New Study Reveals Why Shoppers Keep Coming Back to AI Shopping Assistants

Generative artificial intelligence has quietly transformed the way millions of people shop. What began as a simple ranking of recommended products has evolved into something far more interactive: conversational decision support, cross-platform price and feature comparison, and even the limited execution of shopping tasks on a user’s behalf. Yet a new study published in Current Psychology argues that researchers and designers have been conflating two fundamentally different technologies. The AI shopping assistants most consumers use today are not the fully autonomous, goal-pursuing agents of science fiction. They are bounded collaborators, and understanding that boundary, the researchers suggest, is the key to designing systems people actually want to keep using.

The study, led by Guanghong Xie of Tongji University’s College of Design and Innovation together with colleagues from Wuhan University of Technology and Tongji University, introduces the concept of AI-assisted shopping collaboration. In this arrangement, the human user retains final decision rights and the ability to intervene at any point, while the AI assistant shares the burden of integrating information and carrying out selected task operations. This deliberately limited form of delegation stands apart from agentic AI that can independently plan and pursue goals over extended horizons. By drawing this technological boundary explicitly, the authors aim to give designers and psychologists a clearer shared vocabulary for talking about what these systems do, and what they do not do.

To understand why some users embrace these assistants while others abandon them, the research team turned to two well-established psychological frameworks. The first is self-determination theory, which holds that human motivation flourishes when three basic psychological needs are met: autonomy, the sense that one’s actions are self-chosen; competence, the sense of being effective and capable; and relatedness, the feeling of connection to others. Applied to a shopping assistant, these needs translate into subtle perceptual questions. Does the AI make me feel in control of my own choices? Does it make me feel smarter about my decisions? Does interacting with it feel socially comfortable, almost like talking with a helpful companion?

The second framework is expectation-confirmation theory, a mainstay of information systems research. This model proposes that satisfaction with a technology depends on the gap between what users expected before using it and what they perceive after using it. When performance meets or exceeds prior expectations, confirmation occurs, satisfaction follows, and reuse becomes likely. When performance falls short, disconfirmation sets in and abandonment tends to follow. Combining these two theories, the researchers built an ordered model in which need-related perceptions and perceived trust, together with prior expectations, shape perceived quality and expectation confirmation, which in turn drive satisfaction and ultimately the intention to keep using the assistant.

The empirical work involved 300 users of AI shopping assistants, surveyed in a cross-sectional design that received ethics approval from the Institutional Review Board of Tongji University and was conducted in accordance with the Declaration of Helsinki. What distinguishes the analysis is its methodological ambition. Rather than relying on a single statistical technique, the team triangulated three complementary approaches: partial least squares structural equation modeling to test the structural associations among constructs, artificial neural networks to assess the predictive importance of each factor, and necessary condition analysis to determine whether any factors act as prerequisites rather than mere contributors.

The results carry several surprises with real design implications. Among the three basic psychological need-related perceptions, perceived relatedness emerged as the most prominent. In other words, the feeling of social connection, of the assistant being something more than a cold tool, was the strongest of the three motivational levers in the framework. This finding resonates with a growing body of evidence that people respond socially to conversational interfaces, and it suggests that the warm, dialogue-like character of generative shopping assistants is not a cosmetic feature but a psychological foundation of continued use. Designers who strip conversational sociality out of these tools may inadvertently be removing the very quality that keeps users engaged.

Prior expectations also proved remarkably influential. They showed strong associations with, and high predictive importance for, the two proximal evaluations in the model: perceived quality and expectation confirmation. This underscores a double-edged dynamic familiar from service research. High expectations can set a demanding bar, but they also frame the entire evaluative experience; users who arrive with vivid, concrete expectations judge quality and confirmation through that lens. For companies deploying AI shopping assistants, this means that marketing promises and onboarding experiences are not merely promotional materials. They actively calibrate the psychological benchmark against which every subsequent interaction will be measured.

The necessary condition analysis added a third layer of insight. Perceived quality and expectation confirmation did not merely correlate with satisfaction; they emerged as necessary conditions for high satisfaction. Without a minimum threshold of perceived quality and expectation confirmation, high levels of satisfaction appear to be unattainable regardless of how well other factors perform. This is a stricter and more actionable standard than correlation alone. It tells platform designers that investments in response accuracy, relevance, and reliable task execution are non-negotiable prerequisites, and that no amount of social charm or motivational framing can compensate for a fundamentally underwhelming core experience.

The triangulation of three analytical methods is itself part of the study’s contribution. Structural equation modeling reveals whether theoretical pathways hold, neural networks reveal which inputs matter most for prediction, and necessary condition analysis reveals which inputs gate the outcome entirely. When all three converge, as they did for prior expectations, perceived quality, and expectation confirmation, the evidence is considerably stronger than any single method could provide. The authors argue that this integrated psychological and multi-method perspective offers a template for studying not just shopping assistants but human-AI collaboration in general, particularly in domains where users delegate limited tasks while keeping ultimate control.

As agentic AI systems grow more capable and platforms race to deploy assistants that can browse, compare, and even purchase autonomously, this research offers a timely corrective. The findings suggest that the path to durable adoption runs not through maximal autonomy but through careful collaboration design: preserving the user’s sense of agency, nurturing a feeling of social connection, managing expectations honestly, and delivering a core experience that is unquestionably good. The study, funded by the Shanghai Philosophy and Social Science Planning Project and partner programs, was published in Current Psychology, with data available from the corresponding author on reasonable request. For an industry betting its future on AI-mediated commerce, the message is clear: the psychology of collaboration, not the raw power of the model, may decide who wins the next generation of shopping.

Subject of Research: Psychological factors driving user satisfaction and reuse intention in AI-assisted shopping collaboration

Article Title: AI-Assisted shopping collaboration design: Basic Psychological need-related perceptions, expectation evaluation, and reuse intention

Article References: Xie, G., Bao, Q., Wang, Z., Shen, X., & You, F. (2026). AI-Assisted shopping collaboration design: Basic Psychological need-related perceptions, expectation evaluation, and reuse intention. Current Psychology, 45(17), Article 1488. https://doi.org/10.1007/s12144-026-10054-3

Image Credits: AI Generated

DOI: 10.1007/s12144-026-10054-3

Keywords: AI shopping assistant, AI-assisted shopping collaboration, self-determination theory, expectation confirmation, reuse intention, generative AI, human-AI collaboration, consumer behavior, perceived relatedness, structural equation modeling, artificial neural networks, necessary condition analysis