The rapid integration of AI into multiple spheres of everyday life has brought with it a growing consensus around the need to foster AI literacy across the general population (Kong et al., 2024; Pinski and Benlian, 2024). As AI increasingly influences how we work, communicate, learn and make decisions, understanding its functioning, implications and limitations has become a fundamental competence for responsible and informed citizenship in the digital age.

AI literacy goes beyond technical knowledge or coding skills: it includes a combination of cognitive, social, and ethical competencies that enable individuals to make informed decisions about AI in real-world contexts (Lee et al., 2024). In this sense, AI literacy represents not only a response to technological advancement, but also a civic and educational imperative (Long and Magerko, 2020).

Several attempts have been made in recent years to define and structure the concept of AI literacy. Long and Magerko (2020) described AI literacy as the ability to understand, use, and critically evaluate AI technologies, emphasizing not only conceptual knowledge but also ethical reflection and the capacity to interact meaningfully with intelligent systems. Building on this framework, the following section presents a detailed overview of the five dimensions of AI literacy proposed by Pinski and Benlian (2024), in conjunction with the main insights and contributions from the scholarly literature associated with each component.

Aligned with all said, and with the EU AI Act, UNESCO and other organizations, the OECD defined the AI Literacy as “the technical knowledge, durable skills, and futureready attitudes required to thrive in a world influenced by AI. It enables learners to engage, create with, manage, and design AI, while critically evaluating its benefits, risks, and ethical implications” (Organization for Economic Co-operation and Development - OECD, 2025a). This last definition, when considered alongside the five dimensions of AI literacy proposed by Pinski and Benlian (2024), highlighted the broad scope of AI Literacy (Figure 1).

According to Schiavo et al. (2024), cultivating AI literacy not only enhances individuals’ ability to identify, use, and critically assess AI technologies, but also fosters more positive attitudes toward their adoption by reducing uncertainty and anxiety. Their findings suggested that higher levels of AI literacy are associated with increased perceptions of ease of use and usefulness, which directly influence acceptance. Similarly, Moravec et al. (2024) emphasized that AI literacy enables individuals to better understand the role of AI in daily life, which is essential for informed adoption, for mitigating potential risks, and also for reducing negative prejudices and encouraging more balanced attitudes (Jacovi et al., 2021). Together, these perspectives underlined the value of AI literacy as a foundation for responsible engagement with technology, empowering citizens to navigate an increasingly AI-driven world with greater confidence and awareness.

Promoting comprehensive AI literacy is especially crucial in the field of education, not only for students who are growing up surrounded by AI technologies, but also for the educators responsible for equipping them to navigate and shape the future (Zhang et al., 2024). Educators need AI literacy to understand how AI can be integrated into teaching and learning, how it impacts assessment practices and how to guide students in responsible AI use (Smarescu et al., 2024).

In this regard, as Liu and Zhong (2024) pointed out, introducing AI literacy from early education onwards demands a scaffolding approach, adapted to the cognitive level of each stage of schooling. This approach also highlighted the importance of teacher preparation, both in-service and pre-service, as a prerequisite for meaningful integration of AI literacy into school curricula. This preparation requires not only the capacity to use AI tools effectively in their practice but also the ability to teach about AI and its implications, including its ethical and social dimensions (Casal-Otero et al., 2023).

Fostering AI literacy in primary teachers involves cultivating a multifaceted set of knowledge and skills (Casal-Otero et al., 2023). This includes a cognitive understanding of basic AI concepts and how AI works, enabling teachers to explain it to students and understand its capabilities and limitations (Castro et al., 2025). Furthermore, AI literacy requires the ability to critically evaluate AI outputs and tools, discerning their quality, appropriateness, and potential biases (Abdulayeva et al., 2025). As Castro et al. (2025) determined, this involves understanding ethical principles, data privacy, potential biases in AI systems, and the broader social and environmental impacts — all of which are essential competencies for teachers aiming to integrate AI meaningfully into their practice. In this regard, Castañeda et al. (in press) defined seven dimensions of implications of AI in Primary Education that should also be considered: instrumental/technical/functional, ethical, social, epistemological, ideological, political and commercial.

