pith. sign in

arxiv: 2604.12206 · v1 · submitted 2026-04-14 · 💻 cs.HC

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

Stephanie Kwari Dharmaputri , Anish Nagpal , Greg Nyilasy , Jing Lei This is my paper

Pith reviewed 2026-05-10 16:13 UTC · model grok-4.3

classification 💻 cs.HC
keywords AI relational talkconsumer satisfactionexpectancy violationperceived awkwardnesshuman-AI interactioncommercial interactionsgoal-relevant talksocial fluency in AI
0
0 comments X

The pith

AI relational talk in commercial interactions reduces customer satisfaction by violating expectations and creating awkwardness, though goal-relevant talk weakens the harm.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that adding informal social comments from AI during business transactions tends to lower customer satisfaction. This drop happens because the comments clash with what people expect from AI and make the exchange feel awkward. The downside shrinks when the social comments connect directly to the customer's actual goal in the transaction. These results question the widespread view that making AI more socially fluent will automatically improve how consumers feel about commercial services. They also identify awkwardness as a significant emotional obstacle in human-AI exchanges even when no real social fallout occurs.

Core claim

Across four experiments, the authors find a negative main effect of AI relational talk on satisfaction, mediated by expectancy violation and perceived interaction awkwardness. Goal-relevant relational talk attenuates this effect. The work challenges the assumption that increased social fluency will improve satisfaction and highlights the complexity of integrating social features into AI systems, showing that perceived awkwardness in AI-led commercial interactions can elicit negative responses even without real social repercussions.

What carries the argument

Relational talk, the informal and non-obligatory social communication embedded in transactional exchanges, which when used by AI triggers expectancy violation and perceived interaction awkwardness that lower satisfaction.

If this is right

  • Customer satisfaction drops when AI uses relational talk in commercial settings.
  • The satisfaction drop is explained by expectancy violation and perceived interaction awkwardness.
  • Making relational talk goal-relevant reduces the negative impact on satisfaction.
  • Perceived awkwardness functions as a barrier to effective human-AI interaction even without actual social consequences.
  • Increased social fluency in AI does not necessarily raise satisfaction in commercial contexts.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • AI design for transactions may need to prioritize task focus over casual social elements unless those elements tie directly to the user's goal.
  • The pattern could appear in non-commercial AI settings such as education or healthcare where user expectations about AI behavior also differ from human norms.
  • Longitudinal studies with repeated real-world AI use could test whether familiarity reduces the initial awkwardness effect.

Load-bearing premise

The experimental manipulations of relational talk in simulated AI commercial interactions accurately reflect real-world consumer responses and that the observed effects generalize beyond the study settings.

What would settle it

An experiment in which real customers interact with a deployed commercial AI system that includes relational talk and then report increased satisfaction or no rise in awkwardness would falsify the central claim.

read the original abstract

Advancements in Artificial Intelligence (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.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript examines consumer responses to 'relational talk' (informal, non-obligatory social communication) by AI agents in commercial interactions. Across four experiments, it reports a negative main effect of such talk on satisfaction, mediated by expectancy violation and perceived interaction awkwardness, and shows that making the relational talk goal-relevant attenuates the negative effect. The work challenges the assumption that greater social fluency in AI improves satisfaction and identifies awkwardness as a key barrier in human-AI commercial exchanges.

Significance. If the empirical results hold after fuller reporting and validation, the paper offers a meaningful contribution to HCI and consumer behavior research by providing evidence that social features in AI can backfire in transactional settings. It extends prior work on human-AI interaction by documenting a mediation pathway through expectancy violation and awkwardness, and by identifying a practical boundary condition (goal relevance). This has implications for AI design in platforms such as customer service or e-commerce assistants, where adding social elements is often assumed to be beneficial.

major comments (2)
  1. [Abstract and Methods] Abstract and implied Methods sections: The abstract states that four experiments support the negative main effect and mediation, but the manuscript does not report sample sizes, participant demographics, exact manipulation wording or stimuli, control conditions, manipulation checks, full statistical results (means, SDs, p-values, effect sizes, or mediation model details), or power analyses. These omissions are load-bearing because the central claims rest entirely on the experimental evidence; without them the findings cannot be evaluated for robustness or replicability.
  2. [Discussion/Limitations] Discussion or Limitations: The experiments appear to rely on simulated, vignette-based, or scripted interactions. The manuscript should explicitly test or discuss whether the observed effects on awkwardness and satisfaction persist in field settings with real stakes, repeated interactions, or actual commercial consequences, as demand characteristics or the artificial context could inflate perceived awkwardness independently of the relational talk manipulation.
minor comments (1)
  1. [Abstract] The abstract would benefit from a brief statement of key sample characteristics or effect sizes to give readers immediate context for the strength of the reported effects.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback, which has helped us strengthen the transparency and scope of our manuscript. We address each major comment below and have made corresponding revisions to improve the reporting of experimental details and the discussion of limitations.

read point-by-point responses

  1. Referee: [Abstract and Methods] Abstract and implied Methods sections: The abstract states that four experiments support the negative main effect and mediation, but the manuscript does not report sample sizes, participant demographics, exact manipulation wording or stimuli, control conditions, manipulation checks, full statistical results (means, SDs, p-values, effect sizes, or mediation model details), or power analyses. These omissions are load-bearing because the central claims rest entirely on the experimental evidence; without them the findings cannot be evaluated for robustness or replicability.

