Investigation of Generative AI Adoption in IT-Focused Vocational Secondary School Programming Education

Abstract

The application of artificial intelligence in education, particularly in learning programming, is gaining increasing significance. However, research on secondary school students specializing in IT at an early stage has received relatively little attention in this field. The aim of this study is to assess how vocational secondary school IT students utilize Generative artificial intelligence in learning programming. The study employed a survey-based methodology, where students with varying levels of knowledge were surveyed to understand their AI usage patterns. The sample consisted of students from vocational IT schools, and data were analyzed using descriptive statistics and independent samples t-tests. The results indicate that students with different levels of knowledge use AI tools differently, with ChatGPT being the most popular tool. The study further highlights that AI usage brings significant benefits, such as providing a personalized learning experience and enabling quick error correction. However, excessive reliance on AI tools may hinder students from acquiring fundamental programming skills. The findings support the idea that while AI can effectively complement teachers’ explanations, overdependence on it can be risky, potentially reducing students’ creativity and problem-solving abilities. The study emphasizes the crucial role of educators in teaching the responsible and ethical use of artificial intelligence. The results of this research offer new perspectives on the effective integration of Generative artificial intelligence into vocational secondary school programming education and suggest further studies to compare its applications at the university level. However, the study acknowledges certain limitations, such as the potential bias of self-reported data, which may affect the generalizability of the results. Unlike other studies, the age groups we surveyed, and the cohorts formed from them are nearly evenly distributed, making our sample representative of the region in question.

1. Introduction

Computer science plays a defining role in science, industry, and everyday life today. The rapid development of digitalization creates an increasing demand for professionals with appropriate programming and IT knowledge. Teaching computer science at the secondary school level is already crucial, especially for those studying in IT-focused vocational secondary schools (SIOV, 2025). These students gain a deeper insight into programming, which can lay the groundwork for their further studies and professional careers (Psycharis & Kallia, 2017). Unlike traditional high schools, IT-focused vocational schools-such as vocational secondary schools and technical schools-offer subjects that allow students to immerse themselves in programming and the operation of IT systems at an early age. These schools focus on areas such as algorithms and data structures, which aim to develop students’ problem-solving abilities; software development and coding, where students can learn the syntax of programming languages available in the current market and their practical applications; networking technologies, through which IT students learn the basics of network topologies, IP addresses, protocols, and network security, as well as gain knowledge that will enable them to build and manage networks. Additionally, through database management, they learn how to organize, process, and efficiently query data, while also gaining insight into the application possibilities of artificial intelligence (AI).

Among these, programming education is one of the most important elements, as it develops logical thinking, which is key to technological advancement and innovation. Programming is based on formal and algorithmic thinking: according to the literature, computational thinking involves skills such as abstraction, decomposition, algorithmic thinking, and debugging (Belmar, 2022). Additionally, it lays the foundation for university studies in fields such as applied computer science. The teaching and learning of programming can be examined from several perspectives: these include the development of problem-solving skills, mastering coding, and understanding the syntax of different programming languages. Learning can take place through traditional teacher-led instruction in group settings, or, through the use of Generative artificial intelligence, which has become a more widespread approach in recent times.

Many educators reject the application of artificial intelligence in education. School teachers’ reluctance to adopt AI tools often stems not from the tools themselves but from concerns about increased workload, lack of institutional support, and ethical implications such as data privacy (Cukurova et al., 2023). Teachers’ trust in AI-based educational technology is deeply tied to their perceived benefits versus concerns; the more ethical, pedagogical or technical uncertainties they anticipate, the less likely they are to integrate AI into their teaching practices (Viberg et al., 2024). It is important to emphasize that AI is not intended to replace humans, but rather to support both teachers and students, as well as to make the educational system more efficient (Mishra, 2024). AI technologies can be applied in numerous areas of education, such as personalized learning, automated assessment, the use of predictive models, and through intelligent robotics and AI-based educational tools (Annuš, 2024a). This study examines the potential of Generative artificial intelligence in vocational IT education.

One of the greatest advantages of Generative artificial intelligence in the field of education is its ability to provide an interactive and personalized learning experience. Unlike traditional educational methods, AI-based tools can adapt to the individual pace, abilities, and interests of students. Furthermore, nearly a decade before the widespread adoption of Generative AIs, researchers pointed out that AI could generate personalized learning materials, provide instant feedback, and assist students in problem-solving, regardless of the field of study (Le et al., 2013).

Nevertheless, the behavior and learning strategies of secondary school students may diverge considerably from those of university students. New data is available on students and their learning processes, enabling real-time measurement of learning processes and analysis of their complex relationships (Aksela et al., 2024). This is especially important in the age of artificial intelligence, as such analyses are essential for organizing the future of learning and teaching (Annuš, 2024a). Younger age groups generally exhibit lower levels of learning autonomy, while at the same time demonstrating higher-and sometimes excessive-levels of digital confidence. This is largely attributed to their frequent use of AI tools, a trend documented in recent discourse (Rubin, 2025). This makes them a particularly relevant and underexplored group for studying the impact of generative AI in programming education. Our study aims to fill this gap. According to researchers in the region we examined, university programming education tends to focus more on object-oriented programming and the teaching of corresponding programming languages, while programming languages taught in preparation for the high school graduation exam tend to favor structured programming (Végh & Czakóová, 2023). Furthermore, it can be stated that teaching and learning algorithmization and programming is a long process, especially for beginner programmers (Végh & Takáč, 2021). It is essential not only for students studying computer science but also in other fields that, in the digital era, students have adequate proficiency in using IT tools (Gombkötő et al., 2024). One effective approach to achieving this is the integration of artificial intelligence in programming education, as it can function not only as a programming assistant but also as a tool for generating a wide variety of tasks. Grishin’s research lists some positive aspects of AI tools in programming education, like assisting with information retrieval, bugidentification, and writing comments, and then AI tools can boost their efficacy and motivation to learn, when they were stuck (Zviel-Girshin, 2024).

This study aims to examine IT-focused vocational secondary schools in the southern Slovakia region, with particular attention to how 15–20-year-old students rely on Generative artificial intelligence in programming tasks, for what purposes, and which AI technologies they use. This study aims to compare the Generative AI usage of vocational secondary school students with the results of university-level research, mostly available in Web of Science and Scopus databases. Recent studies highlight the importance of digital competence and adaptive learning among vocational students (Barboutidis & Stiakakis, 2023; Tan et al., 2024; Xu et al., 2025), yet few explore how AI-supported tools affect attitudes, learning strategies, or ethical perceptions in this specific educational setting. Furthermore, Wild’s comparative study (Wild & Schulze Heuling, 2020) indicate that vocational students may possess distinct digital skills and support needs compared to cooperative students. Research on programming instruction in vocational schools also remains limited, even though programming is a key subject for developing logical reasoning and problem-solving skills in these institutions (Maryono et al., 2025). Scholars have repeatedly emphasized the lack of empirical evidence and practical models tailored to vocational contexts (Deschênes et al., 2024), underlining the urgency to better under-stand students’ actual usage patterns and perceptions of AI in this environment.