Furthermore, as Zhang et al. (2024) argued, understanding how AI systems function enables educators to help students recognize potential biases and address the ethical challenges these technologies may pose. Ethical awareness is a core component of AI literacy (Organization for Economic Co-operation and Development - OECD, 2025a; UNESCO, 2024), and teachers who possess this understanding are better positioned to guide students in navigating the social implications of AI (Du et al., 2024).

In this context, the use of AI in education should be driven by pedagogical objectives rather than by what is technologically possible (Abdulayeva et al., 2025; Arroyo-Sagasta, 2024). As AI has a transformative potential in teaching and learning (Holmes and Tuomi, 2022), with specific classroom applications, such as promoting critical thinking, facilitating classroom debates, working with prompts alongside students, and designing activities that help distinguish AI-generated content from human production (McDonald et al., 2025), teachers may also use AI for didactic planning, content creation, personalization of instruction, formative self-assessment tools, and case-based learning scenarios. Castro et al. (2025) reported that teachers increasingly perceive AI as an opportunity to support personalized learning, reduce administrative workload, and address the challenges of multigrade classrooms.

Despite the growing potential of AI in education, one of the main barriers to its meaningful integration in compulsory education is the lack of adequately trained teachers (Liu and Zhong, 2024). The literature does not provide a consensus on how to address this issue. While Chan (2023) emphasized the importance of comprehensive institutional frameworks for integrating AI education at the university level, such initiatives are still rare, especially in teacher education programs. Her model highlighted the need to align technical knowledge, ethical understanding, and policy awareness in order to create a sustainable foundation for AI literacy for professionals. However, when it comes to compulsory education, the implementation of AI-related content often depends on the initiative and preparedness of individual teachers (Ng et al., 2024). This makes teacher training a critical component of AI integration at early stages.

As noted earlier and highlighted in the literature review by Liu and Zhong (2024), the absence of AI-prepared educators is among the most significant obstacles to implementing AI education at early stages. Most teachers have not received specific training in AI, which in turn affects both their confidence and their capacity to engage with AI-related content in their classrooms. In this context, understanding teachers’ perceptions of AI becomes essential (Bae et al., 2024; Cervantes et al., 2024; Nikolic et al., 2024) for identifying the gaps, challenges, and opportunities associated with AI integration in schools.

Recent efforts to introduce AI literacy in secondary education provide useful insights into how these competencies can be meaningfully integrated into school curricula. Ng et al. (2024) identified a range of pedagogical strategies—including project-based learning, inquiry-based approaches, and ethical discussions—that have been employed to engage students in understanding AI concepts and implications. However, they also noted that many of these interventions depend heavily on individual teacher initiative and lack systemic integration into broader educational policies. While their review focused on secondary education, it underscored the need to extend such efforts to earlier stages. If the goal is to cultivate foundational AI literacy from a young age, then primary education must also be considered a strategic entry point—one that requires well-prepared teachers who can introduce AI in developmentally appropriate and pedagogically meaningful ways.

Teachers generally hold a positive perception of AI’s usefulness and perceived benefits for both their teaching and student learning (Emenike and Emenike, 2023; Cervantes et al., 2024; Liu and Zhong, 2024). Key benefits identified include increasing efficiency and saving time in their work (Lee et al., 2024), assisting with lesson preparation, enabling personalized learning and assessment (Liu and Zhong, 2024), and increasing student motivation and engagement (Zulkarnain and Md Yunus, 2023).

Regarding understanding and knowledge of AI, the sources indicated that most teachers have some general knowledge of AI, but their understanding of Generative AI (GenAI) is significantly less comprehensive (Liu and Zhong, 2024). This lack of knowledge and technological skill is perceived as a major challenge (Zulkarnain and Md Yunus, 2023), although some studies found participants rated their knowledge level as average (Smarescu et al., 2024).

Teachers also perceive significant challenges and concerns associated with AI integration (Liu and Zhong, 2024; Lee et al., 2024). Among the most prominent worries are academic integrity and plagiarism, and concerns about the accuracy and reliability of content generated by AI. There is also apprehension about students potentially becoming overly reliant on technology, leading to a decrease in their creativity and autonomy (Liu and Zhong, 2024). Lack of adequate institutional support and professional development is a frequently mentioned challenge (Smarescu et al., 2024). Practical issues like difficulties with class control, content distraction, poor internet connectivity, lack of infrastructure, teacher burnout/workload, and even computer anxiety are also reported challenges (Zulkarnain and Md Yunus, 2023). Ethical issues are a broader concern, encompassing potential bias in AI, security risks, and the need for establishing responsible standards for AI use (Lee et al., 2024; Yusuf et al., 2024).

