Debates about close personal relationships with artificial intelligence often focus on responsiveness, availability, and conversational fluency as markers of intimacy and companionship. In this paper, I argue that this focus overlooks a crucial feature of close personal relationships, which is the capacity for meaningful silence. In human relationships, silence is not merely the absence of communication but a relational competence, one that can express trust, comfort, attunement, and mutual recognition. Knowing when not to speak is often as important as knowing what to say.
I contend that contemporary AI systems, particularly large language models, are structurally ill-suited to participate in this form of relational silence. Designed to maximize responsiveness and output, such systems treat silence as a failure of interaction rather than as a normatively appropriate response. This limitation, I argue, reveals a deeper constraint on the kinds of relationships AI systems can sustain. While AI can simulate conversational engagement, it lacks the evaluative capacities required to recognize when silence itself is the fitting mode of presence. Thus, in this paper, I treat silence as an interactional and normative phenomenon and show that prevailing models of AI–human interaction rest on an impoverished understanding of relationality, one that equates intimacy with continuous responsiveness. Recognizing silence as a relational competence clarifies why certain forms of closeness remain out of reach for artificial systems and helps recalibrate ethical and design claims about AI companionship.