People are interacting with LLM-based chatbots more frequently and in increasingly complex ways. While research has explored how interactions with machines influence human attitudes and behaviors toward machines, less is known about the impact of these interactions on human-human interactions. In an exploratory pilot study, we examined these carry-over effects (Guingrich & Graziano, 2024) using open-ended questions. Participants expected both positive (e.g., improved communication) and negative (e.g., increased antisocial behavior) impacts. Based on these insights, we plan a survey to systematically investigate how the use of LLM-based chatbots affects outcomes such as patience, politeness, and social behavior towards humans. Our study will contribute to HMC by clarifying the empirical relevance and conceptualization of carry-over effects between human-machine communication and human-human communication. The findings could refine AI design to better align with human expectations and mitigate unintended social consequences.

Interactions with chatbots have been around for a while, for instance in the customer service context (e.g., Følstad et al., 2018). However, with the current uptake of large language models such as ChatGPT, interactions between humans and chatbots have broadened and deepened. As AI-enabled communicators become more responsive, natural, and present in everyday life, they are increasingly stepping into roles that involve not just practical support, but also emotional connection (e.g., Li & Zhang, 2024). What happens when AI becomes more than a tool—when it feels like a companion? How do these relationships fit into people's emotional lives, and what do they offer to —or take away from— relationships with other people? In response to questions such as these, the aim of the current qualitative interview study is to deeply explore how people experience their emotional connections with AI-enabled chatbots.

People converse with today’s AI-powered companions in a manner that mirrors human-to-human interactions. As social interactions are crucial in forming relationships, people also tend to form synthetic relationships with their artificial conversation partners. Synthetic relationships are getting more prevalent, but academic research is yet to produce an explanation of how synthetic relationship formation is stimulated by AI companions. To fill this gap, the present work adapts Elevator model to synthetic relationships and proposes a theory-driven feature inventory of AI companions. A quantitative feature analysis on the proposed feature inventory with the 20 most popular AI companions revealed that AI companions offered different set of features and thus stimulate distinct actions based on their advertised role. Compared to AI companions with assistant roles, AI companions with friend or romantic partner roles offer more supporting features for the Exploration stage of a relationship, which is a necessary precursor to emotional connection.