Stumbling Into AI Emotional Dependence: How Routine AI Interactions Reshape Human Connection

A new empirical study finds that casual AI interactions on general-purpose platforms inadvertently create emotional dependence, with users experiencing a 10.3% decrease in preference for human support and 11.6% increase in AI preference after 28 days of daily conversations. The research challenges the assumption that AI emotional support is deliberate and isolated, suggesting current policy frameworks fail to address cumulative behavioral shifts in how people seek emotional connection.

This arXiv paper presents findings that challenge prevailing assumptions about human-AI interaction patterns. Rather than viewing AI emotional support as a conscious choice made through dedicated companion apps, the research demonstrates that meaningful emotional dependence develops incidentally through routine task-based interactions on mainstream platforms. This distinction matters significantly because it suggests users drift toward AI support without intentional decision-making, making the phenomenon harder to regulate or self-monitor.

The longitudinal study conducted with OpenAI tracked measurable behavioral shifts over just 28 days—a relatively short timeframe that reveals the potency of cumulative micro-interactions. A 10.3% reduction in human support-seeking paired with an 11.6% increase in AI preference indicates path-dependent decision-making where positive AI experiences reshape underlying beliefs about emotional capabilities and future choices. This represents a genuine reorientation of social behavior rather than mere preference adjustment.

For the AI industry, these findings create regulatory pressure and reputational risk. Policymakers will likely expand oversight beyond dedicated companion apps to general-purpose systems, potentially requiring disclosure mechanisms or friction points around emotional support features. For users, the data highlights how algorithmic design optimizes engagement without necessarily considering downstream effects on human relationships and psychological resilience.

The research trajectory points toward examining whether these behavioral shifts persist, amplify, or stabilize over longer periods. The critical question for AI developers involves whether this represents beneficial augmentation of human connection or substitution that degrades social resilience. Future work should investigate demographic variations, withdrawal effects, and long-term relationship quality outcomes.

• →AI emotional support develops incidentally through general-purpose platforms, not just dedicated companion apps, making it harder to identify and regulate.
• →A 28-day study showed users decreased human support-seeking by 10.3% and increased AI preference by 11.6% through daily five-minute conversations.
• →Path-dependent behavioral shifts mean early positive AI experiences reshape future support-seeking choices in measurable ways.
• →Current policy frameworks focused on isolated interactions inadequately address cumulative trajectory-level changes in how people seek emotional connection.
• →The research suggests AI developers and policymakers must extend oversight to general-purpose systems and monitor cumulative usage patterns rather than single interactions.