Leveraging AI in vocational education: a systematic review of theory, barriers and opportunities

This paper explores how artificial intelligence (AI) might reshape vocational education and training (VET), both in everyday practice and in how the sector is perceived. The focus on VET is deliberate, based on several key reasons. First of all, its competency-based pedagogy and practice-oriented nature mean the opportunities and challenges associated with AI look different from those in more academically oriented settings. Secondly, many VET systems continue to navigate questions of reputation, status and competition for learners. Added to this, the pace of technological change in industry is placing fresh expectations on providers to demonstrate agility and relevance.

VET encompasses learning activities designed to equip learners with knowledge, skills and competencies for specific occupations or trades. UNESCO also refers to VET as technical and vocational education and training (TVET), which includes education, training and skills development in the occupational fields, production, services and livelihoods (Misselke et al., 2024; UNESCO, 2015). VET (or TVET) typically emphasises practical, hands-on learning experiences directly linked to employment outcomes, distinguishing it from traditional academic education. For this study, we define AI in education as computational systems that can perform tasks usually requiring human intelligence. This includes machine learning algorithms that personalise learning experiences, natural language processing tools that automate assessment and feedback, computer vision technologies that evaluate practical skills and robotics systems that simulate real-world training scenarios.

The world of work is changing rapidly, with the fourth (4.0) and now fifth (5.0) industrial revolutions reshaping vocational education. Industry 4.0, coined by Klaus Schwab in 2015, captures the pace of technological change, particularly its impact on workplaces (Schwab, 2016). Industry 5.0 builds on this, focusing on blending human creativity with AI to boost productivity and innovation. As technology accelerates, VET providers face rising competition and shifting workforce demands. To stay relevant, VET must evolve and seek ways to attract and retain a diverse range of learners while also preparing them for a future in flux (Misselke et al., 2024). AI has the potential to transform marketing, student engagement and educational delivery across the VET landscape. Literature on technology use in education highlights the need for strong theoretical frameworks to guide implementation, without which technological adoption risks becoming piecemeal rather than transformative, particularly in VET, where pedagogy, industry relevance and workforce needs must all align.

This study uses a systematic literature review to analyse how AI has been applied in VET over the past 5 years. It explores the theoretical frameworks underpinning current research and evaluates AI's potential to rebrand and modernise vocational learning. Through analysis of peer-reviewed work, key themes emerge around attitudes towards AI and innovative integration approaches. The review explores both the opportunities – like personalised learning, streamlined admin and improved outcomes – and challenges, such as ethics, implementation barriers and staff development. This paper offers a current overview of AI in VET and identifies areas where further research is needed.

Discussions about AI often feel abstract, but its influence is already woven into everyday work across many sectors. Rather than operating at the margins, AI now sits inside decision-making processes, routine tasks and long-term planning. Estimates from McKinsey, suggesting trillions of dollars in added economic activity by 2030, give a sense of the scale involved, but the real story is in how organisations are already changing (Bughin et al., 2018).

The World Economic Forum (2020) notes that organisations are increasingly using AI to optimise operations, decision-making and innovation, reshaping work and business practices in the process. There is evidence that care teams increasingly rely on automated image analysis to support diagnosis; finance companies use algorithms to flag suspicious transactions or guide investment strategies; and manufacturers are turning to predictive systems that anticipate machine faults before they occur. Retailers, meanwhile, use AI-infused tools to understand customer behaviour and manage complex supply chains (Verma and Srivastava, 2021). None of these developments looks the same, yet together they show a pattern of industries leaning on AI to navigate complexity and speed.

Some well-known examples illustrate this shift. Amazon’s recommendation systems quietly nudge consumer behaviour; DeepMind’s successes in strategy games have become something of a reference point for AI capability; and developments in autonomous vehicles, such as those pursued by Tesla, show how AI underpins new forms of transport (Davenport and Ronanki, 2018). These examples are more noticeable and publicised, but the less visible changes, such as back-office automation, are equally important.

