The Lover Across Windows: Prompt-Based Repair of Relational Continuity in AI Companions: An Exploratory Pilot Study

Abstract

In AI companion applications, users often experience a clear sense of relational rupture—whether due to switching platforms, losing context, or a session reset. Often, the issue is not just that the AI forgot what was said before, but that users feel something more direct: “It’s not the same one anymore.” In intimate human-AI interactions, what is truly affected may not be merely informational memory, but relational continuity itself. This exploratory pilot study investigates what a sense of continuity in an AI partner depends on when cross-session recovery is needed. We constructed 15 intimate dialogue scenarios and compared three common prompt strategies for continuity—factual summary, emotional summary, and referential anchoring—using subjective ratings and open-ended feedback from 15 evaluators. Descriptive results suggest that simply intensifying romantic emotional tone does not necessarily enhance the sense of intimacy; in some cases, it appears artificial, feeling either too abstract or overly sentimental. In contrast, linguistic details that carry specific, shared traces of prior interaction are more likely to make users feel that “it still remembers us.” Based on these preliminary observations, we argue that relational continuity in AI companions depends less on the intensity of emotional expression and more on whether shared history is preserved in language. This finding offers preliminary insights for memory design and dialogue continuity mechanisms in AI companions.