Scaffolded Vulnerability: Chatbot-Mediated Reciprocal Self-Disclosure and Need-Supportive Interaction in Couples

Paper Title

Scaffolded Vulnerability: Chatbot-Mediated Reciprocal Self-Disclosure and Need-Supportive Interaction in Couples

Publication Info

• Topic area: AI-mediated communication for fostering intimacy in romantic relationships.
• Keywords: Chatbots, self-disclosure, Self-Determination Theory, intimacy, couples, need support, reciprocal care, AI-mediated communication, relational scaffolding, well-being.

Background and Problem

• Problem / challenge: Existing technologies for fostering intimacy often fail to balance autonomy, competence, and relatedness support. Many systems either focus on passive sensing or impose constraints, leaving interactions fragile or compromising user autonomy.
• Significance: Intimacy and emotional connection are critical for relationship quality and individual well-being. Addressing gaps in need-supportive communication can enhance relational and personal outcomes.
• Motivation and related work: Prior work on relatedness technologies (e.g., biosignal sharing, telepresence) has struggled to sustain reciprocity or provide guidance for emotionally rich topics. Chatbots have shown potential as mediators but lack grounding in psychological frameworks like Self-Determination Theory (SDT). This study builds on these gaps by operationalizing SDT to scaffold reciprocal care in couples.

Solution

• Proposed approach: A chatbot employing a dual-layer scaffolding framework grounded in SDT to facilitate reciprocal self-disclosure and need-supportive interactions between romantic partners.
• Novelty:
  1. Empirical evidence that chatbot-mediated scaffolding fosters relational closeness and reciprocal care.
  2. Introduction of a dual-layer scaffolding framework combining enabling (instrumental) and mediating (relational) affordances.
  3. Design insights for balancing structural guidance with user autonomy in AI-mediated communication.
• Procedure and key techniques:
  • Layer 1 (Enabling Affordances): Provides instrumental support (e.g., rationales, choices, warm tone) to create a safe environment for self-disclosure.
  • Layer 2 (Mediating Affordances): Deploys structured reflection prompts to scaffold partner-provided autonomy, competence, and relatedness support.
  • Chatbot prompts were adapted from the "36 Questions" paradigm and implemented using GPT-4.1 on Telegram.
  • Three experimental conditions: Partner Support (PS: full scaffolding), Direct Support (DS: enabling only), and Basic Prompt (BP: baseline with no scaffolding).

Results

• Concrete findings:
  • PS condition yielded the longest conversations (M = 107.7 mins) and highest engagement (145 messages, 1077 words).
  • Self-disclosure depth was significantly higher in PS and DS compared to BP (e.g., informational disclosure: PS = 2.28, DS = 1.97, BP = 1.37).
  • Relatedness support was highest in PS (1.50) compared to DS (1.24) and BP (0.51).
  • Controlled motivation decreased across all conditions, while subjective vitality improved only in scaffolded conditions (PS, DS).
• Advantage over baselines:
  • PS uniquely increased perceived closeness (IOS: Pre = 5.17, Post = 6.00) and elicited partner-provided need support through reflection prompts.
  • DS and PS outperformed BP in fostering detailed self-disclosure and relatedness support.
• Experiments / evaluation:
  • Randomized study with 36 couples (N = 72).
  • Metrics: chat duration, message/word counts, self-disclosure depth, partner-provided need support, and well-being measures (e.g., IOS, vitality).
  • Mixed-methods analysis combining quantitative metrics and qualitative coding of chat logs.
• Limitations and future work:
  • Single-session design limits claims about long-term behavioral change.
  • Sample skewed toward younger, short-term relationships; findings may not generalize to older or long-term couples.
  • Future work should explore longitudinal impacts and ecological validity in naturalistic settings.

Summary

This study introduces a chatbot leveraging a dual-layer scaffolding framework grounded in Self-Determination Theory to foster reciprocal self-disclosure and need-supportive interactions in couples. The PS condition demonstrated significant improvements in relational closeness, self-disclosure depth, and partner-provided support compared to baseline. The chatbot's structured prompts effectively transformed individual sharing into reciprocal care, addressing gaps in prior relatedness technologies. While the findings highlight the potential of AI-mediated scaffolding for enhancing intimacy, future research should explore long-term impacts and broader demographic applicability.