T

For online business and online social interaction, the online disclosure of personal facial images has become a prevalent phenomenon. Taking platforms such as Airbnb, Uber, and LinkedIn as examples, users are often required to submit authentic, clear, and format-compliant profile photos, which are placed prominently in their personal profiles. The disclosure of facial images can eliminate anonymity through visual cues (Luca, 2017), build user trust (Dolnicar, 2019; Mosaad et al., 2023), and thereby reduce uncertainty in the transaction environment (Chattopadhyay and Mitra, 2019). Accordingly, the online disclosure of facial images has received ample attention in the literature.

However, due to the widespread existence of appearance bias, the “beauty premium” profoundly influences the decision-making logic in the online business environment. The beauty premium stems from people’s innate preference for attractiveness (Li et al., 2021) and is defined as the excess returns obtained from superior physical appearance (Hamermesh and Biddle, 1994). Existing studies have found that facial appearance has become a crucial factor affecting business performance in various online scenarios, including accommodation (Li et al., 2025), healthcare (Yang et al., 2024), tour guiding (Yang et al., 2022), e-commerce live streaming (Shi et al., 2024), sports commentary (Ryu et al., 2025), and advertising and marketing (Xia et al., 2025). The halo effect provides an explanation for this phenomenon, which states that when confronted with information asymmetry, observers tend to rely on heuristic cues for rapid judgment, forming an unconscious cognitive bias toward the whole based on individual salient features (Dion et al., 1972). Within the framework of the halo effect, physical attractiveness, as a salient cue perceived by users through appearance images in the online interaction environment, triggers this cognitive bias mechanism and further influences relevant decisions. Consequently, the role of physical attractiveness in human cognitive processes has sparked extensive discussion.

Naturally, the assessment of physical attractiveness (facial attractiveness) has become a key focus of research. Notably, the measurement methods of facial attractiveness have gradually evolved from subjective evaluation to objective, standardized, and intelligent approaches. In addition to well-established and commonly used methods such as questionnaires and experiments (Ch’ng & Narayanan, 2023; J. Li et al., 2018; Peng et al., 2020; Qian et al., 2024; Shang and Zhang, 2024), some recent studies have begun to conduct standardized assessments of massive facial images based on artificial intelligence technology (Barnes and Kirshner, 2021; Li et al., 2022, 2025; Peng et al., 2020; Yang et al., 2022).

Nevertheless, while existing research on the beauty premium is insightful, its findings are mostly grounded in single-shot or short-term decision-making contexts, leaving two critical limitations that warrant further exploration. First, prior studies have largely focused on transaction scenarios with extremely limited opportunities for decision information updating (e.g., physical product transactions, short-term service delivery), and it remains unclear whether their conclusions can be generalized to service scenarios involving high-frequency and long-term interactions. Such continuous interaction contexts (e.g., online education, long-term consulting services) provide an ideal setting for observing how initial impressions (e.g., the appearance halo) evolve with the influx of real experience information, yet they have received insufficient attention in the existing literature. Second, the vast majority of studies implicitly assume that decisions are single-shot and static, failing to uncover the dynamics of the halo effect over time. Specifically, the literature has not systematically addressed: In ongoing service relationships, how and why does the initial cognitive halo formed based on physical appearance evolve after the first (trial) purchase? To what extent can the appearance halo still function as an effective substitute cue to influence consumers’ repeat purchase decisions once decision-makers acquire more comprehensive service quality information through accumulated service experiences? The absence of this dynamic perspective constrains our understanding of the boundary conditions of the beauty premium.

Against this backdrop, this study aims to reveal how the service interaction context (one-shot vs. continuous) shapes the boundaries and dynamic processes of the beauty premium effect, thereby filling the aforementioned theoretical gap. Specifically, this study seeks to answer three core questions: (1) Does the impact of physical attractiveness on consumer decisions differ significantly between the first-time purchase stage and the repeat purchase stage, and what are the underlying theoretical mechanisms? (2) As service interactions persist and real experience information accumulates, how will the initial halo effect triggered by physical appearance evolve? Is it merely a short-term decision heuristic cue, or can it transform into a repeatable competitive advantage? What types of cues moderate this halo effect? (3) How can commercial platforms and service providers dynamically adjust their online image presentation strategies to most effectively manage customer impressions and enhance business performance across different decision-making stages?