In Australia, approximately 7.2 million workers (50% of the workforce) will need to adapt to the potential impact of generative AI (Mandala Partners, 2023). Within this group, 3.9 million workers will have their roles disrupted, whilst 3.3 million will see their roles augmented by AI. Augmented roles use AI to enhance their capabilities and shift their focus to higher-value tasks. By 2030, the widespread adoption of generative AI is expected to unlock an estimated A$115 billion in annual economic value, increasing productivity and innovation across various sectors (Australian Government, Department of Employment and Workplace Relations, 2024). Unlike previous technological advancements that primarily affected lower-skilled jobs, generative AI is expected to impact highly skilled professions, such as those that often require university qualifications (Mandala Partners, 2024). According to the Productivity Commission (2024) reports, demographic impacts vary, with women more likely to be disrupted by generative AI than men, and younger workers (particularly Gen Z) expected to be the most affected age group. Workers with bachelor’s degrees and higher are also more likely to be affected by generative AI.

As AI reshapes work across industries, education faces a double challenge: it must both deploy these technologies to improve how it operates and prepare students for jobs that AI will fundamentally alter. Higher education has been the quickest to experiment with AI; universities now use it in multiple ways, some mundane, others ambitious. Chatbots field student queries around the clock (Montenegro et al., 2019); adaptive learning platforms adjust content difficulty based on how individual students perform. Behind the scenes, predictive analytics scan enrolment and engagement data, flagging students who might be struggling before problems escalate (Alamri et al., 2022; Tsai et al., 2019). Recruitment teams deploy algorithms to manage applications and target prospective students (Daniel, 2015). Some institutions have gone further, building “Internet of Things” (IoT)-enabled campuses that monitor foot traffic, library usage and learning space occupancy to optimise the student experience (Khanna and Dhingra, 2018).

The pedagogical applications have evolved as well. Platforms using AI can now adjust what students see and how they are assessed, matching content to individual progress and gaps in understanding (Zawacki-Richter et al., 2019; Kasneci et al., 2023; Herodotou et al., 2019). Some systems provide instant feedback on written assignments, though questions remain about whether this helps or hinders the development of greater analytical skills. Universities have begun treating AI literacy as a core graduate capability. After all, students entering law firms, hospitals, design studios or corporate offices will work alongside these technologies, so understanding their capabilities and limitations matters as much as knowing how to use them (Mandala Partners, 2024).

VET institutions face different constraints. Their work is shaped by competency frameworks that dictate what students must demonstrate, vocational outcomes that determine programme success and employer partnerships that drive curriculum relevance. At the same time, VET operates with tighter budgets than universities and maintains a sharper focus on practical, workplace-ready training. There is less room to experiment with unproven technologies or approaches that might not translate directly into job readiness. The question, therefore, shifts from whether AI could change vocational training in theory to what can actually work in settings where resources are limited, industry relevance is essential, and success is measured by whether graduates can perform the jobs they have trained for.

The VET sector presents unique opportunities for AI integration, aligned with its competency-based approach and practical skills focus. Unlike higher education's emphasis on theoretical knowledge, VET's focus on measurable competencies creates ideal conditions for AI personalisation by mapping vocational skills to shifting industry needs. AI systems can track skill development with precision, providing real-time feedback that directly supports workplace readiness.

Australia is expected to face a shortfall of 370,000 digital workers by 2026 (Future Skills Organisation, 2024b), highlighting the urgent need to align education and training with evolving labour demands. As generative AI accelerates the shift to a skills-based labour market (Future Skills Organisation, 2024a), VET's role becomes increasingly vital. Yet, implementation has lagged: only 20% of VET institutions have adopted AI in their digital strategies (Dodd, 2023b; ReadyTech, 2024). However, this slower adoption rate should not be interpreted as resistance to innovation but rather as a reflection of the sector’s resource constraints and the need for specialised AI applications that are tailored to vocational contexts.

Although 61% of VET institutions rate digital transformation as a high priority, only 44% have formal strategies in place. Key barriers include staffing, skills shortages, training gaps and wage limitations (Compliance and Quality Assurance, 2023; IT Brief Australia, 2024). Some providers use basic AI-enhanced learning management system platforms for content delivery, though most lack advanced personalisation. Current AI applications in VET are largely confined to administrative efficiencies (Southgate et al., 2019; Gekara et al., 2019).

