by Concepción González García and Nina Pallarés Cerdà
Debates about generative AI in higher education often start from the same assumption: students need a certain level of digital competence before they can use AI productively. Those who already know how to search, filter and evaluate online information are seen as the ones most likely to benefit from tools such as ChatGPT, while others risk being left further behind.
Recent studies reinforce this view. Students with stronger digital skills in areas like problem‑solving and digital ethics tend to use generative AI more frequently (Caner‑Yıldırım, 2025). In parallel, work using frameworks such as DigComp has mostly focused on measuring gaps in students’ digital skills – often showing that perceived “digital natives” are less uniformly proficient than we might think (Lucas et al, 2022). What we know much less about is the reverse relationship: can carefully designed uses of AI actually develop students’ digital competences – and for whom?
In a recent article, we addressed this question empirically by analysing the impact of a generative AI intervention on university students’ digital competences (García & Pallarés, 2026). Students’ skills were assessed using the European DigComp 2.2 framework (Vuorikari et al, 2022).
Moving beyond static measures of digital competence
Research on students’ digital competences in higher education has expanded rapidly over the past decade. Yet much of this work still treats digital competence as a stable attribute that students bring with them into university, rather than as a dynamic and educable capability that can be shaped through instructional design. The consequence is a field dominated by one-off assessments, surveys and diagnostic tools that map students’ existing skills but tell us little about how those skills develop.
This predominant focus on measurement rather than development has produced a conceptual blind spot: we know far more about how digital competences predict students’ use of emerging technologies than about how educational uses of these technologies might enhance those competences in the first place.
Recent studies reinforce this asymmetry. Students with higher levels of digital competence are more likely to engage with generative AI tools and to display positive attitudes towards their use (Moravec et al, 2024; Saklaki & Gardikiotis, 2024). In this ‘competence-first’ model, digital competence appears as a precondition for productive engagement with AI. Yet this framing obscures a crucial pedagogical question: might AI, when intentionally embedded in learning activities, actually support the growth of the very competences it is presumed to require?
A second limitation compounds this problem: the absence of a standardised framework for analysing and comparing the effects of AI-based interventions on digital competence development. Although DigComp is widely used for diagnostic purposes, few studies employ it systematically to evaluate learning gains or to map changes across specific competence areas. As a result, evidence from different interventions remains fragmented, making it difficult to identify which aspects of digital competence are most responsive to AI-mediated learning.
There is, nevertheless, emerging evidence that AI can do more than simply ‘consume’ digital competence. Studies by Dalgıç et al (2024) and Naamati-Schneider & Alt (2024) suggest that integrating tools such as ChatGPT into structured learning tasks can stimulate information search, analytical reasoning and critical evaluation—provided that students are guided to question and verify AI outputs rather than accept them uncritically. Yet these contributions remain exploratory. We still lack experimental or quasi-experimental evidence that links AI-based instructional designs to measurable improvements in specific DigComp areas, and we know little about whether such benefits accrue equally to all students or disproportionately to those who already possess stronger digital skills.
This gap matters. If digital competences are conceived as malleable rather than fixed, then AI is not merely a technology that demands certain skills but a pedagogical tool through which those skills can be cultivated. This reframing shifts the centre of the debate: away from asking whether students are ready for AI, and towards asking whether our teaching practices are ready to use AI in ways that promote competence development and reduce inequalities in learning.
Our study: teaching students to work with AI, not around it
We designed a randomised controlled trial with 169 undergraduate students enrolled in a Microeconomics course. Students were allocated by class group to either a treatment or a control condition. All students followed the same curriculum and completed the same online quizzes through the institutional virtual campus.
The crucial difference lay in how generative AI was integrated:
- In the treatment condition, students received an initial workshop on using large language models strategically. They practised:
- contextualising questions
- breaking problems into steps
- iteratively refining prompts
- and checking their own solutions before turning to the AI.
- Throughout the course, their online self-assessments included adaptive feedback: instead of simply marking answers as right or wrong, the system offered hints, step-by-step prompts and suggestions on how to use AI tools as a thinking partner.
- In the control condition, students completed the same quizzes with standard right/wrong feedback, and no training or guidance on AI.