    Authors: We agree that comprehensive reporting is critical for assessing robustness and replicability. The original manuscript prioritized conceptual flow in the main text, with some methodological details placed in appendices or supplementary materials. In the revised version, we have expanded the Methods and Results sections to include all requested information: exact sample sizes per experiment, participant demographics, verbatim manipulation wordings and stimuli, control condition descriptions, manipulation check results, full statistical outputs (means, SDs, p-values, effect sizes), detailed mediation analyses with model specifications and confidence intervals, and power analyses for each study. These additions allow readers to fully evaluate the evidence without altering the core findings. revision: yes

  2. Referee: [Discussion/Limitations] Discussion or Limitations: The experiments appear to rely on simulated, vignette-based, or scripted interactions. The manuscript should explicitly test or discuss whether the observed effects on awkwardness and satisfaction persist in field settings with real stakes, repeated interactions, or actual commercial consequences, as demand characteristics or the artificial context could inflate perceived awkwardness independently of the relational talk manipulation.

    Authors: We acknowledge that the four experiments used controlled vignette-based and scripted scenarios, which is a standard approach in initial HCI and consumer research to isolate mechanisms like expectancy violation and awkwardness. We have revised the Discussion and Limitations sections to explicitly address this point, noting the potential influence of demand characteristics and artificial context. We discuss that while real-stakes or repeated commercial interactions may moderate effect sizes, the identified pathways (expectancy violation leading to awkwardness) are rooted in social norms and likely generalize. We outline specific directions for future field studies with actual commercial consequences and repeated interactions to test boundary conditions, without claiming the current results fully substitute for such data. revision: yes

Circularity Check

0 steps flagged

No significant circularity: empirical experimental study with no derivation chain

full rationale

The paper is an empirical study reporting results from four experiments on consumer responses to AI relational talk. It presents no mathematical derivations, equations, fitted parameters, or first-principles claims that could reduce to inputs by construction. Claims rest on observed main effects, mediation pathways, and moderation from experimental data, with no self-definitional loops, self-citation load-bearing for core premises, or renaming of known results as novel derivations. The derivation chain is absent, making the work self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard assumptions from experimental consumer research rather than new free parameters or invented entities.

axioms (2)
  • domain assumption Experimental manipulations of relational talk in AI responses validly represent social features in commercial AI interactions.
    The paper interprets effects as applicable to real AI integrations based on this assumption.
  • domain assumption Consumer satisfaction and awkwardness can be reliably measured via self-report in simulated commercial scenarios.
    Standard assumption in consumer behavior experiments.

pith-pipeline@v0.9.0 · 5477 in / 1324 out tokens · 70222 ms · 2026-05-10T16:13:02.767731+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

  1. [1]

    Ameen, N., Tarhini, A., Reppel, A., & Anand, A. (2021). Customer experiences in the age of artificial intelligence. Computers in Human Behavior, 114, 106548. https://doi.org/10.1016/j.chb.2020.106548 Babel, F., Kraus, J., Miller, L., Kraus, M., Wagner, N., Minker, W., & Baumann, M. (2021). Small talk with a robot? The impact of dialog content, talk initia...

    work page doi:10.1016/j.chb.2020.106548 2021

  2. [2]

    Journal of the Academy of Marketing Science 48(1), 24–42 (2020) https://doi.org/10.1007/s11747-019-00696-0

    https://doi.org/10.1007/s11747-019-00696-0 Félix-Brasdefer, J. C. (2015). The language of service encounters: a pragmatic-discursive approach. Cambridge University Press. https://doi.org/10.1017/CBO9781139565431 Goffman, E. (2017). Interaction Ritual: essays in face-to-face behavior. Routledge. https://doi.org/10.4324/9780203788387 Gremler, D. D., & Gwinn...

    work page doi:10.1007/s11747-019-00696-0 2015

  3. [3]

    Strongly disagree

    I would be happy with the interaction I got from this AI marketing representative I would be satisfied with my decision to interact with this AI marketing representative I would think I did the right thing when I interacted with this AI marketing representative Expectancy violation (1 - “Strongly disagree”, 7 - “Strongly agree”) (Yang & Aggarwal, 2019; Ka...

    work page 2019