Equally important, there is a clear perceived need for more support and training for teachers (Liu and Zhong, 2024): teachers express a desire for free access to AI resources, guidance on usage, technical support, and the development of practical teaching and assessment systems that integrate AI. Similarly, Smarescu et al. (2024) argued that enhancing teacher training can mitigate fears and misconceptions.

Recent studies have increasingly highlighted the urgent need to incorporate AI literacy into teacher education programs. Ravi et al. (2023) pointed out that, while interest in AI among educators is growing, many feel underprepared to integrate it into their teaching practice due to limited exposure and lack of structured training. Similarly, Hur (2024) emphasized that although teachers recognize the relevance of AI, professional development opportunities remain fragmented and insufficient, often focusing more on technical tools than on pedagogical or ethical implications (Arroyo-Sagasta, 2024). Some authors noted that most existing training initiatives still lack a coherent framework and fail to address age-appropriate pedagogical strategies, particularly for compulsory education (Lee et al., 2024; Liu, 2024). In this regard, Dilek et al. (2025) proposed a scaffolded model of teacher training that gradually builds both technical understanding and reflective competence, adapting to teachers’ prior experience and subject area. Finally, Eun and Kim (2024) provided empirical evidence that targeted and practice-based training programs can significantly enhance teachers’ confidence and readiness to engage with AI in the classroom. Together, these contributions suggested that, while awareness of the importance of AI training is growing, there is still a pressing need for more systematic, pedagogically grounded, and context-sensitive approaches to preparing educators for the challenges and opportunities of AI in education.

While much of the current discourse on AI training for teachers focuses on secondary or higher education, there is a growing consensus that AI literacy must be addressed from the earliest stages of schooling. Liu (2024) underscored that neglecting primary education in AI-related teacher training represents a missed opportunity to build foundational understanding and ethical awareness from a young age.

Understanding how individuals perceive, interact with, and reflect on AI requires the use of valid and reliable measurement instruments. In recent years, various studies have proposed different approaches to operationalize constructs such as AI literacy, awareness, attitude, and trust. These constructs are central to understanding how people engage with AI and how educational interventions might influence such engagement. For instance, Scantamburlo et al. (2025), in the PAICE study, provided a large-scale European framework for measuring AI awareness, attitudes toward AI use in different sectors, and trust in institutions involved in AI governance. Similarly, Nong et al. (2024) developed and validated a multidimensional scale to assess AI literacy, incorporating conceptual knowledge, ethical awareness, and confidence in using AI tools. Krause et al. (2025) also contributed to this field by analyzing how generative AI impacts perceptions and behaviors in higher education, linking constructs such as perceived usefulness, ease of use, and transformative potential. These contributions provide a robust foundation for the design and interpretation of empirical studies in AI literacy and inform the measurement approach adopted in the present research.

This study adopted a quantitative research approach. A quasi-experimental research design with a pre-test–intervention–post-test structure was employed. Although the pre- and post-intervention questionnaires were administered to the same group of participants, the data were collected anonymously and without individual coding. As a result, responses could not be matched at the individual level. For this reason, statistical analyses comparing pre- and post-test results were conducted using methods appropriate for independent samples, focusing on aggregated group-level changes.

This study employed a convenience sample, composed of 60 students enrolled in the Digital Innovation specialization of the Primary Education Degree of Mondragon Unibertsitatea (2023/24 and 2024/25). Participation was voluntary and limited to those who agreed to complete both intervention and associated assessments. No compensation was provided. All students completed the pre-test questionnaire, but only 48 of them participated in the post-test, resulting in a partial response rate for the second measurement point. This discrepancy reflects both attrition and voluntary participation patterns, and it was considered when interpreting the results. Given that all students enrolled in the specialization were included, no further sampling procedures were implemented.

Data was collected using a self-administered online questionnaire. The questionnaire titled “PAICE - Perceptions about AI in Citizens of Europe” is validated (Scantamburlo et al., 2025) and consisted of 14 items including Likert scale, dichotomous, multi-response items, and ranking (Figure 2). These items were organized according to three dimensions: awareness, attitude, and trust regarding AI. The instruments were a pre-test and a post-test, which were identical forms of a validated questionnaire to measure the dependent variables of the study. To ensure comparability, the pre- and post-intervention assessments were conducted using the same questionnaire instrument.