This study aims to address this gap through research questions exploring students’ Generative AI usage patterns, correlations with experience, preferences between Generative AI and teacher instruction, and their perceptions of Generative AI’s educational potential.

In the questionnaire survey, we sought to answer the following research questions (RQs), which we organized into four thematic categories based on their focus areas:

(A). 

Patterns of Generative AI Use

• RQ1: Do secondary school students use artificial intelligence in learning programming?
• RQ2: Which Generative AI systems are used by students?
• RQ3: For what specific purposes do students utilize Generative AI in programming?

(B). 

Role of Programming Experience

• RQ4: Is there a correlation between programming experience and Generative AI usage?
• RQ5: Is there a difference in the perceived usefulness of Generative AI among students with varying levels of programming experience?

(C). 

Perceived Effectiveness and Attitudes toward Generative AI

• RQ6: Is learning through teacher explanations or Generative AI-supported learning more effective? If one is more advantageous, what are the reasons for this?
• RQ7: What are the perceived advantages and disadvantages of Generative AI-based programming learning, and how do students view the future of Generative AI-supported programming education?

(D). 

Cohort Comparisons

• RQ8: Is there a significant difference in the rate of Generative AI usage for learning programming between the two cohorts—those who began programming before and after 2022?
• RQ9: Does cohort membership influence students’ perceptions of the usefulness of Generative AI in learning programming?

The findings of this study align well with the core constructs of the Technology Acceptance Model (TAM), which emphasizes the roles of perceived usefulness and perceived ease of use in predicting individuals’ acceptance of technology, a framework that dates back to the late 1980s (Davis, 1989). In the case of generative AI, students overwhelmingly highlighted accessibility, quick error correction, and tailored explanations as key benefits—elements closely related to these TAM dimensions.

2. Related Work

According to current studies, the use of Generative artificial intelligence in programming learning primarily focuses on university education. This is also confirmed by a recent empirical study published by Deriba et al. (2024). Philip’s work specifically discusses six possible directions for the use of ChatGPT-4 and GitHub Copilot in programming education, among others (Guo, 2023):

• Using AI to create customized programming tutorials for your own domain;
• Using AI to learn complex data visualization libraries;
• Learning to refactor exploratory code into more maintainable software;
• Learning about inherited legacy code;
• Learning new programming languages on demand within the context of your workflow;
• Questioning the assumptions your scientific code is making.

The literature supports the notion that students are increasingly using Generative artificial intelligence, although almost all studies agree that this technology has not only advantages but also potential drawbacks. Baillifard et al. (2025) conducted a controlled study to assess the impact of artificial intelligence on programming education. Their study shows that AI-augmented education has a positive effect on students’ performance. Students who actively used artificial intelligence achieved significantly better results, with an improvement of up to 15% compared to the control group (Baillifard et al., 2025). In line with this, Qureshi conducted an experimental study on the effectiveness of ChatGPT in solving programming problems. The results indicated that the group using ChatGPT outperformed the control group, achieving higher scores in less time (Qureshi, 2023). However, it is important to note that the study involved a small sample size, so the results cannot be generalized. In response to this, the literature highlights the need for larger-scale quantitative studies, which we aim to address with our own research. Researchers suggest that the better outcomes are attributed to the diverse uses of Generative AI, such as deeper understanding of concepts or more dynamic debugging (Biswas, 2023; Haleem et al., 2022).

Lytvynova et al. (2024) conducted a study to investigate which age group most frequently uses artificial intelligence in programming education. The results showed that first-year students, who had not previously studied programming, were the most active users. In contrast, third- and fourth-year students used artificial intelligence more cautiously, justifying their approach by citing the imperfections of existing models. Furthermore, students with less experience in using such technologies also expressed concerns regarding AI use (Keleş & Aydın, 2021). Additionally, research suggests that students voice concerns about excessive dependence on the technology, particularly due to the generation of overly complex code beyond their skill level and the lack of personal interaction (Simaremare et al., 2024). Zviel-Girshin’s study concludes by reinforcing that, instead of creating a collaborative learning environment with their peers, where students work together with artificial intelligence to enhance their understanding, students often allow AI to do most of the work. This dynamic undermines the learning experience and leads to students becoming passive recipients rather than active participants in problem-solving (Zviel-Girshin, 2024).

The application of artificial intelligence in programming education has been shown to yield numerous positive outcomes, according to research; however, the potential negative effects should not be overlooked. This section provides an overview of the currently available relevant literature, which predominantly focuses on university students. In the subsequent sections of our study, our primary aim is to assess the AI usage patterns in programming learning among vocational high school students in computer science, in order to explore the impacts of artificial intelligence at an earlier stage of education.

3. Materials and Methods

This study was conducted in accordance with relevant ethical guidelines. Prior to data collection, written permission was obtained from the leadership of all participating vocational secondary schools, as well as from the IT teachers who supervised the survey administration during class time. Students aged 18 and above provided written informed consent to participate voluntarily and agreed to the anonymous processing of their responses. No personal or identifiable data were collected.

Data collection was carried out between January and the first half of February 2025, while the evaluation was carried out for the second half of February 2025. The timing of the data collection was planned so that first-year students would have gained at least half a year of programming knowledge within the academic year. In selecting these institutions, we aimed to involve students from vocational high schools specializing in computer science, excluding classes from general high schools and non-computer science-oriented vocational schools.

The selection of institutions and students in this study was based on purposive sampling, as our goal was to assess the computer science and Generative AI usage habits of secondary school students in the South Slovak region. Accordingly, the selected institutions are located within this region. A significant portion of students from the vocational high schools in this region continues their studies at our university, making our research relevant for future university students as well. When selecting schools and classes, we ensured that the students were involved in programs focused on programming and computer science subjects throughout their four years of secondary education. The chosen institutions come from three major districts, where secondary schools offer computer science-related departments and programs. Therefore, the findings of this study are representative of the secondary schools in the South Slovak region that teach computer science subjects, as the selected schools and students reflect the local educational environment.

To answer the research questions, we conducted a quantitative questionnaire survey. As part of this, we collected data from several educational institutions in southern Slovakia, involving a total of 200 secondary school students who are pursuing studies in IT-related fields. The identities of the respondents were kept anonymous and unidentifiable.

Beyond demographic data, the questionnaire measured the following key constructs:

• Generative AI usage habits;
• Attitudes toward Generative AI;
• Experiences with Generative AI use in the educational environment.

During the development of the questionnaire, we took into account the tools and scales applied in the previous literature. However, the focus and target group of our study differ from earlier research, which primarily involved university students. Therefore, the measurement instruments were not adopted verbatim; instead, they were adapted to the specific needs, experiences, and context of secondary school students.

The language of the questions was adjusted to match the vocabulary and comprehension level typical of high school learners, avoiding overly theoretical or technical formulations. This adaptation was necessary as students from all grade levels participated in the study, including first-year students who had recently transitioned from primary school. Our aim was to reflect students’ practical learning habits rather than abstract constructs.

In developing the questionnaire, we consulted with vocational IT educators and university lecturers specializing in informatics didactics and programming education. Their expertise ensured that the questions were age-appropriate, context-relevant, and aligned with current pedagogical practices. In addition to our theoretical considerations, we drew upon direct teaching experience in vocational secondary programming education. This provided firsthand insight into the integration of tools such as ChatGPT into students’ learning processes, which we have consistently observed since its emergence in 2022. These practical observations, combined with academic consultations, contributed to a form of content validation by ensuring that the instrument was both theoretically grounded and pedagogically applicable.

When designing the scales, we considered that students tend to choose neutral answers-either out of convenience or to avoid taking a firm stance. To reduce this tendency, we used a 4-point Likert scale in our questionnaire, which encourages more directed responses along the positive–negative spectrum, such as expressing “somewhat positive” or “somewhat negative” attitudes (Zerényi, 2016).

For the analysis of the measurement results, we used Microsoft 365 Excel (Microsoft Corp., Redmond, WA, USA) with university licensing and IBM SPSS Statistics 25 (IBM Corp, Armonk, NY, USA) software tools, which allowed us to subject the data. In the development of detailed and extended statistical analyses, we rely on the guidebook by Kiss Tóth et al. on statistical software systems (Kiss Tóth et al., 2013).

Our research examines a secondary school environment that has not been extensively discussed in the existing literature. With our findings, we aim to provide new perspectives for educators, promoting a deeper understanding of IT vocational students’ knowledge of artificial intelligence, their habits, and the incorporation of Generative AI into practical lessons, as well as uncovering the students’ needs for the development of future programming education. Since this research did not rely on other similar studies in the region or neighboring countries-having found no such studies-we did not reference external validity. The sample, therefore, represents a specific regional focus, and the results are relevant and valuable for the target group under study.

3.1. Grouping of Participants

In this study, we created two distinct cohorts to analyze the use of artificial intelligence (AI) among students, considering when they started programming. The cohorts were defined as follows:

• Cohort 1: These are the students who are currently 15–17 years old and did not attend an IT vocational school before 2022. This group consists of students who had no regular programming education prior to 2022. These students began their programming studies after the widespread adoption of Generative AI technologies.
• Cohort 2: These are students who are currently 18–20 years old and were already vocational school students in 2022, meaning they programmed before 2022 and received programming education without the application of artificial intelligence.

Our goal was to assess the programming skills of the students. Accordingly, we provided three answer options for the students:

• Beginner: This group includes students who have encountered programming, development environments, and general syntaxes only during their school studies. They do not have any hobby-level programming experience.
• Basic knowledge: This group consists of students who already have some programming experience. They are familiar with the IDE environment they use and understand the basic syntax of the programming languages. They comprehend concepts such as variables, conditional statements, and loops.
• Advanced: This group includes students who are able to create algorithms and code them for tasks appropriate to their level of knowledge. They can solve problems that are in line with their studies, including competition problems and even tasks related to their high school graduation exams.

3.2. Research Hypotheses

To better understand the relationship between students’ programming background and their Generative AI usage, we formulated and tested the following hypotheses:

H₀₁. 

There is no significant difference in Generative AI usage between the two cohorts.

H₁₁. 

There is a significant difference in Generative AI usage between the two cohorts, considering the differences in programming experience between the groups.

H₀₂. 

There is no significant difference in the perception of Generative AI usefulness between Cohort 1 and Cohort 2.

H₁₂. 

There is a significant difference in the perception of Generative AI usefulness between Cohort 1 and Cohort 2, considering the differences in programming experience.

H₀₃. 

There is no significant difference in Generative AI usefulness ratings between the three programming proficiency groups.

H₁₃. 

At least one group rates Generative AI usefulness significantly differently from the others.

These hypotheses served as the foundation for the statistical analysis presented in the results section, and were tested using cross-tabulation, chi-square tests, and ANOVA methods where applicable.

4. Results

4.1. General Information

A total of 200 students were involved in the survey, specializing in the following fields: IT network technician, programming IT specialist, applied informatics, and mechanical programming. Among the respondents, 88.5% were male, and 11.5% were female, which indicates that in our region, informatics is still predominantly a field chosen by men. The age of the respondents ranged between 15 and 20 years, with an average age of 17 years.

According to the results of our survey, more than half of the respondents, 64%, categorized themselves as having basic knowledge, while the remaining 36% were distributed between the beginner and advanced groups. This distribution is valid in accordance with the general distribution of education, as the 64% primarily consists of second- and third-year students, while the remaining 11.5% reflects the proportion of first-year students who have not yet engaged in programming, and 24.5% are final-year students who will take their graduation exams within six months.

The golden age of artificial intelligence, particularly Generative artificial intelligence, began at the end of 2022 with the release of ChatGPT, which became widely available to users. In light of this, our survey also examined how many students had programming experience prior to this period. The aim of our research was to determine the number of students who likely learned programming without the support of artificial intelligence. The results show that the students who had previously studied programming are roughly evenly distributed. The exact distribution is illustrated in Figure 1.

4.2. Assessing the Frequency of Generative AI Use in Learning Programming

The second phase of our survey focused on students who already have experience with AI-assisted programming learning. To this end, we needed to filter out those individuals who had not yet used artificial intelligence in this area. Our results show that nearly three-quarters of the respondents have already used artificial intelligence for learning programming. The difference between the two groups is illustrated in Figure 2.

Based on the results presented in Figure 2, the survey further analyzed the data of the 144 students who selected the ‘yes’ option. We were primarily interested in which artificial intelligence tools high school students specializing in informatics had used in their programming learning process. Our question specifically measured three options: ChatGPT, Gemini (formerly Bard), and GitHub Copilot. It is important to note that students could list other options in nearly every question, including this one. Our results are presented in Figure 3.

As we can see, nearly 100% of the remaining 144 students had already used ChatGPT for learning programming, so we can conclude that, among the listed options, this Generative AI was the one most frequently relied upon by the students. The results show that 26 students had used other AI tools besides the ones listed for their programming tasks. Some examples include: Blackbox AI, Perplexity AI, Deepseek, Claude 3.5 Sonnet, My AI, and Sider AI—although the latter is a technology powered by ChatGPT.

In addition to examining overall trends in generative AI usage, the study also explored which programming languages students most frequently used AI tools with. Participants could select multiple languages. The data revealed a clear preference for Python (n = 98), followed by C++ (n = 67), HTML and CSS (n = 57), and C (n = 38). Other languages such as JavaScript (n = 21), PHP (n = 23), Lazarus/Delphi (n = 17), and Java (n = 10) were less commonly associated with AI usage, while Assembly (n = 7) showed the least usage.

The dominant use of generative AI with Python may reflect the language’s simplicity, readability, and beginner-friendly syntax, making it highly compatible with AI-assisted learning. Python is also extensively supported by generative AI platforms like ChatGPT, which often provide more accurate and useful outputs for Python-related queries due to its popularity in educational and AI contexts (Csóka, 2024).

The relatively high usage of AI with C and C++ could be attributed to their presence in traditional school curricula and the inherent difficulty in debugging these languages-something students reported AI helped with. Meanwhile, the significant frequency of HTML and CSS suggests that AI is also being used in web design contexts, where generative tools can quickly suggest layouts or correct syntax (Csóka, 2024).

This emphasizes the importance of considering AI-tool readiness per language in curriculum planning and future tool development.

4.3. Analysis of Generative AI Use in Students’ Programming Learning Based on Knowledge Levels and Cohorts

The aim was to investigate whether the differences in programming experience between the two groups of cohorts influenced their use of Generative AI in learning programming. We formulated the following hypotheses for the analysis:

H₀. 

There is no significant difference in Generative AI usage between the two cohorts.

H₁. 

There is a significant difference in Generative AI usage between the two cohorts, considering the differences in programming experience between the groups.

We examined the Generative AI usage of the two cohorts through a cross-tabulation analysis. The results indicated that approximately 30% of students in both cohorts do not use Generative AI for learning programming, while around 70% in both cohorts do. The Chi-Square test results were χ² = 0.373, p = 0.541. Since the p-value exceeds the significance level (0.05), we cannot reject the null hypothesis, meaning that no significant difference in Generative AI usage was found between the two cohorts. Based on the results, it can be concluded that there was no statistically significant difference in the frequency of AI usage between the two cohorts. This suggests that Generative AI usage is not necessarily dependent on whether students have programming experience. The details of the statistical results can be seen in

Table 1. Crosstabulation analysis between Cohort and Use of Generative AI.

			Used Generative AI for Learning Programming
			No	Yes	Total
Cohort	1	Count	26	60	86
		Expected Count	24.1	61.9	86
	2	Count	30	84	114
		Expected Count	31.9	82.1	114
Total		Count	56	144	200
		Expected Count	56.0	144.0	200.0

and

Table 2. Chi-Square Test.

	Value	df	Asymptotic Significance (2-Sided)	Exact Sig. (2-Sided)	Exact Sig. (1-Sided)
Pearson Chi-Square	0.373	1	0.541
Continuity Correction	0.204	1	0.651
Likelihood Ratio	0.372	1	0.542
Fisher’s Exact Test				0.634	0.325
N of Valid Cases	200

.

Our aim was to examine whether any correlation between students’ knowledge level and the use of AI-based learning tools can be observed at the high school level. For the statistical analysis of the data, we applied cross-tabulation analysis and the Chi-square test. The results of the measurement and evaluation show χ² = 13.099, and since the p-value is less than 0.05, the result is significant, indicating a statistically significant relationship between the students’ knowledge level and Generative AI usage. To evaluate the strength of the association, we calculated Cramér’s V = 0.181, which represents a small effect size according to Cohen’s criteria. This suggests a statistically significant, yet modest association between the variables. The full results are presented in

Table 3. Chi-Square Test.

	Value	df	Asymptotic Significance (2-Sided)
Pearson Chi-Square	13.099	2	0.001
Likelihood Ratio	13.435	2	0.001
Cramér’s V	0.181
N of Valid Cases	200

. The details of the statistical results can be seen in Table 3.

The results indicate that students with higher knowledge levels use AI-based learning tools more frequently. Among beginners, Generative AI usage is lower, while it is significantly higher among advanced students. This may suggest that the use of Generative AI tools has primarily spread among those who are already more confident in their learning and more easily integrate new technologies into their learning processes.

Examining the purposes for which AI technologies are used, the results show that the majority of students use Generative artificial intelligence for debugging, as 36% of respondents indicated this as the primary use. This is followed by generating ready-made code, which 30% of students use, especially when they need to write programs based on given descriptions. The third most popular use of Generative AI is for learning syntax, which represents only 11%, indicating that students value Generative AI support more for solving practical tasks and debugging rather than mastering the syntax rules of programming languages. A smaller proportion of students, only a few respondents, also use artificial intelligence to explain or comment on existing code. Figure 4 clearly illustrates the purposes identified by the respondents and helps to understand that Generative AI primarily assists students in solving practical programming problems.

Based on the purposes of AI usage, we can also observe that students primarily seek effective support in problem-solving, debugging, and coding, while theoretical or grammatical aspects, such as learning syntax, are less important to them when applying AI.

Students’ main motivation lies in the speed and ease of use of Generative artificial intelligence, which allows them to quickly solve problems and ease the learning process. Another key reason for using Generative AI is that students believe it provides clearer, more personalized explanations than traditional learning methods. This can be particularly appealing to beginner students who are looking for quick feedback and understandable explanations.

The results of the survey show that 32 students from the completely beginner group find the use of Generative AI justified, as they feel it helps them learn the basics of programming. However, the results also reveal a concerning trend, as some beginner students overly rely on AI and prefer to solve programming tasks with the help of artificial intelligence, rather than actively attempting to understand the problem and learn the steps of solving the code. Thus, AI does not always function as a learning assistant or tutor, but in many cases, it simply ‘replaces’ the student’s active participation in solving tasks.

This is especially true for students who have no prior programming experience. Such a passive approach could hinder deeper understanding and real learning in the long run, as problem-solving skills do not develop properly if students use artificial intelligence solely for completing tasks.

With the help of Figure 5, a more detailed illustration of the AI usage by beginner programmers can be provided, offering further insight into how students integrate artificial intelligence into their programming learning process.

Figure 6 provides a more detailed overview of why vocational high school programming students chose Generative AI to support their coding. The results indicate that students primarily turn to Generative AI in programming education for practical and immediate benefits. 34% of respondents cited ease of use, while 32% mentioned time-saving as their main reason, suggesting that Generative AI is often integrated into the learning process not necessarily for deeper knowledge construction but rather for efficiency. At the same time, 21% stated that Generative AI helps them better understand programming tasks, and 12% rely on Generative AI due to the absence of someone who could explain the material to them. These responses highlight that Generative AI functions not only as a tool for convenience but, in some cases, also fills a supplementary instructional role. Overall, the findings show that students are already using Generative AI autonomously, adapting it to their individual learning needs-an important implication for the future of education.

The third section of the questionnaire asked students to draw conclusions about the use of artificial intelligence in programming. Among other things, we asked them to compare the effectiveness of Generative AI to that of teacher explanations. 73% of the students believe that both teacher and Generative AI assistance are important in learning programming, while 17% prefer the teacher exclusively, and only 10% would prefer to learn programming exclusively with artificial intelligence. In response to our request to assess the AI’s helpfulness in programming, the respondents provided overwhelmingly positive feedback. Our results are illustrated in Figure 7.

In the study, the students’ perception of the usefulness of AI was measured using a Likert scale, and the responses of the two cohorts were compared to determine whether there is a significant difference between the two groups in their opinion on the usefulness of Generative AI in programming. The following hypotheses were formulated:

H₀. 

There is no significant difference in the perception of Generative AI usefulness between Cohort 1 and Cohort 2.

H₁. 

There is a significant difference in the perception of Generative AI usefulness between Cohort 1 and Cohort 2, considering the differences in programming experience.

The independent samples t-test was conducted to compare the perceived usefulness of Generative AI between Cohort 1 (ages 15–17) and Cohort 2 (ages 18–20). The Levene’s Test for Equality of Variances indicated that the assumption of equal variances was met (F = 0.001, p = 0.975). The results of the t-test showed no significant difference in the perceived usefulness of Generative AI between the two cohorts, t(142) = −1.367, p = 0.174. The mean difference between Cohort 1 and Cohort 2 was −0.133, with a 95% confidence interval of [−0.326, 0.059]. Therefore, we fail to reject the null hypothesis, indicating that the two cohorts did not significantly differ in their evaluation of Generative AI usefulness.

The categorization aimed to enable comparison of students’ attitudes toward the usefulness of Generative AI based on their differing levels of programming experience. After filtering the data, the resulting sample did not follow a normal distribution; therefore, a non-parametric statistical test—specifically the Kruskal–Wallis test—was applied to examine whether the perceived usefulness of Generative AI varied among the three groups. The purpose of the test was to determine whether there was a statistically significant difference between the groups in their evaluation of Generative AI’s usefulness. Below are our hypotheses:

H₀. 

There is no significant difference in Generative AI usefulness ratings between the three programming proficiency groups.

H₁. 

At least one group rates Generative AI usefulness significantly differently from the others.

According to the Kruskal–Wallis test results, the distribution of mean ranks across the groups was as follows: the complete beginner group had a mean rank of 65.39, the basic-level group 74.47, and the advanced group 72.64. The Kruskal–Wallis H value was 1.564, with a corresponding p-value of 0.458. Our results are illustrated in

Table 4. Ranks.

	Student Level	N	Mean Rank
Generative AI Usefulness Level	1	27	65.39
	2	96	74.47
	3	21	72.64
	Total	144

and

Table 5. Kruskal–Wallis Test.

	Generative AI Usefulness Level
Kruskal–Wallis H	1.564
df	2
Asymp. Sig.	0.458

.

These results indicate that although differences in programming experience were expected to influence perceptions of Generative AI usefulness, no statistically significant differences were found between the three groups. This suggests that the application of Generative AI in learning programming appears to be perceived as beneficial across all student levels, and programming experience does not constitute a major factor in its evaluation. As a conclusion, future research should consider additional influencing factors, such as the frequency of Generative AI use.

To examine whether the perceived usefulness of Generative AI was influenced by its specific use cases (e.g., debugging, generating code, developing project ideas, or learning new programming concepts), we conducted a multiple linear regression analysis.

The results indicated that the overall model was not statistically significant (F(4, 139) = 1.607, p = 0.176), and the model explained only a small portion of variance in perceived usefulness (R² = 0.044).

None of the individual predictors were statistically significant (all p > 0.05), suggesting that no particular use case was a strong determinant of how helpful students perceived Generative AI to be.

Furthermore, we summarized the frequency of the usage functions selected by the students and examined how perceived usefulness differed based on the most frequently used Generative artificial intelligence features. The four most commonly used functions were: debugging, developing project ideas, learning a new programming language, and generating new code. One-way analysis of variance (ANOVA) was used to assess whether there are significant differences in perceived usefulness across these different usage purposes.

The ANOVA results showed no statistically significant differences between the groups (F(3, 140) = 0.712, p = 0.547). This indicates that students’ perception of usefulness did not significantly depend on which function they used most frequently. The average usefulness scores for each group were as follows: students who used debugging had an average score of 3.04 (SD = 0.59), those who used project idea development had an average of 2.95 (SD = 0.51), those who learned a new programming language had an average of 3.33 (SD = 1.15), and those who generated new code had an average of 3.00 (SD = 0.61).

The results of the one-way ANOVA indicated that there were no significant differences in the perceived usefulness of ChatGPT or other Generative AI based on the most commonly used functions. These findings imply that perceived usefulness may be influenced by other factors beyond the specific functional purpose for which Generative AI was employed.

4.4. Critical Assessment of the Limitations of Generative Artificial Intelligence

We also asked the students to critically evaluate the potential negative effects of artificial intelligence that could impact the learning of programming. Figure 8 clearly illustrates that students identified the most significant disadvantage of AI as the risk of developing dependency, with 67% of respondents marking this as a major concern. Similarly, 54% perceived a decline in creativity, and 41% highlighted the generation of inaccurate or incomplete code as serious drawbacks. In contrast, the emergence of skill-level disparities and the lack of human interaction were considered among the most harmful effects by only 17% and 15% of the participants, respectively. Lastly, the unethical use of code received much less emphasis, with only 7% of respondents identifying it as a serious negative impact. This is particularly concerning, as if even at such a young age, students specializing in computer science are not concerned with ethical and legal use, it does not bode well for the future.

The literature highlights the imperfections of Generative language models, particularly in code generation (Lytvynova et al., 2024). In this context, we investigated what types of tasks the students encountered where the artificial intelligence could not provide assistance. More than half of the respondents (51%) reported encountering a problem that the AI could not solve. We asked them to specify the programming tasks where the Generative AI could not provide a proper solution. The students listed the following problems, with varying frequency:

• HTML and web syntax issues;
• For lesser-known programming languages, mainly difficulties in handling differences between versions;
• In cases of overly complex or lengthy code, the AI could not identify the error;
• The AI could not always provide an accurate explanation;
• After code completion, the original working functions of the code were lost;
• In Arduino programming, the AI could not provide any help;
• Database queries;
• Overcomplicating simple tasks-using unfamiliar functions that the student was not aware of.

Among the listed issues, several students criticized ChatGPT and its Plus version, particularly during HTML and CSS programming. In addition, Blackbox AI also received negative feedback regarding learning the Java programming language.

Reflecting on the results presented in Figure 5, we examined whether beginner students encountered any of the previously mentioned issues. The results indicate that more than 50% of these students experienced such problems. With this, we aim to draw the students’ attention to the importance of focusing on understanding the problem and using Generative AI primarily as an explanatory assistant rather than a task solver, as Generative AI is not capable of solving every programming task.

4.5. The Future Impact and Role of AI in Programming Education as Perceived by Students

Finally, we asked the students to express in their own words how they see the role of AI in the future of programming education. The responses we received are presented below, unchanged, using machine translation (used Deepl Free version).

For better readability and interpretation, student responses were organized into four main thematic categories. These groups were formed based on the content similarities of the answers, revealing the dominant trends in students’ attitudes toward Generative artificial intelligence in the context of programming education.

$$

Positive Impacts on Education—AI as a Learning Tool

Includes responses that highlight the advantages of using AI to support the learning process.