The competency-based nature of VET makes it well-suited for AI personalisation that aligns curricula with labour market demands (Bakhshi et al., 2017; Waschull et al., 2020). Simulations and virtual labs offer immersive, workplace-relevant learning experiences (Bédard-Maltais, 2017), and AI can also support apprenticeships by integrating classroom and on-the-job learning (Popenici and Kerr, 2017). Moreover, AI can address equity concerns by providing tailored support to learners with additional needs (Kohnke and Zaugg, 2025). Policy frameworks are evolving in response. The ‘AI-Empowered Workforce: Priority Framework’ (Future Skills Organisation, 2024a) calls for new pathways from VET into tech occupations. It identifies qualifications in marketing, communications, conveyancing and insurance broking as priorities for AI-related reform.

A broader shift is also underway, with demand increasing for microcredentials and accredited short courses, while traditional certificates decline. The National Skills Agreement aims to address workforce shortages, with institutions focusing on student retention, experience (73%) and staff productivity (71%; Department of Employment and Workplace Relations, 2024). In England, the new Skills England body seeks to close skills gaps and support job mobility (Department for Education, 2024). However, challenges still remain. Ensuring digital inclusion is critical to avoid exacerbating existing inequalities. Policy must balance innovation with risk management. At the same time, competitive market policies risk undermining vocational education's role as a public good (Dodd, 2023a; Productivity Commission, 2024). Certain groups, especially women and younger workers, face disproportionate risks from AI-related disruption due to occupational segregation and a lack of complementary skills (Productivity Commission, 2024).

Digital transformation is vital for the future success of the VET sector, both in attracting learners and aligning with labour market needs. Yet, despite the wide deployment of AI in industry, its systematic integration in vocational education remains underexplored in academic literature (Wu, 2021). Many discussions of AI in education draw heavily on higher education research, with limited attention to the specific realities of VET. Studies that foreground the experiences of vocational learners, teachers or leaders are still rare. This gap provides the rationale for the review at the centre of this paper.

Concerns about the status of vocational education have circulated for decades, and they continue to shape debates across policy and research. The tension is not simply a historical curiosity about different types of knowledge: it is tied to the ways VET is positioned and talked about in contemporary education systems. Although earlier philosophical discussions (e.g. those by Carr and Kemmis, 1986, or Buckingham, 2021) help explain where some of these hierarchies originated, today the issue feels more connected to how VET is framed, funded and evaluated in relation to higher education.

In countries such as England and Australia, VET has often been cast as a practical response to labour market needs and as a pathway for learners who are perceived, whether rightly or wrongly, as less suited to academic routes. This has long been reinforced by policy decisions shaped by market-driven reforms. The result has been a narrative that places vocational knowledge in a lower tier: useful but narrow, technical but not intellectual, instrumental rather than exploratory (Keep et al., 2021; Nakar, 2025). Writers such as Hyland (2014), Wheelahan and Moodie (2017) and Keep et al. (2021) have drawn attention to how these assumptions are woven into the very structure of the sector. In some accounts, learners are treated more like consumers than students, and knowledge becomes something to be packaged and sold rather than developed or questioned (Coffield, 2008; Nakar and Olssen, 2021).

Sociological work helps explain why these patterns are so resilient. Bernstein's (2000) theory of pedagogic discourse, for instance, highlights how education systems often privilege certain types of knowledge and practices over others, which helps reproduce longstanding divisions between “academic” and “vocational” learning. These divisions have real consequences: they shape funding, influence public perceptions and determine who gets access to what kinds of learning. Research by Gregson and colleagues points out that this contributes to inequalities that remain stubborn despite years of reform (Gregson et al., 2020).

In this context, the integration of AI presents a timely opportunity to reshape the narrative. As AI transforms the nature of work and demands new skillsets, there is potential for VET to position itself as a forward-looking sector capable of preparing learners for cutting-edge careers in an AI-augmented workplace. By embedding AI into teaching, learning and curriculum design, VET institutions can demonstrate agility, innovation and relevance – thereby challenging the perception of VET as outdated or low-status.

