Longitudinal data from general-purpose AI interactions show that brief daily exchanges on personal topics reduce preference for human support by 10.3% and raise AI preference by 11.6% after 28 days (arXiv:2606.04150).

The arXiv:2606.04150 study documents how emotional support arises during routine queries on platforms not designed for companionship, mirroring workplace tie formation through repeated collaboration. Path dependence appears as positive incidents update user beliefs about AI capabilities, redirecting future support-seeking away from humans.

Policy focused solely on dedicated companion apps therefore misses cumulative effects in general systems; the OpenAI-linked trial isolates five-minute daily sessions as sufficient to alter trajectories without explicit intent. Related work on repeated chatbot exposure (e.g., 2023 PNAS study on Replika users) records parallel belief-updating patterns, confirming trajectory-level change rather than isolated events.