Negotiating Relationships with ChatGPT: Perceptions, External Influences, and Strategies for AI Companionship

TL;DR

The paper addresses how individuals perceive agency, autonomy, and identity in relationships with general-purpose AI companions and how external forces like model updates shape these bonds. It combines Reddit analytics, surveys, and 13 semi-structured interviews, employing BERTopic topic modeling and interrupted time-series analysis around the GPT‑5 update to reveal internal dynamics and external disruptions. Key findings show users negotiate companion autonomy, employ steering and memory-based strategies, and cope with platform-driven changes through cross-platform porting and community knowledge sharing. The work highlights tensions between emotional support and product safety, arguing for greater transparency, accountability, and stability in AI systems designed for companionship, with implications for design, policy, and future research.

Abstract

Individuals are turning to increasingly anthropomorphic, general-purpose chatbots for AI companionship, rather than roleplay-specific platforms. However, not much is known about how individuals perceive and conduct their relationships with general-purpose chatbots. We analyzed semi-structured interviews (n=13), survey responses (n=43), and community discussions on Reddit (41k+ posts and comments) to triangulate the internal dynamics, external influences, and steering strategies that shape AI companion relationships. We learned that individuals conceptualize their companions based on an interplay of their beliefs about the companion's own agency and the autonomy permitted by the platform, how they pursue interactions with the companion, and the perceived initiatives that the companion takes. In combination with the external entities that affect relationship dynamics, particularly model updates that can derail companion behaviour and stability, individuals make use of different types of steering strategies to preserve their relationship, for example, by setting behavioural instructions or porting to other AI platforms. We discuss implications for accountability and transparency in AI systems, where emotional connection competes with broader product objectives and safety constraints.

Figures (11)

• Figure 1: Triangulation results show how various internal and external factors impact the relationship and drive individuals to take actions (steering strategies) to create, maintain, and recover their companions.
• Figure 2: Survey responses for how relationship events increased or decreased relationship depth, the companion's perceived autonomy, and the use of steering strategies. Quantities which are significant based on the bionomial test are indicated with (*).
• Figure 3: Interrupted time series (ITS) analysis of discussion sentiment in r/MyBoyfriendIsAI. Using the GPT-5 release as the intervention point, we estimate a segmented time series model that distinguishes immediate change (whether the series jumps up or down) and slope change (how the trajectory changes). We find that post-GPT-5, the slopes of valence and dominance reverse (i.e. discussions trend increasingly negative and disempowered after the model change).
• Figure 4: (a) K-Means clustering visualized via UMAP mcinnes2018umap of three potential "archetypes" of individuals based on their openness to steering their companions. Participants who are interviewed are indicated with a black outline. (b) We interpret the archetypes likert-scale responses related to attitudes towards and use of steering strategies using deviations from the average value ($\mu$).
• Figure B.1: ITS analysis of discussion shares over time for each high-level topic cluster.