What Drives University Students to Adopt Generative AI as a Learning Tool?

Moon Su-ji’s study, “The Impact of University Students’ AI-Related Digital Literacy on Their Intention to Use Generative AI as a Learning Tool: The Moderating Effect of Self-Efficacy,” asks what leads students to choose generative AI for learning rather than merely how frequently they use it. The researcher surveyed 255 students enrolled at four-year universities across South Korea and examined the relationships among AI-related digital literacy, self-efficacy, and intention to use generative AI as a learning tool.

The results show that both digital literacy and self-efficacy positively influence students’ intention to use generative AI. However, self-efficacy does not significantly strengthen or weaken the relationship between digital literacy and usage intention.

From “Do Students Use AI?” to “Do They Choose It for Learning?”

Generative AI has already become an everyday technology that most university students can access. Yet access to a tool and the sustained integration of that tool into learning are two different matters. Asking AI a few questions, using it to complete an assignment, and repeatedly employing it to understand and verify learning materials cannot easily be treated as the same form of “use.”

This study focuses on the distance between these activities. Rather than measuring only how much students currently use generative AI, the researcher investigates whether they intend to choose AI for future study, coursework, and educational activities. Usage intention may appear less concrete than actual behaviour, but it is an important variable for understanding how a new technology becomes incorporated into learning habits.

Knowing that a tool exists does not necessarily lead students to choose it, and choosing it does not guarantee that they will use it effectively for learning. The central question therefore shifts from “Are university students using AI?” to “What abilities and perceptions lead them to choose AI as a learning tool?”

From this perspective, the educational spread of generative AI cannot be explained solely by the availability of services or user accounts. Students must develop both the judgement required to connect the technology to learning and the belief that they can use it competently.

Digital Literacy Is a Structure of Judgement, Not Just Technical Skill

The AI-related digital literacy measured in the study goes beyond the ability to launch a program or type a prompt. It includes whether students can use AI-based digital tools to perform academic tasks, locate and evaluate learning resources, improve their learning through digital tools, and understand how to protect and manage information safely.

Taken together, these survey items reveal how the researcher understands digital literacy. It is less a matter of knowing numerous functions than of being able to position a tool appropriately within a task. It involves identifying what needs to be done, finding relevant information, evaluating the quality of the result, and managing risks involving privacy, copyright, and information security.

Such abilities become particularly important with generative AI. Conventional software generally performs user commands in relatively predictable ways. Generative AI, by contrast, can produce responses that sound convincing while remaining factually wrong. It may generate different answers to the same question, and the quality and direction of its output can change considerably depending on how the user frames the request.

A competent user is therefore not merely someone who obtains results quickly. Competence lies in the ability to ask again, compare outputs, verify information, and revise the result.

The finding that digital literacy positively influences usage intention suggests that students who possess these evaluative abilities are more likely to accept generative AI as part of their learning. Once students understand how the technology works and know how to control and assess its outputs, AI begins to appear less like an unpredictable automation system and more like a manageable learning resource.

Knowing How to Use AI Becomes a Sense of Being Able to Use It

The study also finds that AI-related digital literacy positively affects self-efficacy. In the structural model, the standardised path coefficient from digital literacy to self-efficacy was .494, while the direct effect of digital literacy on usage intention was .282. The effect of self-efficacy on usage intention was .686, making it the strongest of the three paths.

Self-efficacy differs from vague confidence or optimism. The items used in the study asked whether students believed they could search for various kinds of information with generative AI, whether they were comfortable retrieving and sharing information, whether they could use multiple AI functions, and whether they could handle AI tools proficiently. Self-efficacy therefore refers to an individual’s judgement that they can successfully perform a particular action.

When students understand the functions and risks of AI and gain experience searching for information and evaluating results, they begin to develop the feeling that “I can use this tool.” At first, they may hesitate because they do not know how to write a prompt. As they accumulate experience revising outputs and asking follow-up questions, however, they develop the expectation that they can guide and control the system.

