GenAI literacy linked to perceived e-commerce decision quality in Chinese student survey

In an online survey of 401 Chinese college students, people reporting greater generative artificial intelligence literacy also rated their shopping decisions as higher quality and experienced less information overload when using GenAI tools for product research. The study found a statistically significant indirect association between GenAI literacy and perceived decision quality through perceived information overload, while a direct association between literacy and perceived decision quality remained after the proposed mediator was included. Consumer involvement moderated the first stage of this pathway: the negative association between literacy and overload was stronger at higher involvement. These results describe statistical relationships in a cross-sectional self-report study, not evidence that literacy caused objectively better purchasing decisions.

The research, published in Frontiers in Psychology, examined four linked constructs in the context of GenAI-assisted online shopping. GenAI literacy was defined as the knowledge and skills needed to understand AI systems, use them effectively, evaluate their outputs and recognize ethical considerations. Perceived information overload captured the sense that the demands of processing shopping information exceeded a consumer’s capacity. Perceived decision quality reflected evaluations of the decision process, including whether it appeared systematic and whether relevant alternatives had been considered. Consumer involvement represented the personal importance attached to a product or purchase. The analysis therefore examined subjective evaluations of shopping decisions rather than objective measures of whether the selected product was the best available option.

Participants were instructed to consider their general experiences and capabilities with tools such as ChatGPT and DeepSeek and to apply those experiences to online shopping activities, including searching for product information, comparing product features and reading recommendations. Previous use of GenAI during an actual purchase was not an eligibility requirement, so the sample was not restricted to experienced GenAI shoppers. Data were collected in January 2026 through Wenjuanxing, an online survey platform used in China. Participants were recruited through social media and email invitations under a non-probability convenience sampling strategy, with no financial or material compensation. The final sample comprised 184 men and 217 women, and included students from first year through graduate level or above. The statistical models controlled for gender, grade level and self-reported socioeconomic status.

The researchers used established scales adapted to GenAI-assisted shopping. The ten-item GenAI literacy measure covered awareness, usage, evaluation and ethics, with responses on a five-point strongly disagree to strongly agree scale. Perceived information overload was assessed with 14 items on seven-point scales, covering sensitivity to the amount of information, anticipation of negative affect, anticipation of mistakes and perceived information-processing ability; four positively worded ability items were reverse-coded so that higher scores consistently represented greater susceptibility to overload. Four items assessed perceived decision quality, and seven bipolar semantic-differential items measured consumer involvement. All four constructs were retained as continuous mean scores. Their Cronbach’s alpha values ranged from 0.862 for perceived decision quality to 0.956 for perceived information overload.

Measurement analysis supported the reliability, convergent validity and discriminant validity of the four measures. A confirmatory factor analysis of the correlated four-factor model showed a comparative fit index of 0.992, a Tucker–Lewis index of 0.991, a root mean square error of approximation of 0.018, with a 90% confidence interval of 0.008 to 0.025, and a standardized root mean square residual of 0.029. A single-factor model fit substantially worse, with a comparative fit index of 0.577 and root mean square error of approximation of 0.132. All standardized factor loadings in the four-factor model were statistically significant, ranging from 0.751 to 0.796. Average variance extracted values were between 0.593 and 0.611, and heterotrait–monotrait ratios ranged from 0.372 to 0.523, below the study’s conservative 0.850 threshold. The analyses therefore supported treating the four constructs as related but empirically distinguishable in this sample.

The main model found that GenAI literacy had a positive adjusted association with perceived decision quality, with a standardized coefficient of 0.403, a bootstrap standard error of 0.043 and a 95% bootstrap confidence interval from 0.322 to 0.488. The model accounted for 17.7% of the variance in perceived decision quality. GenAI literacy was negatively associated with perceived information overload, with a coefficient of −0.418, while overload was negatively associated with perceived decision quality, with a coefficient of −0.372. The resulting indirect association through overload was 0.156, with a bootstrap standard error of 0.026 and a 95% bootstrap confidence interval of 0.108 to 0.210. The direct association between literacy and perceived decision quality was 0.248, with a confidence interval of 0.156 to 0.342. This pattern was consistent with partial statistical mediation: overload represented one pathway linking literacy to evaluated decision quality, while an association between the two focal constructs remained.

