The Relational Capacity Framework: A Lens for AI-Mediated Relationality

Emily Tseng, Patrick Gage Kelley, Renee Shelby, Ashley M. WalkerarXiv 2026 · arXiv

This paper introduces the Relational Capacity Framework (RCF), a new conceptual model for understanding how general-purpose AI (like ChatGPT or Claude) mediates human-to-human relationships. By analyzing a 100-person study, the authors move beyond the simplistic "AI as a substitute for friends" narrative, demonstrating instead how people use AI as a metacognitive tool to supplement their own social skills, relational bandwidth, and contextual information.

TL;DR

The authors propose the Relational Capacity Framework (RCF) to explain how people use general-purpose AI to manage their human relationships. The framework identifies three pillars—skills, bandwidth, and information—and shows how AI can either expand or contract a person's relational capacity depending on their situational needs and metacognitive awareness.

What Problem Does This Paper Address?

While much research has focused on "purpose-built" AI companions (like Replika), there is a gap in understanding how people use general-purpose AI for socioemotional needs in their real-world relationships.

Existing theories often fall into a binary: either AI connects us or it isolates us (the "substitution hypothesis"). This paper argues that this binary is too narrow. It addresses the need for a unified framework that explains the massive heterogeneity in user practices—from a parent using AI to regulate their emotions during a toddler's meltdown to a manager using it to draft a sensitive email to a difficult employee.

Methods

The researchers conducted a 100-person mixed-methods study involving:

  1. Qualitative Surveys: Characterizing AI use for socioemotional needs over the past 30 days.
  2. Think-Aloud Interactions: Participants recorded themselves using their chosen AI to address a real-world relational hurdle.
  3. In-depth Interviews: 23 follow-up interviews to probe the "why" behind AI use and its place in the user's broader social ecosystem.
  4. Abductive Analysis: An iterative process of coding data and integrating it with established psychological and sociological frameworks to develop the RCF.

Datasets

The study utilized a newly created corpus of:

  • 100 qualitative survey responses.
  • 100 think-aloud interaction recordings.
  • 23 interview transcripts.

The participant pool was purposively sampled to ensure diversity in demographics and AI usage frequency.

Table 1: Participant Demographics (N = 100)

Key Insights and Ideas

  • Relationality as a Three-Pillar Capacity: The authors argue that our ability to relate to others depends on Social Skills (e.g., communication, interpretation), Relational Bandwidth (internal energy and external support), and Contextual Information (understanding the norms or facts of a situation).
  • AI as a Flexible Subsidy: AI doesn't just "provide advice"; it serves as a resource that users "spend" to bolster whichever of the three pillars is currently depleted.
  • Relational Metacognition: The RCF frames relational management as a metacognitive process. Users "monitor" their relational world for deficits and use AI to "control" or execute corrective actions.
  • Expansion vs. Contraction: AI use is not inherently good or bad. It can expand capacity (e.g., by providing a low-stakes rehearsal environment) or contract it (e.g., by encouraging rumination or providing inaccurate, generic advice that masks harmful dynamics).
Table 2: Domains of Relationality Management and Corresponding Skill Types

Key Results

The paper identifies eight "Relationality States" (RS1–RS8) based on different configurations of high/low skills, bandwidth, and information. These states are dynamic; a person might be in a high-capacity state (RS8) one day and a low-bandwidth state (RS7) the next due to a life crisis.

Table 3: States of Relational Capacity

Qualitative Findings include: * Skill Mastery: Users leverage AI to calibrate tone (e.g., "sounding more professional") or practice difficult conversations. * Bandwidth Management: AI is used for "micro-doses of decompression" or to quickly handle logistical relational tasks when the user is situationally exhausted. * Information Seeking: AI acts as a synthesizer for complex social contexts, such as navigating healthcare bureaucracies or understanding a partner's ADHD.

Limitations and Future Work

  • Self-Reporting Bias: The study relies on participants' own accounts of their skills and needs, which may be subject to social desirability bias.
  • Cross-Sectional Nature: The study provides a snapshot. Because relational states are dynamic, the authors call for longitudinal research to see how AI use affects relational capacity over months or years.
  • Design Implications: The authors suggest that AI systems should be designed to scaffold (teach and encourage reflection) rather than bypass (simply providing a script to copy-paste) human evaluative processes.
  • Erosion of Friction Tolerance: A key concern for future work is whether "frictionless" AI interactions make the inevitable messiness of human relationships feel less tolerable by comparison.

Summary updated

Abstract

A growing literature examines the impact of engaging with AI chatbots on our relationality: our ability to form relationships with others, maintain them through life's changes, and exit when needed. Currently, substantial work focuses on purpose-built AI companions (e.g. Replika, Character.AI), with less work investigating socioemotional uses of the more widely-used general-purpose chatbots. In this paper, we present a 100-person investigation into how people engage general-purpose AI for socioemotional needs. Participants' AI uses varied, from practicing specific social skills to seeking additional bandwidth and information as they navigated their social contexts. To explain this heterogeneity, we provide 8 states that describe how AI expands or contracts relationality for people with different levels of skills, bandwidth, and information. We then present a unified conceptual model for the role of technology in mediating human relationality, the Relational Capacity Framework, and outline future research designing appropriate roles for AI in human connection.
BibTeX
@misc{tseng2026relational,
  author = {Emily Tseng and Patrick Gage Kelley and Renee Shelby and Ashley M. Walker},
  title = {{The Relational Capacity Framework: A Lens for AI-Mediated Relationality}},
  year = {2026},
  eprint = {2611.08877},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.08877},
}
Figure