Online shopping can turn a simple question into an exhausting comparison. A shopper looking for a laptop, insurance policy or household appliance may now ask an AI assistant to summarize dozens of options in seconds. Yet the speed of an answer does not necessarily make the decision easier, especially when the resulting information is difficult to assess.
A new peer-reviewed study suggests that knowing how to question and evaluate generative AI may matter as much as having access to it. In a survey of 401 university students in China, those who reported stronger generative-AI literacy also tended to report less information overload and more favorable judgments about the quality of their online shopping decisions.
The relationship was particularly pronounced among students who regarded the purchasing decision as more important. But there is a vital distinction: the researchers measured how good participants believed their decisions were, not whether they objectively chose the best products or paid lower prices.
Published on 21 October 2026 in Frontiers in Psychology, the research provides an evidence-based look at the growing role of AI literacy in consumer behavior while underscoring how much remains unknown about the effects of AI-assisted shopping.
When shopping information becomes another problem to solve
Online marketplaces already give customers access to enormous numbers of specifications, reviews, promotions and competing products. Generative AI adds another layer. Instead of reading every product page, a user can request a comparison, ask a follow-up question or have a long review summarized in ordinary language.
That convenience is valuable, but it can also create a new challenge. An answer may be convincing without clearly identifying its sources. It may mix supported facts with uncertain claims or present a long list of recommendations that looks more useful than it is. The user still has to decide which details are trustworthy and relevant.
The study examined this problem through the concept of perceived information overload: the feeling that the volume or complexity of available material makes a purchasing decision harder to process. This is not simply a measure of how many product options appear on a screen. It concerns the burden consumers experience while trying to turn information into a choice.
Yiming Guo, the study’s author, considered whether users who understand generative AI’s strengths and limitations may be better equipped to navigate that burden. The researchers’ framework draws on information-processing theory, which emphasizes that attention and cognitive capacity are limited resources.
A person who can ask specific questions, detect missing evidence and organize comparisons may find the shopping process more manageable. Someone who treats every generated recommendation as equally useful may face more material to evaluate without a reliable way to prioritize it.
These explanations are plausible mechanisms, not behaviors directly observed in the study. The data describe reported literacy, overload and decision quality at one point in time.
What the researchers asked 401 students
The evidence comes from an online survey conducted in January 2026 using Wenjuanxing, a survey platform in China. Participants were recruited through social media and email invitations. The final analysis included 401 students from Chinese higher education institutions, with no cases removed from the final dataset.
The sample consisted of 184 male respondents (45.89%) and 217 female respondents (54.11%). Students at different stages of study were represented, including undergraduates and a smaller group of graduate students. Participation was voluntary, and the study did not use random population sampling.
Respondents were instructed to consider their experiences and capabilities with generative AI tools, including services such as ChatGPT and DeepSeek, in the context of online product searching, feature comparison and recommendations.
Importantly, the researchers did not require participants to have previously used an AI assistant to complete a real purchase. Their responses therefore reflected perceptions in an AI-shopping context rather than verified records of shopping sessions.
The survey measured four areas. Generative-AI literacy covered awareness, practical use, critical evaluation and ethical understanding. Information overload captured respondents’ difficulties with too much or confusing shopping information. Decision quality reflected how systematic and satisfactory the participants considered the process. Consumer involvement described how personally important a purchase felt.
The literacy instrument contained 10 questions scored on a five-point scale. Information overload used 14 questions, perceived decision quality four questions and consumer involvement seven questions, with the latter three constructs measured on seven-point scales.
These separate instruments matter because the study did not treat familiarity with an AI brand as equivalent to the ability to use that tool critically. Knowing what a chatbot is and knowing when its recommendations require verification are different skills.
The main result: greater literacy, less overload
The simplest associations already revealed a consistent pattern. Generative-AI literacy correlated positively with perceived decision quality, at r = 0.406, and negatively with perceived information overload, at r = −0.417. Both correlations were statistically significant at p < 0.001.
Information overload was itself negatively related to reported decision quality, with a correlation of r = −0.475. In ordinary terms, participants who felt more overwhelmed by AI-assisted shopping information tended to judge their shopping decisions less favorably.