• “It’s important, because I haven’t found a teacher who explains better than AI. But this may partly be because AI answers when I don’t understand something and tries to explain it more simply.”
• “It may be important for students who fall behind in terms of knowledge compared to others.”
• “It could be useful when learning new programming languages if there isn’t someone to explain it. If we encounter problems, quick help can be a great advantage.”
• “If someone can talk to AI, they already know something about coding, and even if they don’t understand much, they can still learn better, because if there’s a problem with the code, AI can explain what went wrong. Since people learn from mistakes, AI can be a great help.”
• “AI plays a significant role in learning programming because it offers personalized learning experiences, taking into account the learner’s individual needs and progress. It helps in detecting mistakes quickly, fixing code, and simplifying the learning process by generating code snippets. This makes learning more efficient, especially for beginners.”
• “I think AI can help a lot and will continue to assist in learning programming in the future, because if you don’t understand something one way, it can explain it twenty other ways.”
• “It has a very important role because it teaches better than some teachers.”
• “If I have a question, sometimes it’s faster than Stack Overflow or Reddit, but not as reliable. If I need to quickly find something out while programming, I use AI.”
• “It’s still a good method for helping us with ideas or fixing potential errors.”

$$

AI as a Useful but Limited Tool—The Importance of Balance

Encompasses answers that recognize AI’s usefulness but warn about overreliance.

• “With more frequent use and teacher supervision, it can be very useful.”
• “I think despite all the challenges, there’s potential in it. The key is to use the technology properly. We shouldn’t overdo it, AI can be very helpful, but it’s just a tool, just like any other, and we should treat it as such!”
• “Finding the right balance will be very important.”
• “It will play a major role, as it can help a lot, but we need to find the right balance so that students don’t rely solely on AI.”
• “AI can help a lot, but it’s not good if someone relies too much on it. It can help with solving certain problems or be good for ideas. These days, AI can already create complete programs with few errors. If it continues to develop like this, it might be able to write entire programs without human help.”
• “A good option, but we cannot rely entirely on it.”
• “It’s useful because we can quickly get information, but we also need experience without AI to understand what it writes to us.”

$$

Potential Negative Impacts of AI on Education

Captures concerns regarding AI’s growing role-such as fears about automation replacing human expertise, reduced educational value, or the reliability and originality of AI-generated content.

• “It definitely has its drawbacks, especially when the person overly relies on the help given by AI. In the future, we may reach a point where AI-generated code can make anything work, and human coding will become obsolete.”
• “It can help a lot, and there are many programming-related help resources in it, but we must be careful with it because it can also become dangerous.”
• “I think the use of AI will be excessive in the future.”
• “I think it mostly helps in learning, but sometimes it writes complete nonsense, and it can mislead those who aren’t very experienced in programming.”
• “I see AI taking over the work of programmers.”
• “If AI continues to develop this fast, programming education will slowly become useless, because even laypeople can write simple programs with AI help now.”
• “AI won’t help in education, and I disagree with using AI. The information is inaccurate, creativity and problem-solving abilities are not strengthened, and the lack of interaction with people is a big problem.”

$$

The Future Importance of the Teacher’s Role

Reflects views stating that AI cannot substitute the human aspects of teaching. These responses stress the value of teacher guidance, clarity, and personal interaction in the learning process.

• “It won’t replace a good teacher or an explanatory video.”
• “It explains a lot, but it doesn’t reach the level of a teacher’s explanation.”
• “I think it’s an effective method, and it will only get better in the future. However, I believe that for mastering programming, a well-prepared and explanatory person is the best solution.”
• “Teachers won’t have much of a role.”

To deepen the understanding of students’ attitudes, open-ended responses regarding the role of AI in the future of programming education were thematically grouped and analyzed in connection with the closed-ended questionnaire results.

A significant portion of students emphasized the supportive role of AI in learning, often highlighting its accessibility, patience, and clarity. One student noted, “I haven’t found a teacher who explains better than AI,” while another emphasized, “It explains things twenty different ways if I don’t understand the first one.” The practical results suggests that students perceive Generative AI not just as a tool for task completion but as a dynamic learning partner that offers personalized and immediate support-particularly valuable for beginners and students without external assistance. These views align with the quantitative results: 34% of respondents cited ease of use and 32% mentioned timesaving as their main reasons for using Generative AI, while 21% indicated that it helps them better understand programming tasks. Notably, 36% of students primarily use Generative AI for debugging, and 30% use it for generating code-suggesting that learners are leveraging Generative AI as a practical, real-time tutor for overcoming coding challenges.

At the same time, many students expressed a balanced view, acknowledging the usefulness of Generative AI while cautioning against overreliance. “It’s just a tool, and we should treat it as such,” summarized one response, while another stated, “It can be very helpful, but we need to find the right balance so that students don’t rely solely on AI.” These perspectives echo concerns identified in the quantitative section about ethical boundaries and the reliability of AI-generated outputs because the most frequently selected drawbacks of AI were dependency, reduced creativity, and inaccurate or incomplete code as we demonstrated above in Figure 8. Interestingly, this cautious approach was more pronounced among students from higher cohorts and those who self-identified as more advanced in programming-indicating a growing awareness of AI’s limitations as students’ progress.

The third group of responses emphasized the potential negative impacts of AI on programming education. Students expressed concerns that overreliance on AI could diminish their own coding skills, originality, and independent problem-solving abilities. Additional issues included the generation of inaccurate or misleading code explanations and ethical concerns related to inputting and using code snippets provided by AI tools. These viewpoints reflect the presence of a more critical and cautious minority-students who, despite actively using AI tools, are aware of their limitations. Interestingly, in the quantitative part of the survey, ethical risks were rated as the least concerning aspect by most respondents. This contrast suggests that while the majority may not yet recognize the long-term risks, a smaller group is already demonstrating a more reflective and critical approach to the role of AI in education.

Finally, several responses clearly emphasized the continued importance of teachers in programming education. Many students noted that AI cannot replace the human aspects of teaching: “It won’t replace a good teacher or an explanatory video,” and “The best way to learn programming is with a well-prepared, explanatory person.” These remarks highlight a persistent belief in the irreplaceable value of interpersonal guidance. They also align with previous findings indicating that students still require structured and supervised AI use in classroom environments and continue to rely on teacher explanations-currently the second most frequently used learning resource between personal notes and digital technologies at the secondary school level (Annuš, 2024b).

Together, these qualitative themes not only reinforce the patterns seen in the quantitative analysis but also deepen the understanding of students’ nuanced views. While many recognize the transformative potential of AI, there is a clear call for structured integration, ethical oversight, and the preservation of the human elements of teaching-especially in the early, formative stages of programming education.

5. Discussion

Our research examines a question largely unexplored in the literature: the experiences of vocational high school students specializing in IT with Generative AI in learning programming. This target group is particularly important as they are not yet university students but already possess a high level of technological affinity. As a result, their relationship with AI tools and their usage strategies may differ from those of previously studied groups.

While the study did not explicitly control for socioeconomic status, the uniform access to digital tools among participants due to the institutional context mitigates potential disparities in AI tool usage. Students attending IT-focused vocational schools are generally expected to have access to laptops and internet at home.

Our results lead to several interesting conclusions, and our answers to the research questions allow comparison with prior university-level studies, in some cases confirming and in others contradicting them. Although overlaps can be observed in several aspects—such as the most popular tools or the main purposes of AI usage—the novelty of our investigation lies in variations in age, learning autonomy, and educational context influence how and to what extent AI is integrated into the learning process. While Zviel-Girshin’s study was based on students from one specific course-limiting generalizability (Zviel-Girshin, 2024), our research examines the effectiveness of programming language learning in various IT vocational schools through the application of artificial intelligence. Although education is based on a national curriculum, different perspectives and outcomes emerge because schools vary in local syllabi, programming languages taught, and instructor attitudes toward AI integration.