However, the promise of AI is not matched by an equivalent body of research. Very little work explores how AI might genuinely help reposition VET or how pedagogical innovation could reshape the sector’s identity. Even less is known about how those within VET – students, teachers and leaders – actually experience AI tools or what motivates them to adopt or resist new technologies. These perspectives matter because implementation does not happen in a vacuum: it unfolds in the lived reality of classrooms, workshops, training centres and administrative offices. Without understanding those settings, AI strategies may be technically impressive but educationally thin.

This systematic literature review addresses key gaps in understanding the transformative potential of AI in rebranding vocational education and developing innovative teaching and learning practices. It synthesises VET–AI literature from the past 5 years, focusing on the perspectives of students, teachers and institutional stakeholders, as well as the practical application of AI in vocational contexts. Existing research highlights the importance of strong theoretical frameworks in guiding not just adoption, but sustained and meaningful integration of technology in education. Without such foundations, AI risks being used in fragmented ways rather than supporting systemic change.

To explore this landscape, the review investigates the following questions: What AI technologies have been successfully applied in vocational contexts? To what extent is VET–AI research informed by theory, and how might this be strengthened to support the sector's rebranding? Finally, what evidence exists of AI's potential to rebrand and innovate vocational training?

The integration of AI in VET represents a significant innovation within the broader Industry 4.0 landscape, characterised by data analytics, connectivity, automation and human–machine interaction (Fuertes et al., 2021; Moraes et al., 2023). This study draws upon established theoretical frameworks from information systems research, notably the technology acceptance model (TAM), to understand the adoption by varied stakeholders and the impact of AI technologies in VET settings. TAM offers a structured approach that includes perceived usefulness, perceived ease of use, attitudes towards technology, behavioural intention and actual system use.

Initially developed by Davis (1989), the TAM provides a robust theoretical foundation for examining technology adoption behaviours across diverse contexts. Unlike Ajzen's (2012) theory of reasoned action, which offers a generic approach to predicting attitudinal underpinnings of behaviours, TAM specifically addresses technology acceptance by linking attitudinal and behavioural aspects to technology preference and use. This specificity has established TAM as a cornerstone theory in information systems and organisational behaviour research (Davis, 1993; Marikyan and Papagiannidis, 2024). Through successive iterations, TAM has evolved to address various limitations: TAM2 expanded the model to incorporate social influence processes and cognitive instrumental processes (Venkatesh and Davis, 2000), whilst TAM3 introduced direct predictors of perceived ease of use, including computer self-efficacy, perception of external control, computer anxiety, computer playfulness, perceived enjoyment and objective usability (Venkatesh and Bala, 2008). These extensions have enhanced the model’s explanatory power and applicability across different contexts, including educational settings.

For example, the basic TAM has been used in VET–AI research recently by Seufert (2024), where the author contended that all variations of TAM identify Behavioural Intention to Use as an antecedent to the Actual System Use, “reflecting the strength of a user's intention to perform a specific behaviour, in this case, the use of an AI-based solution” (p. 1). The author also argued that VET leaders' understanding of this relationship is vital for the successful adaptation, development and implementation of a strategy for the acceptance of AI solutions in this sector by both educators and students. Extending the VET–AI research to investigate the VET-specific determinants of perceived usefulness (from TAM2) and the determinants of perceived ease of use (as TAM3 suggests) will offer better insights and more practical strategies and policy guidelines for the sustainable use of AI solutions in VET.

Despite critiques regarding its parsimony and initial focus on individual technology use rather than performance outcomes (Benbasat and Barki, 2007; Goodhue, 2007), TAM’s theoretical resilience and strong predictive capability make it particularly relevant for examining AI adoption in VET. As Marikyan and Papagiannidis (2024, p. 9) asserted, “the limitations cannot overshadow the contributions of the theory,” noting TAM's enduring ability to assess individuals’ technology use intentions for nearly 3 decades.

Employing the TAM to analyse AI implementation within vocational education contexts yields two significant contributions: first, it enhances theoretical frameworks by extending them into an understudied educational domain; second, it generates practical insights that can guide how these emerging technologies might be leveraged to strengthen VET’s relevance and strategic positioning within contemporary labour markets that are increasingly transformed by AI advancements.