The researchers then estimated models that accounted for gender, grade level and socioeconomic status. Stronger AI literacy remained associated with higher perceived decision quality, with a standardized coefficient of β = 0.403 and a 95% bootstrap confidence interval of 0.322 to 0.488.
This first model explained 17.7% of the variation in perceived decision quality. That is a meaningful statistical association within the sample, but it leaves most variation unexplained. Other factors beyond AI literacy clearly contribute to the way shoppers judge their decisions.
The relationship with information overload was also evident after adjustment. AI literacy was negatively associated with overload (β = −0.418), while information overload was negatively associated with perceived decision quality (β = −0.372) once literacy and the control variables were considered together.
The model incorporating both literacy and overload accounted for 29.0% of variation in reported decision quality. These percentages describe statistical explanatory power, not the proportion of people who made better purchases.
Why information overload matters to the relationship
The research went beyond asking whether knowledgeable users simply liked their choices more. It tested an indirect statistical pathway: was the association between AI literacy and perceived decision quality partly linked to the amount of overload users experienced?
Using 3,000 bootstrap resamples, the study estimated a positive indirect association of β = 0.156, with a 95% confidence interval from 0.108 to 0.210. The interval did not include zero, supporting the proposed pattern within the statistical model.
The direct association between literacy and decision quality remained significant after information overload was included (β = 0.248). In the paper’s terminology, the findings were consistent with partial statistical mediation.
This means that reduced information overload was associated with part, but not all, of the relationship between AI literacy and better-rated decisions. It does not prove that teaching AI skills would automatically reduce overload and then improve real-world purchase outcomes.
All three constructs were measured in the same questionnaire. A respondent who generally felt confident may have rated several aspects of their experience favorably. The statistical analysis can account for measured variables but cannot fully separate the direction of influence from other unmeasured characteristics.
The distinction is especially important for commercial claims. A company cannot use these results alone to promise that an AI-literacy tutorial will increase conversion rates, reduce returns or improve customer satisfaction. Those would require studies tracking actual behavior before and after an intervention.
Higher-stakes purchases showed a stronger pattern
One of the paper’s more interesting findings concerns consumer involvement, or how personally important a purchase is to the shopper. A low-involvement decision might involve an inexpensive everyday product. A high-involvement decision might involve a costly device, a service contract or another choice with consequences the buyer takes seriously.
The association between AI literacy and reduced information overload was present at both levels, but was stronger when involvement was higher. At one standard deviation below the average involvement score, the standardized slope was −0.186. At one standard deviation above the average, it was −0.480.
The difference was supported by a statistically significant interaction between literacy and involvement (β = −0.147, p = 0.002). Involvement was not divided into two actual groups for the analysis; those slopes describe modeled values at lower and higher points on a continuous scale.
The indirect literacy-to-decision-quality association through information overload was also larger at higher involvement (0.179) than at lower involvement (0.069). The reported index of moderated mediation was 0.055, with a 95% confidence interval of 0.020 to 0.095.
A practical interpretation is that AI skills may be more closely linked to the experience of managing complex information when a person cares deeply about a choice. Such shoppers have a reason to compare more alternatives and examine recommendations in greater detail.
However, the analysis cannot establish that importance caused people to use their AI skills more effectively. It identifies a pattern of conditional associations, not the outcome of an experiment in which the importance of a purchase was manipulated.
What the findings suggest for online retailers
For digital commerce teams, the study draws attention to the customer’s ability to interpret information, rather than focusing exclusively on how advanced the retailer’s AI assistant appears. More suggestions, longer explanations and additional generated text may not help every user equally.
One design possibility is a shopping interface that offers concise comparisons by default, with clear access to supporting evidence and deeper details when needed. Consumers making consequential purchases may want structured side-by-side comparisons, sources and explicit indications of uncertainty.
Another possibility is lightweight guidance on verifying generated information. For example, shoppers could be encouraged to check delivery terms against the retailer’s actual policy, distinguish sponsored placements from product specifications and confirm important prices or contract conditions through primary sources.
These are potential applications informed by the study, not tested recommendations. The researchers did not conduct a randomized interface trial or measure which layout would produce better decisions in a live store.