One of the goals of our research was to assess the frequency of AI use among high school students learning programming. Our findings show that Generative AI already plays a significant role in high school programming education, although the frequency of use varies by knowledge level. Based on our survey, beginner students use Generative AI less frequently than advanced students. This observation highlights that technological confidence does not necessarily correlate with early adoption of AI tools, especially in a school environment. AI adoption appears more common among confident learners who more readily integrate new technologies. This contradicts the findings of Lytvynova et al.’s university-level research, where lower-year students were more likely to use AI (Lytvynova et al., 2024). While in Lytvynova’s study the majority of respondents were first-year students, which resulted in the study not equally representing all academic years, in our research, we evenly analyzed all academic years of IT vocational schools. Additionally, by creating cohorts based on the students’ age, we examined their openness to using Generative AI and evaluated its effectiveness. At the same time, our findings partially confirm that high school graduates are likely to be the first university students to adopt Generative AI in their learning. This transitional phase has received little prior attention, positioning our study as a bridge in understanding AI use between secondary and higher education. Overall, while first-year university IT students are typically the most active AI users, lower-year vocational students are the least likely to use it. This answers RQ1 and RQ4.

Another aim of our study was to identify the most commonly used artificial intelligence technologies among students. Our results indicate that generative AI, especially ChatGPT, was by far the most popular tool. Other AI technologies such as BlackBox AI and Deepseek were also frequently mentioned. This answers RQ2.

Students also noted programming limitations of each AI, as discussed earlier. These limitations relate to other research questions, including RQ3, which explores students’ purposes for using generative AI in programming. Most respondents cited personalized code explanations, self-paced learning, debugging, and new code generation as primary uses. These use cases align with the work of other researchers as well (Simaremare et al., 2024; Haindl & Weinberger, 2024a, 2024b; Hanifi et al., 2023; Vaithilingam et al., 2022). While purposes such as code generation or debugging are consistent with those reported in higher education, the depth of understanding and critical reflection on AI outputs is often more superficial among secondary students. This contrast highlights not only differences in usage patterns, but also in metacognitive engagement and levels of self-regulation. Our findings clearly indicate that high school students value teacher involvement in the use of AI tools and consider it important that educators themselves be competent in these technologies. One of the key conclusions of our study at the secondary level is that teacher guidance is essential-not only in terms of technical use but also in fostering a critical approach to AI. Students may benefit from being encouraged to set explicit learning goals before engaging with AI, reflect on the relevance and reliability of its outputs, and compare these with their own solutions. Such structured reflection activities and teacher-facilitated discussions can help learners develop a more thoughtful stance toward AI-generated content, distinguishing between beneficial support and over-reliance that may hinder the development of independent problem-solving skills.

Qureshi’s study identified several limitations of ChatGPT in generating correct code or programs. Specifically, it struggled to interpret questions accurately and lacked the capacity to apply knowledge effectively to produce precise solutions. In some instances, even after repeated prompts to revise the code, the output failed to compile within the IDE, leading to frustration among users. Furthermore, inaccuracies in the output caused test cases to fail (Qureshi, 2023). Our own findings align with these observations. Similar issues were detected during our research with high school students, many of whom expressed frustration due to these shortcomings. This is particularly important to consider, as an overreliance on AI tools-especially when they fail to provide the expected support-may quickly diminish students’ motivation to engage with programming. It is also crucial to emphasize that excessive dependence on AI is not advisable, as it can hinder the development of independent problem-solving skills.

Our survey asked students to draw conclusions about learning programming with Generative AI and summarize the advantages and disadvantages of relying on AI. The results showed that students require both teacher-led instruction and AI assistance in their learning process. Many students highlighted the desire for instructors to be more competent in using Generative AI and to incorporate it more frequently in practical training. Some students would prefer AI over teacher explanations, as they found that Generative AI provides much more personalized explanations. In the words of the students, AI can explain a single task in up to twenty different ways. This suggests that even in high school-level programming education, personalized learning experiences are essential. Students emphasized the importance of finding the right balance, as excessive reliance on AI should be avoided. Overuse of code-generating tools could result in a superficial understanding of programming principles and concepts. This not only limits students’ ability to write and understand code independently but also makes them more vulnerable to handling technical problems and algorithmic errors. Several studies support this finding (Zviel-Girshin, 2024; Bird et al., 2022; Lau & Guo, 2023). Regardless of the level of the educational institution, excessive reliance on AI does not aid in the development of problem-solving skills for students learning to program. This is supported by the fact that students mentioned the greatest disadvantages of AI as the development of dependency, reduced creativity, and incomplete or inaccurate code. Additionally, issues such as programming solutions that are not appropriate for their level, lack of human interaction, and unethical use of source code-such as sharing with Generative AI-also emerged, reinforcing the viewpoints of other students (Simaremare et al., 2024). This answers our remaining research questions, RQ6 and RQ7. One of the key conclusions we have drawn is that this issue primarily affects total beginner students, more than half of whom use AI for coding without fully understanding the generated code. This approach is not ideal, and students need to be guided to avoid it. This warning is also supported by the study of (Saari et al., 2024).

In response to RQ5, statistical tests were conducted to examine how students evaluate the applicability of Generative artificial intelligence in learning programming. Based on the results, no significant differences were found among the three groups categorized by programming proficiency levels. The results imply that perceived usefulness is not exclusively tied to prior programming knowledge but may instead relate to broader factors such as individual learning strategies or exposure to technology. This is also supported by the open-ended responses of the students. The qualitative responses revealed that the most valued benefits included easy access, fast feedback, and personalized code explanations. Students highlighted that AI-assisted debugging was especially helpful in identifying specific syntactic or semantic errors and understanding the logic of their own solutions. However, the data also showed that some learners chose to adopt AI-generated code without fully understanding it, especially when unfamiliar functions were used. This tension between support and overreliance points to the necessity of guided integration, where educators facilitate AI use to deepen-rather than replace-students’ programming comprehension.

Although we hypothesized that students’ perceived usefulness of ChatGPT and other Generative AI would vary depending on how they used the tool (e.g., for debugging, idea generation, code generation, or learning new programming concepts), the regression analysis did not reveal any statistically significant relationships.

This suggests that students generally perceived Generative AI as helpful regardless of its specific application. One possible interpretation is that students’ overall attitude toward AI support in programming tasks may be shaped more by their broader experience or disposition toward technology rather than by the nature of the task itself.

