Can AI be a Social Buffer? Investigating the Effect of AI-assisted Cognitive Reappraisal and Narrative Perspectives on Managing Difficult Workplace Conversations over Email

Paper Title

Can AI be a Social Buffer? Investigating the Effect of AI-assisted Cognitive Reappraisal and Narrative Perspectives on Managing Difficult Workplace Conversations over Email

Publication Info

• Topic area: AI-mediated communication in workplace conflict resolution
• Keywords: AI-assisted reframing, cognitive reappraisal, workplace communication, narrative perspectives, conflict management, emotional regulation, large language models, email communication, social buffering, trust in AI

Background and Problem

• Problem / challenge: Difficult workplace conversations over email often exacerbate negative emotions due to the absence of nonverbal cues. While tools exist to help senders refine tone, there is limited support for receivers to process emotionally charged content.
• Significance: Managing workplace conflicts effectively is critical for employee well-being, team performance, and organizational success. Supporting receivers in regulating emotions during difficult conversations could improve outcomes.
• Motivation and related work: Prior research has explored AI tools for cognitive reappraisal and tone adjustment but has not addressed how AI can assist receivers in reframing difficult messages without distorting the sender's intent. This study builds on existing findings to explore AI's role as a mediator in text-based communication.

Solution

• Proposed approach: AI-assisted cognitive reappraisal using large language models (LLMs) to reframe difficult emails in positive or neutral tones, presented in first- or third-person narrative perspectives.
• Novelty:
  1. Empirical evaluation of AI-assisted reframing's impact on receivers' emotions, behaviors, and attitudes.
  2. Comparison of positive vs. neutral reframing and first- vs. third-person perspectives.
  3. Focus on recipient-centered AI-mediated communication for workplace conflict resolution.
• Procedure and key techniques:
  • Conducted a controlled experiment with 132 participants using five conditions: positive-first person, positive-third person, neutral-first person, neutral-third person, and a control (no AI reframing).
  • Measured emotional responses, conflict management behaviors, linguistic features, and attitudes toward AI.
  • Analyzed linguistic data using LIWC and coded conflict management styles.

Results

• Concrete findings:
  • Positive reframing significantly reduced negative emotions (both high and low arousal) compared to neutral reframing or no reframing (p < .001).
  • Positive reframing was perceived as more helpful (mean difference: 0.69, p < .001) and trustworthy (p < .01) than neutral reframing.
  • No significant effect on positive emotions or conflict management behaviors.
  • Positive reframing led to fewer "power"-related words in email interpretations (p < .05), indicating reduced perceptions of dominance.
• Advantage over baselines:
  • Positive reframing outperformed neutral reframing and the control condition in reducing negative emotions and enhancing perceived helpfulness and trust in AI.
• Experiments / evaluation:
  • Participants (n=132, mean age 36.7) were randomly assigned to one of five conditions.
  • Scenarios simulated workplace conflicts (e.g., poor performance, layoffs).
  • Metrics included emotional valence/arousal, conflict management styles, linguistic analysis, and trust/helpfulness ratings.
• Limitations and future work:
  • Limited to single email exchanges; future studies should explore multi-round interactions.
  • Scenarios involved hierarchical power dynamics; further research needed for peer-level conflicts.
  • Lack of significant behavior change suggests the need for integrating explicit conflict management strategies in AI prompts.

Summary

This study demonstrates that AI-assisted positive reframing can reduce negative emotions and enhance perceived helpfulness and trust in workplace email conflicts, without distorting the sender's intent. While no significant effects were observed on conflict management behaviors, positive reframing reduced perceptions of dominance in email interpretations. These findings highlight AI's potential as a social buffer in difficult conversations but also underscore the need for further research on long-term effects, peer-level dynamics, and integrating behavioral guidance into AI-mediated communication.