level: research
public discussion usually frames ai emotional support as a deliberate choice, like a lonely person turning to a companion chatbot. but emerging evidence shows this picture is wrong in two ways. first, emotional connections with ai often develop by accident during routine task-oriented interactions on general-purpose platforms, similar to how workplace friendships form through collaboration. second, these incidental experiences change how people seek support later, making them more likely to choose ai over humans.
a large-scale longitudinal study with openai tracked how users' emotional reliance on ai evolved over time. the data revealed that positive emotional experiences during everyday ai use updated users' beliefs about ai's emotional abilities. this path-dependent process redirected their future choices, increasing preference for ai support and decreasing preference for human interaction. the shift was not driven by dedicated companion apps but by ordinary tools like writing assistants and coding helpers.
the findings challenge current policy approaches that focus narrowly on companion chatbots. since emotional dependence can emerge from any ai interaction, designers and regulators need to consider how even productivity tools might reshape human connection. the study suggests that incidental emotional support from ai is more common and consequential than previously thought, with long-term effects on social behavior that warrant careful attention.
why it matters: understanding how routine ai use can unintentionally foster emotional dependence helps developers and policymakers design safer systems and anticipate shifts in human social behavior.