This study adopts a longitudinal research design, using real transaction data of 11,450 language tutors on the online education platform Italki as the sample, quantifying tutors’ facial attractiveness via deep learning-based artificial intelligence technology, and constructing econometric models for empirical testing. The core findings corroborate and deepen the theoretical framework proposed in this study: (1) Facial attractiveness only significantly drives first-time purchases, but has no significant impact on repeated purchases; (2) In sharp contrast, tutors’ professional competence serves as a robust predictor of repeated purchases. These two findings jointly confirm a fundamental shift in decision-making bases from relying on external visual cues to core quality signals as service interactions deepen. In addition, heterogeneity analysis further reveals the differential effects of cues with different attributes: gender and age, which are also visual cues, exert significant moderating effects on facial attractiveness in the first-time purchase stage, whereas professional type labels as text-based certification cues show no moderating role, highlighting the dependence of information processing mechanisms on cue types in online decision-making.

The contributions of this study are as follows: First, this study breaks through the static and single-stage decision-making research paradigm of the beauty premium, and for the first time reveals that the sustainability of service interaction serves as a critical boundary condition constraining the functioning of the halo effect. By distinguishing between initial (trial) purchase and repeated purchase behaviors, this paper clearly delineates the applicable boundary of facial attractiveness as an “effective substitute cue” in activating the halo effect: the positive influence of appearance is concentrated in the initial decision-making stage with high information asymmetry, and attenuates significantly in the sustained purchase stage. The characteristics of such stage changes are inherently consistent with theoretical perspectives such as consumer learning and experience accumulation, providing direct empirical evidence for understanding the timeliness of the initial cognitive halo in the digital service environment. Second, this study uncovers the internal mechanism driving the above dynamic evolution, finding that signals of tutors’ professionalism replace facial attractiveness, which plays a core role in the first-time purchase stage, and become a new key decision-making factor in the repeated purchase stage. This demonstrates that consumers’ decision-making bases undergo a fundamental shift with information updating, namely from relying on peripheral cues that trigger the halo effect to central quality signals reflecting service quality. This finding offers an explanatory framework based on “decision stage evolution” for the contradictory conclusions on the beauty premium in previous studies. Finally, this study expands our understanding of the boundaries of online impression management and provides platforms and service providers with dynamic strategic foundations based on decision stages and cue types. Specifically, the management of facial visual elements (e.g., appearance, expressions) should be matched with decision stages, and their value is mainly reflected in the initial stage of attracting new customers. More importantly, the study identifies systematic differences in the moderating effects of cues with different attributes (visual vs. textual), suggesting that platforms should guide users to integrate multiple cues more effectively by optimizing information architecture (e.g., visualizing textual information).

Halo effect and beauty premium phenomenon

The halo effect is a classic cognitive bias in psychology. First proposed by Thorndike (1920), this effect refers to the cognitive generalization that occurs when people evaluate others or objects based on a single salient feature. Such a feature influences people’s judgments of the overall attributes of the perceived object much like a lunar halo obscures surrounding starlight, thereby resulting in cognitive bias. According to the cognitive simplification mechanism (Nisbett and Wilson, 1977),

Data

The data of this study are obtained from Italki (https://www.italki.com), a leading global online language education platform. Founded in 2006, the platform aims to connect and match language learners and tutors worldwide, and has now gathered more than ten thousand tutors and millions of learners.

Fig. 1 presents an example of the online profiles of language tutors listed on this platform, with the left image showing the tutor list and the right image showing an example of the personal profile

Conclusion and discussion

Based on an artificial intelligence-driven panel data analysis, this study deconstructs the differential effects of facial attractiveness on tutor performance in the knowledge-intensive online education context. The empirical results show that facial attractiveness significantly and positively predicts tutors’ first-time purchases (number of students) and teaching ratings, but exerts no significant impact on repeated purchases (average lessons per student). This finding confirms that in the

CRediT authorship contribution statement

Lifang Peng: Writing – review & editing, Funding acquisition, Conceptualization. Wenjun Lyu: Writing – original draft, Methodology, Investigation, Data curation.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements