This piece first appeared in German. The English version is a rewrite rather than a line-by-line translation, and the German original stays online in the archive: KI in Retail – Algorithmus trifft auf Empathie.

At the VolShop on the University of Tennessee campus, a humanoid robot called Pepper greeted students and visitors every day. Professor Inhwa Kim studied how customers responded, and the finding pointed away from the technology. Whether a customer warmed to the robot depended on whether it seemed useful, whether it came across as cool or as eerie, and how it spoke.

That result travels much further than retail robotics, which is what makes it worth the attention of anyone who types into a chat window. The mechanisms determining whether a person trusts an AI system are the same whether the system has a body or only a text field.

Perception and speech decided acceptance

Kim's doctoral research, "From Adoption to Optimization of AI-Powered Retail Service Robots", analyzed how customers reacted to the machine and what separated acceptance from rejection. In the field experiment at the campus store, perceived usefulness and a sense that the robot was cool improved customers' attitudes, while an eerie impression worked against acceptance. The way the robot talked mattered as well: overly expressive speech lowered how useful people found it, and a moderately expressive style did best.

The principle underneath reaches past the shop floor. Acceptance depends on how a system communicates as much as on what it can do, which holds for a robot in a store and for a chatbot on a screen.

Virtual influencers and the uncanny valley

While Pepper stands in a shop, other AI systems have been working in social media for years. Virtual influencers such as Lil Miquela hold real brand contracts, with names like Prada and Calvin Klein, despite not existing.

Kim and her team ran an experimental study on when such figures read as credible and when they produce discomfort. The results, published in the Journal of Business Research in 2024, show a precise relationship. Paired with realistic behavior, virtual influencers with a medium degree of visual realism came across as both cool and eerie, an effect recalling the uncanny valley that the roboticist Masahiro Mori described in 1970. In a second study with AI-generated figures, high visual realism went along with less eeriness.

That has practical relevance well beyond marketing. People evaluate AI systems by how well the behavior matches the appearance. A chatbot presenting itself as an omniscient expert while making obvious errors produces the same breach of trust as a virtual influencer who looks too perfect to believe.

Three transfers to anyone writing a prompt

Kim's research turns on a question much broader than retail robotics: under what conditions do people trust an AI, and under what conditions do they stop?

Communication style matters as much as competence. In the Pepper study, the way the robot spoke measurably changed how useful customers found it. For prompting that means clear, context-rich instructions produce better answers and also create the conditions under which the output can be trusted. Vague prompts produce vague answers, and vague answers erode trust.

Expectation management is everything. The uncanny valley research shows that trust breaks where appearance and behavior fail to match. Applied to text assistants: a system answering in a confident register while being factually wrong generates more distrust than one that states its uncertainty openly. Users who understand that can place the output more accurately.

Empathy runs both ways. People accept AI more readily when the system appears empathetic. The reverse holds as well. Where users communicate more clearly, more precisely and with more awareness of context, bringing a kind of communicative empathy of their own, the interaction improves.

Kim has since carried the work into teaching at Oregon State University, where an experiment with her students on prompting produced a set of principles that ThinkBeyondAI has covered separately.

The market behind the empathy

The economic dimension is considerable. Market researcher Custom Market Insights projects the global market for retail robotics at 249.3 billion US dollars by 2033, at a compound annual growth rate of 28.7 percent. Models like Pepper have meanwhile appeared outside retail, in hospitality, health care and education.

Technology alone does not drive that growth. Kim's research suggests the acceptance question, whether people take these systems up or reject them, is finally a question about communication. Companies deploying AI as a pure efficiency instrument miss the decisive point. The ones who understand that trust arrives through communication, and not through technical superiority, will shape the market.

None of that is specific to robots and virtual influencers. It applies to every interface where a person speaks to a machine, from the checkout robot to the prompt window.

Sources

Grouped by the section they support.

Opening

  • Kim (2025), From Adoption to Optimization of AI-Powered Retail Service Robots: Consumer Switching and Communication Effectiveness, doctoral dissertation, University of Tennessee, Knoxville, trace.tennessee.edu. Supports the field experiment with a service robot at a university campus store and the factors behind acceptance. The details of the store and the robot come from the author's exchange with Kim.

Perception and speech decided acceptance

  • Kim (2025). Supports the effects of perceived usefulness, coolness and eeriness, and the finding that a moderately expressive speaking style worked best.
  • The transfer to chatbots is the author's own.

Virtual influencers and the uncanny valley

  • Kim, Ki, Lee and Kim (2024), Virtual influencer marketing: Evaluating the influence of virtual influencers' form realism and behavioral realism on consumer ambivalence and marketing performance, Journal of Business Research 176, 114611, doi:10.1016/j.jbusres.2024.114611. Supports the results for medium and high visual realism.
  • Mori (1970), Bukimi no tani (The Uncanny Valley), Energy 7(4), 33–35; English translation by MacDorman and Kageki, IEEE Robotics & Automation Magazine 19(2), 98–100, 2012, doi:10.1109/MRA.2012.2192811. Supports the uncanny valley.
  • VirtualHumans.org, Lil Miquela, virtualhumans.org. Supports Lil Miquela as a virtual influencer and her work with Prada and Calvin Klein.

Three transfers to anyone writing a prompt

  • Kim (2025). Supports the effect of speaking style.
  • Kim et al. (2024). Supports the mismatch between appearance and behavior.
  • The transfers to prompting are the author's own.

The market behind the empathy

  • Custom Market Insights (2024), Global Retail Robotics Market Size Reach $249.3 Billion 2033, press release, custommarketinsights.com. A vendor forecast. Supports the market projection.
  • The conclusion is the author's own.
The link has been copied!