To answer RQ8 and RQ9 we also examined whether generational differences-based on when students began learning programming-affect AI adoption. Cohort 1, comprised of younger students who began programming after the release of ChatGPT, was compared to Cohort 2, who started before Generative AI tools became widespread. Contrary to expectations, the frequency of AI usage did not significantly differ between the two groups. Both cohorts demonstrated similar usage patterns, indicating that the timing of introduction to programming (pre- or post-ChatGPT) does not strongly influence whether students adopt AI. This may reflect the rapid normalization of AI tools in educational contexts, regardless of when a student’s programming journey began. Finally, we analyzed whether the two cohorts differ in how useful they perceive Generative AI to be in their programming studies. Although there was a slight difference in mean scores, with older students rating AI marginally higher, the difference was not statistically significant. This finding warrants deeper interpretation. One possible explanation is that the rapid diffusion and intuitive interfaces of generative AI tools may have leveled the playing field, allowing even novice students to use these tools with similar intensity as their more experienced peers. This trend aligns with findings by Prather (Prather et al., 2024), who observed that novice programmers frequently used generative AI tools regardless of experience level, although often without deep understanding. Moreover, since AI integration into classroom instruction is still limited, as indicated by student feedback, the adoption of these tools may be more influenced by individual initiative than by prior programming education. Shahzad (Shahzad et al., 2024) similarly found that perceived usefulness and ease of use, rather than technical background, were primary drivers of ChatGPT adoption among students. This outcome suggests that AI use is shaped less by formal experience and more by accessibility, perceived usefulness, and personal learning preferences-an insight that calls for further qualitative investigation.

While this study provides valuable insights into the AI-related learning behaviors of secondary school students, several limitations must be acknowledged. Although the questionnaire was developed with careful attention to content relevance and linguistic appropriateness for secondary school students, no formal psychometric validation procedures-such as a pilot test, factor analysis, or test–retest reliability checks-were conducted. The questions were reviewed by experienced educators with PhD. in field of informatics didaction and programming education, which provided a form of informal content validation. However, the absence of statistical validation means that the internal consistency and measurement precision of the instrument could not be formally established. This limitation should be considered when interpreting the results, particularly in terms of generalizability and replicability. Next, this study employed purposive sampling, which is appropriate for exploratory research focused on specific populations-in this case, secondary school students learning programming. However, purposive sampling limits the generalizability of findings to the broader student population. The focus on IT-oriented students may introduce a selection bias, as these learners likely possess higher levels of digital literacy and may be more inclined to experiment with emerging technologies. Future studies employing random or stratified sampling across diverse school types and regions would be needed to validate and extend the findings presented here. Furthermore, part of our data is based on self-reported experiences and perceptions, which are inherently susceptible to inaccuracies in self-assessment. Students were asked to self-identify their programming level. While this approach allowed for efficient data collection across a large sample, it also introduces potential bias and subjectivity, as students may either underestimate or overestimate their own skills. Future research could mitigate this limitation by incorporating objective skill assessments or teacher-evaluated proficiency levels. Another limitation lies in the rapidly evolving nature of AI tools and platforms, with new models and updates emerging almost daily potentially rendering some findings outdated over time. This underlines the importance of continuous, adaptive research in this field. One of our planned next steps is to conduct a comparable survey among IT-focused students at local universities and compare those results with the present secondary school data. Given that many future university students in our region come from the very vocational schools involved in this study, such a comparison could reveal how AI-supported learning behaviors evolve over the transition into higher education. To support deeper understanding, we are also designing longitudinal interventions where students’ access to large language models would be routed through controlled APIs (application programming interface), allowing us to log real-time interactions. These usage logs could help identify students’ authentic engagement patterns and shed light on which programming tasks most frequently trigger reliance on Generative AI. This data could, in turn, support the development of more personalized programming curricula, enabling instructors to better distinguish between tasks that foster independent problem-solving and those where AI support becomes a crutch.

6. Conclusions

Our study examines the AI-supported programming learning habits of IT-focused vocational high school students, a topic that is rarely addressed in the literature. Even in high school programming education, maintaining the right balance is essential. The objectives of programming education should be defined within clear frameworks already during high school studies, not only in university-level education. The fundamental goal of programming education supported by innovative technologies should be to teach responsible and ethical use of artificial intelligence throughout students’ studies. This includes teaching them how to use AI tools and how to apply them effectively and ethically (Saari et al., 2024). As our survey results also showed, students expect this, as they expect educators to be competent in incorporating AI technologies into practical classroom activities. Teachers must play a key role in ensuring this guidance and direction, especially in the ethical use of programming codes. However, our research indicates that, unfortunately, a large portion of the students take this issue lightly, and in the survey, they only marked it as the least risky aspect. As trained IT professionals, future programmers must have a clear understanding of the ethical and legal use of software, and the same applies to managing the source codes that build the software.

The results of our descriptive and frequency analyses indicate that students most often used Generative AI for debugging and code generation, while fewer reported using it for conceptual learning or project planning. Cross-tabulation results further revealed interesting patterns in how these usage habits relate to students’ programming experience levels and perceived usefulness. Our cohort comparison showed no significant difference in perceived usefulness between beginner and advanced programmers, suggesting that AI tools are seen as similarly helpful across proficiency levels. Moreover, statistical analyses including multiple linear regression and ANOVA found no significant relationships between specific AI usage types and perceived usefulness. These results suggest that students regard Generative AI as broadly supportive across different programming tasks, not tied to a single purpose. The consistency of these findings across cohorts and function types suggests a generalized acceptance and perceived value of Generative AI in programming education within our region’s vocational secondary schools specializing in computer science, regardless of students’ experience level or the specific way the tool is applied.

Our survey results indicate that, in pedagogical practice, teachers should not merely permit the use of AI tools but should deliberately integrate them into the curriculum. Although the sample size was adequate for conducting basic statistical analyses, one of the study’s limitations lies in the reliance on self-reported data. Therefore, we encourage computer science educators to purposefully incorporate Generative AI into classroom practice through supervised, structured tasks. This approach could pave the way for future longitudinal research that systematically monitors the personalized and practical impact of Generative AI on students-potentially tracking progress from the first year through to graduation at the class level. It is advisable to design task structures in which students compare AI-generated code with their own solutions through guided analytical questions and reflect on the correctness and limitations of the AI’s suggestions. Institutions must concurrently provide the necessary infrastructure-up-to-date AI licenses and a secure network environment-and offer regular, practice-oriented training to prepare teachers for using AI platforms and addressing ethical and data-protection issues. Educational policymakers, in turn, should establish a framework that both fosters digital pedagogical innovation and regulates the application of Generative AI, while supporting the continuous development of teachers’ digital competencies. We recommend that the organization, digitization, and modernization of education be guided by the European Union’s 2025 “A European Model for Artificial Intelligence” (European Commission et al., 2025). Under this framework, institutions should establish AI-friendly learning environments that ensure personalized, inclusive, and ethical AI applications governed by strict data-protection and compliance standards, while guaranteeing continuously improving learning quality through ongoing teacher training and robust technological infrastructure support.

In the future, we aim to broaden the focus of our research. Our goal is to involve students from universities where the computer science departments specialize in applied informatics and where programming and software development are key subjects. Since most of the literature is based on smaller group studies typically focusing only on university students’ experiences, the findings of our research are particularly noteworthy, as they also consider the experiences of vocational high school programming students. In the future, we aim to complement our research by conducting a practical comparison of student habits, practical outcomes, and different programming languages, integrating results from university students. This will provide instructors with valuable insights that can help them better prepare for and organize AI-supported programming education.