Previous Research Today coverage of AI evaluation described how model judgments can change with the opinions supplied to them. The broader lesson is relevant to shopping: confidently expressed AI output should still be checked against reliable information.
Similarly, research into how companies turn social-media data into marketing decisions illustrates how interpreting digital information remains a business challenge even when tools can process large amounts quickly.
The present survey adds the consumer side of that problem. Instead of assuming that greater access to information means a better choice, it asks how the user’s understanding of the tool relates to the burden of evaluating its answers.
Why self-reported confidence is not actual shopping accuracy
Perceived decision quality is an important outcome, particularly in customer experience research. Consumers who believe a decision was systematic may feel more comfortable continuing with a purchase. But confidence and accuracy are not interchangeable.
A person can feel thoroughly informed and still choose a product with unsuitable features. Conversely, someone who remains uncertain may nevertheless have identified the most appropriate option. Determining actual decision quality would require evidence about prices, quality, suitability or later satisfaction with the product.
The study measured respondents’ agreement with statements about the quality and systematic character of decisions in AI-assisted online-shopping settings. It did not independently score whether they chose the objectively best option from a set of alternatives.
The result also should not be presented as a demonstration that AI-based shopping makes people more financially responsible. Research Today reporting on buy-now-pay-later behavior highlights a different aspect of digital commerce: consumers’ financial circumstances and anxieties can shape outcomes beyond the informational choices made at checkout.
In short, even a clear statistical association between AI literacy and confidence cannot tell us whether users save money, select safer products, avoid debt or experience fewer returns.
What the study cannot tell us
The authors acknowledge several limitations. First, the sample consisted of Chinese university students recruited through convenience sampling. Their digital familiarity, educational experiences and shopping patterns may differ from those of older consumers, rural shoppers or people in other countries.
Second, the survey was cross-sectional: respondents completed the measures at one point in time. This prevents firm conclusions about causality or timing. It is possible that confident shoppers develop stronger AI literacy over time, rather than AI literacy being the sole source of confidence.
Third, all the main variables were self-reported. That creates the possibility of common measurement influences, even though the researchers used statistical checks to examine whether their measures represented distinct concepts.
The measurement analysis was comparatively thorough. The main four-factor model showed strong reported fit, including CFI = 0.992 and RMSEA = 0.018. Nevertheless, some proposed subdimensions within AI literacy and information overload were not sufficiently distinct in this sample, so the researchers used overall scores for those constructs.
Fourth, prior actual use of AI for a real purchase was not an eligibility requirement. Participants were asked to consider generative AI in a shopping setting, which is not the same as observing how they behaved while deciding whether to spend their money.
Finally, the research offers no direct evidence about which consumer-education program or online-store interface would improve results. Those possibilities would need experiments or longitudinal studies involving shoppers, actual product choices and objective outcomes.
A more useful question for AI-assisted shopping
Generative AI can make product information easier to retrieve, but it cannot remove the need for critical judgment. This study suggests that users who report greater ability to understand and evaluate the technology also report less information overload and more favorable shopping decisions, especially when a choice feels important.
The question for retailers and consumers is therefore not just whether AI can produce a recommendation. It is whether people can recognize what the recommendation is based on, evaluate competing claims and know when to seek more reliable evidence.
For now, the evidence concerns how shoppers experience decisions, not whether AI literacy produces objectively better purchases. That boundary is precisely what future research will need to test before turning an encouraging survey relationship into a confident promise about consumer outcomes.
Source Information
Study title: GenAI literacy and perceived decision quality in e-commerce: the mediating role of perceived information overload and the moderating role of consumer involvement
Authors: Yiming Guo
Journal: Frontiers in Psychology, Volume 17, Article 1919451
Publication date: 21 October 2026
Study design: Cross-sectional quantitative online survey, with confirmatory factor analysis, covariate-adjusted path models and 3,000-bootstrap mediation and moderation analysis
Sample or scope: 401 university students in China recruited by convenience sampling in January 2026; experiences and perceived decisions in AI-assisted online shopping
Key finding: AI literacy was positively associated with perceived shopping decision quality (β = 0.403, p < 0.001) and negatively associated with perceived information overload (β = −0.418); the overload association was stronger at higher consumer involvement
