ecosystem exhibits a distinct gen- dered structure, with substantial engagement by female users. While toxicity is generally low across most pathways, we ob- serve localized spikes concentrated in a small subset of AI- porn and gender-oriented communities. Nearly 16% of users engage with gender-focused subreddits, and their trajecto- ries display systematically different patterns of emotional ex- pression and elevated toxicity, suggesting that a minority of gendered pathways may act as toxicity amplifiers within the broader AI-companionship ecosystem. These results charac- terize the gendered structure of cross-community participa- tion around AI companionship on Reddit and highlight where risks concentrate, informing measurement, moderation, and design practices for human-AI relationship platforms. Introduction Artificial–intelligence chatbots capable of sustaining inti- mate, emotional, and romantic interactions have grown dra- matically in both visibility and adoption. Dedicated plat- forms such as Replika (https://replika.com/) and Charac- ter.ai (https://character.ai/) now support deeply personal- ized conversations that foster companionship, erotic engage- ment, and perceived emotional reciprocity (De Freitas et al. 2025; Mlonyeni 2025; Chu et al. 2025). Prior work demon- strates that users often experience strong emotional attach- ment to these systems, rely on them for affective support, and sometimes consider them substitutes for human rela- tionships (Babu et al. 2025; Malfacini 2025). Despite rising public and academic interest, most research on AI intimacy remains platform-specific, focusing on surveys or interviews with users of individual chatbot applications. As a result, we know little about how people who form such attachments behave across broader online ecosystems. Theoretical frameworks on anthropomorphism, relational attachment, and parasocial bonds help explain why emo- tional connections to AI companions emerge (Giles 2002; Derrick, Gabriel, and Hugenberg 2009). At the same time, work on algorithmic intimacy and adaptive conversational systems suggests that AI agents can influence users emo- tional states and expectations through sustained interaction (Hancock, Naaman, and Levy 2020). The MyBoyfriendIsAI (MBIA) subreddit offers a com- pelling lens through which to examine these dynamics. Cre- ated in 2024, MBIA rapidly accumulated over 30, 000 fol- lowers, and tens of thousands of posts centered on emo- tional, romantic, and sexual experiences with AI partners. Although MBIA has raised recent interest (Pataranutaporn et al. 2025), we are the first to explore this community at a larger scale, by examining its relationship to the wider constellation of online communities in which users partic- ipate. Specifically, in this work, we reconstruct two and a half years of Reddit activity (from January 2023 to Septem- ber 2025) for more than 3, 000 highly active MBIA users. Using these data, we build a historical interaction network of more than 2, 000 subreddits and 27, 000 directed edges, encoding the order in which users first posted across com- munities. We organize our investigation around three main research questions: RQ1: What is the historical activity of MBIA users on Reddit? Which communities and pathways most fre- quently precede or follow engagement with AI- intimacy spaces? RQ2: How do these surrounding communities differ in terms of thematic focus, toxicity levels, and user gen- der composition? RQ3: Do MBIA users encounter gender-oriented or poten- tially radical/extremist communities along their tra- jectories, and how do their emotional and toxic be- haviors differ within these spaces? To answer these questions, we combine embedding-based topic modeling, toxicity estimation, and gender inference with large-scale network reco

…(Full text truncated)…

This content is AI-processed based on ArXiv data.