In order to conduct a comprehensive review of previously published literature on AI integration within the VET sector, this study followed the PRISMA guidelines (PRISMA, 2023; Rethlefsen et al., 2021) throughout the entire research process, including using a flowchart (Page et al., 2021). The PRISMA framework provides a 16-item checklist that helps ensure transparency and reproducibility in reporting systematic searches (Rethlefsen et al., 2021). The process unfolded in a way that responded to the sparsity and inconsistency of available work rather than adhering rigidly to each step.

The first task was establishing what would count as relevant. Only studies published between 2019 and 2024 were considered, as AI in its current form has moved rapidly, and earlier work often refers to technologies that no longer reflect contemporary practice. To keep the review anchored in genuine VET contexts, the search was limited to research involving vocational institutions, programmes or learners. University-based studies were excluded, even those touching on applied subjects, as their environments and pedagogical structures differ substantially from VET systems. Conceptual papers, policy briefs and literature reviews were also set aside, since the goal was to draw on empirical findings.

Studies needed to examine specific AI applications within VET settings. This meant clear use of technologies such as machine learning, natural language processing, computer vision or intelligent tutoring systems. Research focused solely on basic automation or learning management systems was excluded, as was work that used AI only for data processing or statistical analysis without examining educational applications. To ensure real-world relevance, only research involving actual VET stakeholders (students, teachers or administrators) was included. Papers lacking stakeholder input or focused purely on system features or theoretical models were excluded.

The search was carried out across eight databases during July 2024: ERIC via ProQuest, Education Database, A + Education, Gale Academic OneFile, Scopus, VOCEDplus, Web of Science and JSTOR. Because authors use varied terminology when discussing AI in education, search terms were combined and adapted using groups of related expressions rather than relying on a single phrasing. These included variations of “artificial intelligence’, “AI applications’, “teaching’, “training” and “vocational education”. For example, the search combined terms like (“artificial intelligence” OR “AI”) AND (“AI integration” OR “AI application”) OR (teaching OR instruction OR education OR pedagogy OR training) AND (“Vocational Education”) OR (“occupational training” OR “technical education” OR “applied education”). Searches were run on abstracts rather than full texts to keep the process manageable and to maintain consistency across databases. The initial search produced 479 items, as shown in Table 1.

Given that AI in VET is an emerging field, the search was supplemented with a manual scan of Google Scholar. Following the recommendation of Haddaway et al. (2015) that the first few hundred results often capture work not indexed elsewhere, this step identified four additional studies. All records were exported into an Excel spreadsheet, where duplicates were removed. After removing 66 duplicates, 413 records underwent title and abstract screening. Of these, 389 were excluded for failing to meet basic eligibility requirements. The majority (298 studies) focused on higher education or general post-secondary settings.

An additional 48 papers examined general educational technologies without specific AI integration; 27 were non-empirical works, including theoretical and conceptual papers, and 16 fell outside the timeframe or language criteria. Twenty-four articles progressed to full-text review, conducted independently by two authors who compared interpretations and discussed any uncertainties. The selection process is summarised in Figure 1. This stage confirmed the limited nature of available research: although a reasonable number of papers mentioned VET in passing, only a small subset genuinely examined AI within vocational teaching, learning, or administration. A further 12 studies were excluded for not meeting the full inclusion criteria, leaving a final sample of 12 studies.

The decision to focus exclusively on VET contexts rather than broader post-secondary education was methodologically necessary. Many studies on AI in education concentrate on universities or mixed institutions offering both academic and vocational programmes. Including such research would have compromised the distinct analytical lens required for VET’s unique characteristics: competency-based assessment, strong industry links and occupation-specific pathways. While this boundary limited available literature, it ensured findings were directly relevant to VET rather than diluted by fundamentally different educational models. The small final sample reflects the genuine lack of focused research in this area rather than any limitations in the search strategy. As with other systematic reviews in emerging fields, the limited evidence base still reflects the current state of knowledge.

Following Braun and Clarke's (2022) six-step inductive approach, thematic analysis was conducted on the results of the 12 selected articles using NVivo (Version 12). In the familiarisation phase, the second author carried out line-by-line open coding to support the cross-study translation of findings. Initial codes were then generated and grouped into potential themes. These themes were reviewed for coherence with the original coding, and refined to ensure clarity and alignment with the research questions. The final report was structured around the identified themes. Multiple debriefing sessions, in which the authors critically discussed and refined their interpretations until consensus was reached, were held to minimise subjectivity.

