Socially Fluent, Socially Awkward: Artificial Intelligence Relational Talk Backfires in Commercial Interactions

Abstract: Advancements in AI technologies' social fluency are being integrated into commercial interactions. As tools such as OpenAI's assistant are integrated into platforms such as Shopify, Klarna, and Visa, understanding consumer responses to AI social features become essential. One such feature is relational talk, an informal and non-obligatory social communication embedded in transactional exchanges. Across four experiments, we find: 1) a negative main effect of AI relational talk on satisfaction, mediated by expectancy violation and perceived interaction awkwardness, and 2) goal-relevant relational talk to attenuate this effect. This paper extends the literature by challenging the assumption that increased social fluency will improve satisfaction, and highlights the complexity of integrating social features into AI systems. It also identifies awkwardness as a key emotional response and barrier to effective human-AI interaction, showing that even in the absence of real social repercussions, perceived awkwardness in AI-led commercial interactions can elicit negative responses.

• The paper reveals that relational talk in commercial AI significantly reduces user satisfaction by triggering expectancy violations and perceived awkwardness.
• It employs four robust experiments with mediation analyses and validated scales to rigorously measure the emotional and behavioral outcomes across diverse service domains.
• The study finds that aligning small talk with transactional goals can mitigate negative effects, highlighting the importance of contextual congruence in AI design.

AI Relational Talk in Commercial Interactions: Expectancy Violations and the Paradox of Social Fluency

Introduction

The proliferation of AI-driven commercial interfaces, such as chatbots integrated into ecommerce and financial platforms, has accelerated an industry-wide push towards anthropomorphic and socially fluent digital agents. However, the persistent assumption that enhancing the social fluency of AI will necessarily improve consumer satisfaction in commercial contexts lacks rigorous empirical substantiation. This paper, "Socially Fluent, Socially Awkward: Artificial Intelligence Relational Talk Backfires in Commercial Interactions" (2604.12206), systematically examines the impact of AI relational talk—non-task-oriented, informal conversational behavior—on consumer satisfaction, unpacking the mediation through expectancy violation and perceived interaction awkwardness across four well-powered experimental studies.

Theoretical Framework and Research Design

The central argument is grounded in Expectancy Violation Theory, leveraging the observation that consumers typically construe commercial AI as task-centric and efficient, rather than socially expressive. This theory predicts negative affective outcomes when interactions deviate from this expectation space. The authors uniquely operationalize relational talk within AI dialogues and measure outcome variables using validated multi-item scales.

Methodologically, the work employs four between-subject experiments crossing relational talk and goal relevance manipulations, sampling participants primarily via Prolific. The studies encompass interventions across retail, online banking, and web assistance domains, consistently operationalizing relational talk in either goal-relevant or goal-irrelevant forms. Covariates such as prior AI experience and general attitudes toward AI were factored through rigorous mediation analyses using bootstrap resampling.

Empirical Results

Main Effects: Relational Talk Reduces Satisfaction

Contrary to prevailing intuition in applied machine learning and HCI design, the introduction of relational talk by commercial AI agents produced a robust, significant negative effect on user satisfaction. This was demonstrated in Study 1 (Cohen's d = –0.62, $p < .001$) and replicated with comparable magnitude and reliability across additional studies. Notably, participants interacting with AI agents displaying small talk reported lower satisfaction than those engaging in strictly task-oriented exchanges. The effect sizes were substantial for this genre of service research.

Mechanism: Expectancy Violation and Awkwardness

A serial mediation model revealed two critical mediators: expectancy violation and perceived interaction awkwardness. Experimental analyses (e.g., Study 2, $B_{\text{mediation}} = -0.26$, 95% CI [–0.35, –0.19]) evidenced that relational talk significantly increased both expectancy violation (d > 1.0) and perceived awkwardness (d ≈ 0.98), both of which in turn decreased satisfaction. Critically, awkwardness emerges as an affective construct distinct from frustration or anger, specifically capturing the discomfort from subtle social norm violations in AI-led contexts.

Consequential and Goal-Relevant Moderation

Study 3 extends findings into consequential decision-making by tying users' point allocation (with believed monetary value) to the AI agent’s future development. Here, again, relational talk decreased the incentive to reward the agent, an effect mediated by awkwardness.

Study 4 identifies goal relevance as a statistically significant moderator. When relational talk was congruent with the transaction goal (e.g., discussing coffee preferences during a coffee machine inquiry), expectancy violations and awkwardness were attenuated, and overall satisfaction increased (Cohen’s d = 0.74 for satisfaction between goal-relevant and goal-irrelevant relational talk). Serial mediation models confirmed that full mediation via expectancy violation and awkwardness explained the boundary effects of goal alignment.

Contradictory and Bold Claims

The findings challenge a widespread industry and academic claim that social fluency—mimicking “small talk” and informal banter—should universally enhance user satisfaction with AI agents. Instead, this research demonstrates the inverse: Socially expressive AI features can backfire, producing expectancy violations and awkwardness, ultimately diminishing user satisfaction in commercial transactions. This result is at odds with much of the literature on social robotics and human social interaction, where relational talk typically enhances rapport and satisfaction.

Implications for AI Interaction Design

Practically, the results indicate that the uncontextualized introduction of relational talk is a liability in commercial AI deployments. The work demonstrates the importance of contextual alignment between relational talk and user goals to mitigate negative affective outcomes. Designers of conversational agents should avoid non-instrumental social flourishes unless such behaviors are directly instrumental toward the human's transactional objective.

From a theoretical perspective, the identification of awkwardness as a key emotional barrier in human-AI interaction is notable and has not received adequate attention in prior studies on user experience with AI. The mediation role of expectancy violations suggests that consumer mental models for AI remain fundamentally distinct from human-human interactions, reinforcing the need for careful boundary management in AI anthropomorphism.

Limitations and Directions for Future Research

The controlled experimental design employing scenario-based interactions and standardized dialogue may limit ecological validity, especially as actual consumer-AI exchanges become increasingly dynamic. Future research should explore real-world commercial deployments, examine variable-length interactions, more nuanced prosodic markers, and user agency (e.g., toggling social features on/off). Alternative interventions, such as strategic use of humor or responsive personalization, could potentially mitigate the negative effects delineated here.

Furthermore, the role of individual differences (e.g., cross-cultural variation in social norm expectations, technology experience) warrants deeper analysis, as does extending the model into richer multimodal conversational settings (audio, video) and high-stakes service environments.

Conclusion

This paper delivers a rigorous empirical challenge to assumptions that enhancing social fluency in commercial AI agents is unambiguously desirable. AI-led relational talk in commercial exchange contexts provokes expectancy violation and awkwardness, undermining user satisfaction—unless carefully aligned with user goals. These findings delineate critical design constraints for AI sociality in transactional settings and elucidate the affective complexity of human-AI interaction dynamics. Such insights provide foundational guidance for both designers and theorists concerned with socially intelligent AI in marketplace contexts.