Digital Literacy Is Making Consumers More Critical of AI, Not More Resistant

The assumption that better digital literacy automatically breeds greater trust in artificial intelligence (AI) is beginning to look increasingly fragile. A new study from Germany finds that people who understand AI better are simultaneously more likely to recognize its usefulness and more likely to question its risks, suggesting that technological competence produces a more demanding consumer rather than a more compliant one.

The study, "The Paradox of Digital Literacy: Assessing Consumer Scepticism and Perceived Utility in the Age of AI-Driven Commerce," was published in the Journal of Theoretical and Applied Electronic Commerce Research. Authored by Ionut Laurentiu Petre, Georgiana-Raluca Ladaru, Raluca Andreea Ion, Maria Claudia Diaconeasa and Steliana Mocanu of the Bucharest University of Economic Studies, the research uses nationally representative German survey data to examine how digital literacy shapes perceptions of AI's benefits, risks and overall desirability.

Using Structural Equation Modelling, the researchers found that digital literacy significantly increased both perceived utility and consumer scepticism. However, utility had by far the stronger relationship with whether respondents viewed AI as an opportunity, while scepticism had no statistically significant direct effect on the overall evaluation.

Digital Literacy Is Creating Critical Consumers, Not Automatic AI Believers

The study challenges one of the most common assumptions embedded in digital-skills policy: that greater knowledge should reduce uncertainty and therefore increase acceptance. Instead, digitally literate users appear better equipped to see both sides of AI, appreciating its efficiency and potential while also identifying weaknesses involving misinformation, privacy, opacity and declining human control.

The relationship is visible in the model's coefficients. Digital literacy had a positive and statistically significant association with perceived utility, at β = 0.398, but it also significantly increased scepticism, at β = 0.182. The finding supports the idea of an "informed and critical consumer" who recognizes technological value without abandoning caution.

Scepticism is treated as a problem to be eliminated through education. The research suggests the opposite interpretation may sometimes be more appropriate: greater scepticism can reflect deeper understanding. Consumers who know more about algorithms may become more sensitive to manipulation, bias, fake news or the possibility that automated systems could exceed meaningful human oversight.

Utility Still Dominates the Final Judgment

While literacy strengthens both enthusiasm and concern, the strongest relationship in the study is between perceived utility and the belief that AI represents an opportunity. The standardized coefficient was β = 0.703, with a p-value below 0.001, while the direct effect of consumer scepticism was effectively zero and statistically non-significant, with a p-value of 0.975.

The authors interpret this pattern through the logic of the Privacy Paradox and the broader utility-risk trade-off. Consumers may recognize that AI brings uncertainty, surveillance concerns or reduced control, yet still accept it when the immediate benefits feel tangible enough. In other words, concern does not automatically translate into rejection.

The descriptive survey results reinforce that picture. Respondents who saw AI as an opportunity cited future technological development, greater convenience, work simplification, medical benefits and increased efficiency. Among those perceiving risk, the most prominent concern was loss of control at 21.3%, followed by fraud and abuse at 15.4%, unclear impacts at 10.5%, fake news at 9%, job losses at 7.5% and AI dominance at 7.4%.

The asymmetry is revealing. Benefits are often concrete, immediate and easy to imagine: faster tasks, better services, easier daily life or improvements in healthcare. Risks, although serious, are frequently more diffuse or future-oriented. The difference may help explain why consumers remain willing to engage with AI even while expressing strong reservations.

For digital businesses, this means trust-building alone may not drive adoption if customers cannot see what they gain. Firms deploying recommendation systems, conversational agents or AI-supported services must therefore pair transparency and responsible design with clear functional value rather than assuming that reassurance alone will overcome resistance.

AI Policy Cannot Treat Education, Trust and Value as the Same Problem

The study separates three issues that are often bundled together: literacy, trust and adoption. More literacy does not necessarily create more trust, and more scepticism does not necessarily stop adoption. Consumers may understand risks clearly and still choose an AI-enabled service because they believe the value outweighs the exposure, making digital-literacy programmes more complex than traditional technology training. Governments should not focus only on teaching citizens how AI works or how to use AI-enabled tools. The study suggests curricula should also address algorithmic accountability, misinformation, privacy, human oversight and the ethical dimensions of automation.

For regulators, the findings also challenge the idea that public confidence can be secured merely through disclosure. Explainability and transparency remain important, but more information can itself increase awareness of weaknesses. Regulation may therefore need to focus not only on making AI understandable, but on ensuring that systems remain contestable, accountable and genuinely useful once people understand them better.

In developing countries, where digital-literacy initiatives are often tied to financial inclusion, e-commerce expansion and digital public services, better-informed citizens may become more demanding about data use and automation. That could strengthen governance if institutions are prepared to respond, but it could also expose weak accountability arrangements that were less visible when technological literacy was lower.

For international development agencies and public-sector innovators, the lesson is that adoption should not be treated as a communication challenge alone. If people perceive clear value, they may tolerate some risk; if they do not, no amount of promotional messaging is likely to compensate. The strategic goal should therefore be trustworthy usefulness, not simply trust.

Evidence Demands Caution

The study uses nationally representative German survey data and by a modelling strategy that considers multiple psychological relationships simultaneously. It also makes an important conceptual contribution by treating scepticism as an active response to knowledge rather than as the mere absence of acceptance.

However, the authors acknowledge substantial limitations. The structural model showed only moderate fit: the TLI was 0.750 and RMSEA 0.098, both outside commonly recommended thresholds, while the CFI, IFI and NFI approached but did not reach the conventional 0.90 benchmark. The authors therefore caution that the path coefficients are better interpreted as indicative relationships than as precise estimates.

Measurement is another concern. Digital literacy was represented by a single self-assessment item, while reliability coefficients for the utility and scepticism constructs were below the conventional 0.70 threshold. Because the original GESIS survey was not designed specifically as a psychometric instrument for this theoretical model, the constructs capture useful signals but cannot fully represent the complexity of AI literacy or resistance.

Additionally, Germany combines relatively high digital capacity with a distinctive regulatory culture and a major role in European AI governance. Consumer reactions may differ in countries where institutional trust, digital access, regulation or experience with AI services vary substantially, making cross-national replication an important next step.

Future research should test the same relationships using validated multi-item AI-literacy measures, different national samples and specific use cases such as healthcare, banking, public services or e-commerce. Institutional trust and regulatory context may prove especially important in determining when scepticism remains manageable and when it becomes strong enough to outweigh utility.