Digital literacy does not remain a collection of knowledge. It becomes a practical sense that allows students to begin acting.

The strong effect of self-efficacy on usage intention also shows that students do not select a technology based only on its objective performance. Even an advanced tool may not be adopted for learning when users believe they cannot operate it properly. Conversely, students with incomplete knowledge may still incorporate the technology into their learning when they believe they can tolerate trial and error and improve the results through continued interaction.

Self-Efficacy Matters, but It Does Not Determine the Effect of Literacy

The most interesting finding is that the moderating effect of self-efficacy was not statistically significant. It is reasonable to expect that students with higher self-efficacy would have a stronger intention to use AI. Yet the study does not find that digital literacy becomes effective only among highly confident students, nor that its effect disappears among those with lower confidence.

The researcher divided participants into a low-self-efficacy group of 113 students and a high-self-efficacy group of 142 students. Digital literacy had a significant positive effect on usage intention in both groups. The path coefficient was .450 in the low-self-efficacy group and .651 in the high-self-efficacy group, but the difference between the two groups was not statistically significant.

This result invites a different interpretation of the relationship between self-efficacy and digital literacy. Self-efficacy appears to function less as a switch that activates the effect of literacy and more as an independent factor that adds further momentum to usage intention. Even students with relatively low confidence become more willing to use AI for learning when their digital literacy improves. Students with high confidence also continue to benefit from digital literacy.

The result does not support the immediate conclusion that students should be divided into high- and low-confidence groups and given entirely different forms of digital literacy education. Regardless of their level of self-efficacy, digital literacy education is likely to increase their intention to use AI.

At the same time, it would be unwise to assume that literacy education automatically resolves issues of confidence. Learning environments must also provide practical experiences through which students can develop self-efficacy.

Digital Literacy and Self-Efficacy Reduce Different Forms of Friction

Viewed from the perspective of the learning process, digital literacy and self-efficacy reduce different obstacles.

Digital literacy reduces friction in judgement. When students do not know what to ask, whether an answer is trustworthy, or what information should be checked next, they can easily become stuck in front of an AI-generated response. Higher literacy allows them to evaluate the output and decide what to do next, making it easier to connect AI to an ongoing learning process.

Self-efficacy reduces friction in beginning and sustaining action. It helps students believe that they can try again when a result differs from what they expected, explore unfamiliar functions, and avoid treating a failed output as evidence of their own incompetence. The belief that they can handle AI lowers the psychological cost of beginning and supports repeated experimentation.

The two factors move in the same direction, but they do not perform identical roles. A student may possess sound judgement yet fail to act because of low confidence. Another student may be highly confident but unable to assess the quality of AI-generated information. The study indicates that the educational use of generative AI depends on combining two processes that remove different forms of friction rather than choosing between knowledge and confidence.

AI Education Must Shift from Explaining Functions to Organising Experience

The study argues that educational environments should strengthen both AI-related digital literacy and self-efficacy if they aim to encourage students to use generative AI for learning. An important premise here is that digital literacy is a dynamic capability that develops and accumulates over time. A brief introduction to new functions or a one-time lesson on prompt writing is therefore insufficient.

Students need repeated opportunities to formulate questions, inspect generated outputs, compare them with other sources, and correct errors through actual learning tasks. These activities should also include decisions about source verification, privacy management, bias, copyright, and ethical use. Explaining functions may provide a starting point, but literacy develops as students make a series of practical judgements.

Self-efficacy cannot be built through encouragement alone. Educational tasks should initially have clearly defined scopes, allow students to revise results, and demonstrate that outputs improve when users intervene. Students develop self-efficacy when they experience themselves not as passive recipients of AI’s first answer, but as people who correct its output and determine its direction.

The study’s central contribution lies in showing that the educational adoption of generative AI cannot be judged by technological performance alone. Students’ decisions depend on what they understand, how they evaluate the results, and whether they believe they can perform the process successfully.

The question for education therefore changes over time. Rather than asking which AI system performs best, educators must ask what kinds of experiences will enable students to examine, control, and connect AI to their own learning.