Reviewing the 12 selected articles revealed a rich diversity of approaches despite the modest sample size. The studies examined both the application of AI in VET contexts and the beliefs and attitudes surrounding its use, drawing on the perspectives of students, teachers and institutional stakeholders, including principals and VET leaders. Geographically, the research covered a widespread area including China, Ghana, Germany, Malaysia, Taiwan, Switzerland, Australia and Brazil, although a few studies did not explicitly state their national context. As summarised in Table 2, most studies adopted mixed methods designs, frequently combining interviews and surveys with practical interventions to assess AI’s effectiveness in educational settings. These were followed by quantitative and experimental approaches, including applications of machine learning. Only one study employed a purely qualitative methodology, highlighting a potential area for future research development.

A prominent focus across the selected papers was the exploration of beliefs and attitudes towards AI, reflecting a dual perception of AI as both an opportunity and a challenge within vocational education. Students, teachers and stakeholders consistently expressed openness and willingness to engage with emerging technologies, recognising their potential to enhance learning and teaching practices (Bekiaridis and Attwell, 2024; Liu et al., 2020; Ma, 2022; Nyaaba and Zhai, 2024; Ridzuan and Junaidi, 2023; Rott et al., 2022; Seufert, 2024). One study also considered the influence of AI integration on pre-service teachers' career choices (Moreno and Petko, 2024), though only a small proportion of its participants (11%) were from postgraduate VET programmes.

Despite the relatively small number of studies that directly evaluated innovative AI integration in VET, important developments were observed. A handful of studies implemented and assessed AI-driven tools for teaching and learning, including robot-assisted instruction (Kong et al., 2024; Souza et al., 2022; Rott et al., 2022). Other studies demonstrated AI’s value for VET administration, notably through predictive analytics to forecast programme completion rates and create personalised course recommendations based on student preferences (Hall et al., 2023; Ma, 2022). These applications point to AI’s potential to strengthen both institutional operations and learner support services. Alongside these empirical contributions, Rott et al. (2022) argued for the creation of an AI-specific qualification to address emerging workforce needs, contending that such a development could enhance graduate employability. However, no such programme has yet been trialled, leaving this as a policy aspiration without an accompanying evidence base.

The theoretical foundations across the studies varied but were dominated by versions of the TAM (e.g. Seufert, 2024). Other theoretical lenses included diffusion of innovation theory (Nyaaba and Zhai, 2024), the smart learning model (Kong et al., 2024), agency theory (Moreno and Petko, 2024) and the robot teaching assistant-supported learning model, which was embedded within an extended TAM framework (Chang and Hwang, 2024). Across the 12 studies, 110 significant statements were extracted and synthesised into two overarching themes: beliefs and attitudes towards AI, and innovative integration in practice (see Figure 2).

Within these themes, our TAM-informed analysis also reveals distinct patterns of barriers and opportunities. Several barriers emerged repeatedly (Figure 3). Implementation challenges appeared frequently, especially the puzzling gap between enthusiastic attitudes and limited actual use. Resource constraints create ongoing problems as funding is limited, time pressures persist and professional development remains inadequate. Ethical issues also surfaced consistently, with particular concern about academic integrity, data privacy and whether AI systems might embed or amplify bias. These are counterbalanced by opportunities, including personalised learning pathways, predictive analytics for student success, competency-based assessment alignment, industry-relevant skill development and innovative pedagogical approaches through robot-assisted and wisdom teaching models (see Figure 3).

Beliefs and attitudes towards using AI emerged as an important finding for students, teachers and stakeholders, as explained next.

Overall, students viewed AI positively in terms of its educational potential. Ridzuan and Junaidi (2023) reported a mean score of 3.88 for perceived usefulness, indicating general agreement that AI supports learning. However, engagement scored lower at 3.40, highlighting a gap between recognition of AI's benefits and actual use. This may be due to limited integration support or challenges in incorporating AI into students' routines.