Consumer involvement also showed a statistically significant interaction with GenAI literacy in the information-overload equation. The interaction coefficient was −0.147, with a bootstrap standard error of 0.047, p = 0.002 and a 95% bootstrap confidence interval from −0.241 to −0.059. Simple slopes showed that literacy remained negatively associated with overload at one standard deviation below mean involvement, with a coefficient of −0.186 and a 95% confidence interval from −0.318 to −0.054. At one standard deviation above the mean, the coefficient was −0.480, with a confidence interval from −0.608 to −0.352. The conditional indirect association through overload was 0.069 at low involvement, with a 95% bootstrap confidence interval of 0.017 to 0.125, and 0.179 at high involvement, with an interval of 0.117 to 0.250. The index of moderated mediation was 0.055, with a bootstrap standard error of 0.019 and a confidence interval of 0.020 to 0.095. In statistical terms, higher involvement was associated with a stronger literacy–overload relationship; the design did not show that involvement caused that difference.

The findings fit the Information Processing Theory used by the researchers, which treats attention and working memory as limited resources. Generated shopping responses can summarize many attributes, compare products or provide recommendations, potentially reducing the work required to find and organize information. At the same time, the material still needs to be assessed for relevance, consistency and credibility. Fluent language can present unsupported or inaccurate claims confidently, leaving consumers to decide what to verify and what can safely be ignored. The study treated these verification demands as part of perceived information overload rather than measuring them as a separate psychological construct. Consumers with stronger GenAI literacy may be better able to formulate focused requests, identify relevant comparison criteria, recognize uncertainty and organize generated material around a purchase objective.

The authors discuss possible implications for platforms, developers and educators, but the evidence comes from a single self-reported survey and does not prove that a particular interface or training program will improve outcomes. Traceable sources, structured comparisons and explicit indications of uncertainty could potentially help consumers evaluate generated information, while concise default responses with optional detail might reduce demands during routine purchases. The authors also propose extending AI-literacy education beyond prompt formulation to source evaluation, uncertainty recognition, comparison of alternative outputs and awareness of potential bias. They note that such initiatives should be evaluated against behavioral measures such as decision accuracy and response time, rather than perceived decision quality alone. The study itself did not test these interventions or compare interface designs.

Several constraints limit the conclusions. The sample consisted exclusively of Chinese university students recruited through convenience sampling, limiting generalization to older consumers, people with different educational or socioeconomic backgrounds, and other cultural settings. Participants were not required to have used GenAI in an actual purchase, so their responses may not represent consumers with established GenAI shopping experience. All focal variables were captured at one time through the same questionnaire, leaving common-method influences, temporal ordering and causality unresolved, although the single-factor diagnostics did not indicate that one common factor fully explained the covariance among the measures. Perceived information overload was also self-reported, and the outcome did not capture objective decision accuracy or performance. Longitudinal, experimental and multi-source designs, combined with shopping tasks, accuracy measures, response times and process-tracing methods such as eye tracking, would be needed to determine whether the observed associations reflect a causal process and whether they apply to actual purchasing behavior.

Subject of Research: Associations among GenAI literacy, information overload, consumer involvement and perceived e-commerce decision quality

Article Title: GenAI literacy and perceived decision quality in e-commerce: the mediating role of perceived information overload and the moderating role of consumer involvement

Article References: Guo, Y. (2026). GenAI literacy and perceived decision quality in e-commerce: the mediating role of perceived information overload and the moderating role of consumer involvement. Frontiers in Psychology, 17, Article 1919451. https://doi.org/10.3389/fpsyg.2026.1919451

Image Credits: AI Generated

DOI: 10.3389/fpsyg.2026.1919451

Keywords: generative artificial intelligence, GenAI literacy, e-commerce, consumer behavior, information overload, decision quality, consumer involvement, information processing, online shopping, China, GenAI, literacy