Shoppers in five experiments were less likely to leave early when they dealt with AI retail assistants rather than human shop staff. The surprising mechanism was not speed or product knowledge. Researchers found that human assistants made participants feel more judged and intruded upon, while robot assistants weakened that social pressure.
AI retail assistants changed the feeling of being watched
The study, published in the Journal of Business Research, examined a familiar but rarely measured part of shopping: the discomfort of being approached by somebody who appears to be evaluating what you are doing. The researchers describe this as social-evaluative judgement. When shoppers believed a human retail assistant was observing them, that judgement increased perceived intrusiveness and made premature departure more likely.
Across five experimental studies in fashion and skincare settings, the authors compared human retail assistants with artificial-intelligence retail assistants, or AIRAs. Their proposed pathway was consistent across the work: the type of assistant affected perceived judgement, perceived judgement affected intrusiveness, and intrusiveness affected the urge to leave the store before the shopper otherwise would.
The PolyU institutional research record identifies the work as a September 2026 peer-reviewed Journal of Business Research article. Its abstract describes the same five fashion and skincare studies, the judgement-to-intrusiveness pathway and the moderating roles of smiling and shopping-goal clarity.
That is a different argument from the usual case for retail automation. A robot is often justified because it can work long hours, answer routine questions or reduce staffing costs. Here the potential advantage is psychological. The machine can feel less socially demanding precisely because the shopper does not experience it as another person forming an opinion about them.
The robot advantage was not universal
The paper does not conclude that shoppers always prefer machines or that a robot should replace every human assistant. Two boundary conditions changed the effect: whether the assistant smiled and how clear the shopper was about what they wanted. Those details matter because intrusiveness is not a fixed property of a technology. It emerges from an interaction between the assistant, the customer and the situation.
A shopper who knows exactly which product they want may value fast, direct human help. Somebody browsing without a firm goal may experience the same approach as pressure. Likewise, a smile can change the social meaning of an interaction. The experiments therefore support a more nuanced conclusion than ‘robots are friendlier than people’.
The results also come from controlled studies, not a nationwide trial in live shops measuring purchases, repeat visits or staff costs. The paper supports a causal pathway within the experimental designs, but it cannot establish that installing robots in real stores will automatically increase sales or satisfaction. Real retail environments contain queues, children, noise, complex questions, staff relationships and many other influences.
Why judgement can matter more than helpfulness
Retailers often train staff to approach quickly because unattended shoppers may need help. The problem is that the same behaviour can be interpreted in two ways. One customer sees attentive service. Another sees surveillance, sales pressure or an implied judgement about what they can afford, how they look or whether they belong in the shop.
That mechanism belongs to a wider question about how AI changes human decisions through presentation and interaction. LiveAIWire has reported that AI review summaries changed stated purchase choices, while a separate experiment found that disclosing that a persuasive chatbot was AI did not by itself remove its influence. Neither study proves the same effect in a physical shop, but both show why interface design can alter behaviour without changing the underlying product.
A machine can strip some of that social meaning away. Asking a robot where a size is located may feel closer to using a search box than admitting uncertainty to another person. That can be valuable in categories where people feel self-conscious, but it may be less valuable when expertise, empathy or reassurance are the reason the customer wants assistance.
This helps explain why technology adoption does not always follow age or technical enthusiasm. LiveAIWire recently reported that older travellers were more likely to adopt an AI shopping assistant on Ctrip. The common thread is that people may adopt AI because it removes friction from a task, not because they are enthusiastic about artificial intelligence itself.
Human warmth and human pressure can arrive together
Much of the debate about service robots assumes a trade-off between efficient machines and warm humans. The new research suggests the human side of that equation can contain both warmth and pressure. A person can notice hesitation, improvise and show genuine empathy, but their attention can also make the customer more aware of being observed.
That tension appears elsewhere in human-AI interaction. Warnings about AI flattery lowered trust without reliably removing the system’s influence, showing that people’s reactions to machines are not captured by a single question such as ‘Do you trust it?’ In retail, the relevant feeling may be comfort rather than trust in the abstract.
A retailer therefore needs to ask what job the assistant is doing. If the task is locating a product, checking stock or answering a routine specification question, low-pressure automation may work well. If the customer is anxious about fit, needs expert comparison or is making a high-value decision, a skilled person may add something the robot cannot reproduce.
The result is also a warning about how algorithmic advice feels
People sometimes treat machine output as more objective than an equivalent human judgement. LiveAIWire covered an experiment involving more than 1,300 teachers in which the same harsh grade seemed fairer when it was labelled as algorithmic. The retail study does not test grading or professional decisions, but both findings show that changing the perceived source can change the social meaning of identical or similar information.
That can be useful or dangerous. A robot that feels less judgemental may help a shopper ask questions they would otherwise avoid. Yet the appearance of neutrality can conceal commercial priorities. A retail assistant powered by software may still be designed to maximise conversion, promote particular products or collect information about the customer.
The absence of human embarrassment does not mean the absence of influence. It may simply move the influence into a different channel. Good retail design would make clear when recommendations are sponsored, personalised or optimised around business goals, rather than letting the machine’s emotionally neutral appearance imply independence.
A better shop may combine distance with access to people
The study points towards a hybrid service design rather than an obvious staff-replacement strategy. A robot or kiosk could handle the first contact and let customers request help without social pressure, while human staff remain available when judgement, creativity or reassurance are useful. That gives the shopper more control over when the interaction becomes personal.
The same principle can be achieved without a humanoid robot. A store app, screen, shelf interface or handheld device can provide private information and summon a person only when requested. The research is about the psychological characteristics of AI-mediated assistance, not an endorsement of a particular robot body.
Retailers would still need to test these systems in real stores. Useful measures would include how long people stay, whether they complete purchases, whether they return, which customers avoid assistance entirely and how often automated interactions escalate to staff. The key is to measure behaviour alongside satisfaction rather than assume novelty is a benefit.
There is also a workforce question. If automation handles low-pressure informational contact, human roles may move towards cases requiring expertise and judgement. That can improve the job if staff gain time for meaningful help, or make it more demanding if they receive only the hardest interactions. The technology’s effect depends on how work is redesigned around it.
The unexpected advantage is social, not mechanical
The most memorable result from the five studies is that a robot may sometimes be easier to deal with because it feels less like an audience. That is almost the reverse of the familiar criticism that AI makes service impersonal. In situations where personal attention itself creates pressure, a degree of impersonality can be useful.
It would be a mistake to turn that finding into a general claim that customers prefer robots. The experiments identify one pathway in particular settings, with clear moderating conditions. They do not erase the benefits of human service or establish long-term commercial outcomes.
What they do offer is a better design question. Instead of asking whether a shop should be staffed by humans or machines, ask when a customer wants to be noticed and when they want to be left alone. AI retail assistants may earn a place not by imitating the perfect salesperson, but by knowing when not to feel like one.
That distinction also matters for accessibility and privacy. Some shoppers may actively prefer a low-pressure machine interface, while others will need a person who can adapt to disability, language or an unusual request. The useful design is one that preserves choice rather than assuming a single service style fits everyone.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.