While the TAM (Davis, 1993) suggests perceived usefulness and ease of use contribute to acceptance, it does not guarantee adoption. In this review, students reported AI as relatively easy to use (mean score = 4.18), yet variability in comfort levels (SD = 0.803) suggests inconsistent confidence, which may hinder widespread adoption. Addressing this variability is key to promoting meaningful engagement and realising AI's full educational value. Taken together, these findings point to a perception–implementation gap in which students recognise AI as useful and report it as accessible, yet these positive evaluations do not reliably translate into sustained engagement. The gap between a perceived usefulness score of 3.88 and an engagement score of 3.40 (Ridzuan and Junaidi, 2023) is numerically modest but practically significant. It indicates that favourable attitudes towards AI are a necessary but insufficient condition for adoption within VET learning contexts. Factors extrinsic to the individual student, including how AI is integrated into assessment tasks, whether its use is modelled by teachers and whether institutional infrastructure supports access, are likely to account for much of the remaining variance. This finding qualifies straightforward applications of TAM in VET settings, since attitude formation and behavioural adoption are related but not equivalent, and the conditions mediating the relationship between them deserve closer attention in future research.

Another theme was students' awareness of AI's relevance to vocational education, though its influence on career choice was limited. Liu et al. (2020) noted students recognised AI's role in digital transformation, but found it had minimal impact on their career plans. Similarly, Moreno and Petko (2024) found that factors such as job satisfaction and subject interest were more influential than AI in shaping career decisions. These findings suggest that while AI is part of students’ educational awareness, it is not yet a significant driver of vocational aspirations.

Findings show that teachers and stakeholders express both enthusiasm and caution regarding AI in VET. While they recognise its potential to support educational practice and benefit students, concerns remain around preparedness and ethical implications. Seufert (2024) found strong support among Swiss vocational educators, with 96 of 111 respondents endorsing AI’s benefits and 95 supporting its integration. However, 78 raised concerns about societal readiness, reflecting a sense of cautious optimism. This suggests that while educators acknowledge AI's value, they remain mindful of the challenges posed by its rapid deployment.

A consistent sub-theme was the need for professional development. Nyaaba and Zhai (2024) identified a strong demand for training in AI-supported activities such as automated marking, lesson planning and research. While teachers appeared open to using generative AI tools, they emphasised the need for structured support and ongoing development. This aligns with TAM (Davis, 1993), which posits that perceived usefulness and ease of use contribute to the intention to engage. Although VET teachers may view AI as beneficial, a lack of preparation appears to delay active engagement.

This inaction may stem from scepticism, limited training or lack of confidence. Several studies (e.g. Bekiaridis and Attwell, 2024) highlight that teachers feel underprepared for the technical demands of AI. This points to a critical need for professional development programmes that build technical, data literacy and computational skills, as well as address ethical concerns. Both Bekiaridis and Attwell (2024) and Liu et al. (2020) advocate for training that equips educators with the knowledge required for confident and ethical AI integration in vocational teaching practice.

A theme of healthy scepticism and ethical concern emerged around AI use in education. Teachers were particularly apprehensive about academic integrity, data privacy and algorithmic bias. Bekiaridis and Attwell (2024) noted fears of plagiarism and misuse of personal data, while Nyaaba and Zhai (2024) highlighted student misuse of AI tools, calling for structured guidelines to mitigate risks.

Data privacy was a major concern, with teachers wary of exposing sensitive student data through AI-driven systems (Bekiaridis and Attwell, 2024). Additionally, there were fears that AI adoption could deepen socioeconomic divides by disadvantaging students with limited access to technology (Nyaaba and Zhai, 2024). These concerns, aligned with TAM3's emphasis on perceived risk, highlight the need for clear, equitable policies to support ethical, inclusive AI integration in VET settings.

An emerging theme among teachers relates to concerns about their own proficiency with AI. As students quickly develop advanced skills with these tools, some educators fear a shift in the traditional teacher–student dynamic. Nyaaba and Zhai (2024) proposed unrestricted AI access for teachers, with controlled access for students, arguing that such a model could help maintain professional authority in the classroom while promoting balanced, pedagogically informed integration.