Importantly, the intervention did not encourage students to outsource solutions to AI. Rather, it framed AI as an interactive study partner to support self-explanation, comparison of strategies and self-regulation in problem solving.
We administered pre- and post-course questionnaires aligned with DigComp 2.2, focusing on five competences: information and data literacy, communication and collaboration, safety, and two aspects of problem solving (functional use of digital tools and metacognitive self-regulation). Using a difference-in-differences model with individual fixed effects, we estimated how the probability of reporting the highest level of each competence changed over time for the treatment group relative to the control group.
What changed when AI was taught and used in this way?
At the overall sample level, we found statistically significant improvements in three areas:
- Information and data literacy – students in the AI-training condition were around 15 percentage points more likely to report the highest level of competence in identifying information needs and carrying out effective digital searches.
- Problem solving – functional dimension – the probability of reporting the top level in using digital tools (including AI) to solve tasks increased by about 24 percentage points.
- Problem solving – metacognitive dimension – a similar 24-point gain emerged for recognising what aspects of one’s digital competences need to be updated or improved.
In other words, the AI-integrated teaching design was associated not only with better use of digital tools, but also with stronger awareness of digital strengths and weaknesses – a key ingredient of autonomous learning. Communication and safety competences also showed positive but smaller and more uncertain effects. Here, the pattern becomes clearer when we look at who benefited most.
A compensatory effect: AI as a potential leveller, not just an amplifier
When we distinguished students by their initial level of digital competence, a pattern emerged. For those starting below the median, the intervention produced large and significant gains in all five competences, with improvements between 18 and 38 percentage points depending on the area. For students starting above the median, effects were smaller and, in some cases, non-significant.
This suggests a compensatory effect: students who began the course with weaker digital competences benefited the most from the AI-based teaching design. Rather than widening the digital gap, guided use of AI acted as a levelling mechanism, bringing lower-competence students closer to their more digitally confident peers.
Conceptually, this challenges an implicit assumption in much of the literature – namely, that generative AI will primarily enhance the learning of already advantaged students, because they are the ones with the skills and confidence to exploit it. Our findings show that, when AI is embedded within intentional pedagogy, explicit training and structured feedback, the opposite can happen: those who started with fewer resources can gain the most.
From ‘allow or ban’ to ‘how do we teach with AI?’
For higher education policy and practice, the implications are twofold.
First, we need to stop thinking of digital competence purely as a prerequisite for using AI. Under the right design conditions, AI can be a pedagogical resource to build those competences, especially in information literacy, problem solving and metacognitive self-regulation. That means integrating AI into curricula not as an add-on, but as part of how we teach students to plan, monitor and evaluate their learning.
Second, our results suggest that universities concerned with equity and digital inclusion should focus less on whether students have access to AI tools (many already do) and more on who receives support to learn how to use them well. Providing structured opportunities to practise prompting, to critique AI outputs and to reflect on one’s own digital skills may be particularly valuable for students who enter university with lower levels of digital confidence.
This does not resolve all the ethical and practical concerns around generative AI – far from it. But it shifts the conversation. Instead of treating AI as an external threat to academic integrity that must be tightly controlled, we can start to ask:
- How can we design tasks where the added value lies in asking good questions, justifying decisions and evaluating evidence, rather than in producing a single ‘correct’ answer?
- How can we support students to see AI not as a shortcut to avoid thinking, but as a tool to think better and know themselves better as learners?
- Under what conditions does AI genuinely help to close digital competence gaps, and when might it risk opening new ones?
Answering these questions will require further longitudinal and multi-institutional research, including replication studies and objective performance measures alongside self-reports. Yet the evidence we present offers a cautiously optimistic message: teaching students how to use AI can be part of a strategy to strengthen digital competences and reduce inequalities in higher education, rather than merely another driver of stratification.
Concepción González García is Assistant Professor of Economics at the Faculty of Economics and Business, Catholic University of Murcia (UCAM), Spain, and holds a PhD in Economics from the University of Alicante. Her research interests include macroeconomics, particularly fiscal policy, and education.
Nina Pallarés is Assistant Professor of Economics and Academic Coordinator of the Master’s in Management of Sports Entities at the Faculty of Economics and Business, Catholic University of Murcia (UCAM), Spain. Her research focuses on applied econometrics, with particular emphasis on health, labour, education, and family economics.