Between the pre- and post-test, participants completed an 8-h training session focused on AI literacy. The session introduced foundational concepts such as the principles of machine learning, how these systems are trained using data, and the distinction between different types of AI. Rather than focusing on advanced technical details, the training prioritized a clear and accessible understanding of how current AI systems function and are integrated into everyday tools, aligning with Kong et al. (2024), who consistently identify conceptual understanding as the foundational level of AI literacy (cognitive dimension).

Delivered in person, the session interwove explanatory content, illustrative examples, and practical exercises to support comprehension, which simultaneously nurtured the metacognitive and affective dimensions of AI literacy (Kong et al., 2024). Table 1 provides an overview of the content areas covered during the intervention, along with a description of each topic and the corresponding PAICE questionnaire’s dimensions targeted.

The questionnaire was hosted on a secure online platform (Encuesta.com). Invitations containing a link to the survey were provided to the target student population through the course Learning Management System (LMS) of Mondragon Unibertsitatea. A brief explanation of the study’s purpose, estimated completion time, and clear statements regarding voluntary participation, anonymity, and confidentiality of responses were provided at the beginning. Participants were required to provide electronic informed consent before accessing the questions.

Once the pre-test questionnaire was completed and following the intervention, the identical questionnaire (post-test) was administered to the same participants using the same LMS. Only the research team had access to the collected data.

All statistical analyses were conducted using Jamovi (version 2.6.26). In addition, Python was employed for data preprocessing, the calculation of Cronbach’s alpha for internal consistency, and the generation of visualizations.

Descriptive statistics (means, standard deviations, frequencies, and percentages) were computed to summarize participant responses at both measurement points. Although the pre- and post-test responses were collected from the same population, the lack of individual identifiers prevented a paired analysis.

Before examining the specific dimensions and their items, an overview was provided of how the Likert-scale items were analyzed. Initially, the data were assessed for normality using skewness and kurtosis statistics. For items that met the assumption of normality, Levene’s test was conducted to evaluate the homogeneity of variances. Based on the results of these assumption checks, appropriate inferential tests were selected following the parametric or non-parametric approach. Specifically, when both normality and homogeneity of variances were confirmed, the parametric Student’s t-test was used. In cases where normality was confirmed but equal variances could not be assumed, Welch’s t-test was applied. Conversely, when the assumption of normality was violated, the non-parametric Mann–Whitney U test was employed to compare groups. The significance threshold for all tests was set at α ≤ 0.05.

In case of dichotomous items, a chi-square test of independence was used to compare the proportion of affirmative responses (“yes”) before and after the intervention. This approach treats the two datasets as independent samples, enabling the detection of significant differences in overall response distributions.

For multiple-response and ranking items, the proportion of participants who selected each option was calculated separately for the pre- and post-test. These proportions were then compared to identify changes.

Descriptive statistics (means, standard deviations) and inferential statistics (t-tests, t-Welch and U-Mann–Whitney) were used to compare pre- and post-intervention scores on each dimension.

To contextualize the findings of the current study, we conducted an additional comparative analysis using aggregated data from the PAICE project (Scantamburlo et al., 2025), which surveyed over 4,000 European citizens across eight countries. While individual-level data were not available, the published aggregate results offer a valuable benchmark for interpreting our local findings.

For each of the three core dimensions —awareness, attitude, and trust— visual comparisons between the results of our study and those reported in PAICE were created. This comparison was performed item by item, where possible, using bar Figures to highlight patterns and discrepancies in response distributions.

Although no statistical inference is possible due to the lack of raw data from PAICE, this visual and descriptive comparison offers meaningful insights into how our participants’ perceptions align with or diverge from broader European trends. Particular attention was paid to items where our intervention appeared to produce significant shifts, allowing for reflection on context-specific effects and possible implications for AI literacy and policy at the local level.

For this comparison, we used the post-intervention results from our study, as they reflect the participants’ perceptions after being exposed to the educational intervention. This choice allows us to examine whether, and to what extent, the intervention brought participants’ levels of awareness, attitude, and trust closer to or further from those observed in the broader European population, as reported in the PAICE study (Scantamburlo et al., 2025).