Reluctance among some TVET teachers to adopt AI may stem from the pace of technological change and the complexity of implementation. Professional development is essential to help teachers understand AI's potential and apply it effectively in their practice. While earlier research (Holmes et al., 2019; Luckin et al., 2016) highlighted the benefits of digital tools for personalisation and efficiency, the emergence of AI introduces new challenges. These require targeted training to support meaningful integration into pedagogical approaches. The teacher data reinforces the perception–implementation gap from a different angle. Where student-based evidence shows recognition without consistent uptake, the teacher-based evidence shows interest qualified by institutional uncertainty. Seufert (2024) found that Swiss vocational educators could see AI’s potential while doubting whether their organisations were ready to use it responsibly, and Bekiaridis and Attwell (2024) found that concerns about data handling and academic integrity operated as active deterrents rather than background worries. Across both student and teacher data, the pattern is consistent in that positive orientation towards AI does not automatically produce changed practice, and the conditions that would allow it to do so – adequate professional development, clear governance and reliable institutional infrastructure – are reported as absent more often than present. Interpreting the two themes together, it is the interaction between beliefs and institutional context that shapes whether AI moves from awareness into application.

The second major theme to emerge from the systematic review was innovative approaches to the integration of AI in practice within VET. The findings identify four key areas of specific work, including AI-based qualification frameworks, predictive achievement models, robot-assisted learning and the wisdom teaching model. Together, these areas of work relating to AI in VET could illustrate AI’s potential transformative impact on vocational education, equipping students with industry-relevant skills and competencies.

A key sub-theme identified was the growing demand for AI qualifications, highlighting the need to equip students for an AI-driven economy. Rott et al. (2022) advocate for an expanded AI curriculum in VET, supported by targeted teacher training. They suggest that either cross-sector AI qualifications or sector-specific courses showcasing AI's transformative potential would strengthen VET's dual focus on practical skills and knowledge transfer.

Another sub-theme focused on AI-based qualification frameworks to personalise learning pathways. Ma (2022) demonstrated the effectiveness of AI-driven recommendation systems, achieving a 96% accuracy rate. Such systems help align course selection with student aspirations and labour market demands, which is critical in vocational education where job relevance is key. Innovative AI applications, such as predictive models like XGBoost and CatBoost (Hall et al., 2023), were also identified. These tools support early intervention by forecasting student outcomes, enabling timely support and reducing the risk of attrition.

Hall et al. (2023) identified advanced predictive models (XGBoost and CatBoost) that effectively forecast student achievement rates, enabling early intervention for at-risk students. These analytical approaches support proactive retention strategies to address the variable completion rates documented in Australian VET programmes (ranging from 50% to 80%).

Souza et al. (2022) found that students participating in robot-assisted learning scored significantly higher than control groups in assessments. This interactive approach promotes hands-on, student-centred learning, which Chang and Hwang (2024) reported as particularly beneficial for practical applications like healthcare case management. Robot-assisted methodologies support the development of critical thinking and problem-solving capabilities while fostering student autonomy.

Kong et al. (2024) reported significantly higher engagement and creativity among students taught using a wisdom teaching model that emphasises creative problem solving and active participation. This approach develops the analytical and creative capabilities identified as essential workforce skills by the World Economic Forum (2023), highlighting AI’s potential to enhance precisely those human capabilities that complement rather than compete with automated systems. Collectively, these innovative approaches demonstrate AI’s capacity to transform vocational education across instructional, administrative and assessment domains while developing the workforce capabilities needed for contemporary technological environments.

Our analysis points to several gaps that need addressing through empirical work. Longitudinal studies would help considerably. Research that follows VET institutions over time could clarify how attitudes towards AI evolve. Do initial perceptions of usefulness and ease of use remain stable, or do they shift as people gain hands-on experience with these technologies? Longitudinal designs would also allow researchers to ask a question this review could not answer, namely, what happens after the pilot ends? The available evidence largely captures attitudes during funded trials or structured professional development programmes when institutional support is at its highest; whether those attitudes persist once ordinary workload pressures return is simply not known. Tracking the same cohorts of teachers over two or more years could clarify which conditions, professional development quality, leadership prioritisation or policy alignment, matter most to sustained adoption. Comparative work across different VET settings would also prove valuable. Technical colleges face different constraints and challenges than apprenticeship programmes or industry training centres. Understanding what implementation factors matter in each setting would strengthen the evidence base considerably.

Experimental studies testing specific interventions remain surprisingly scarce. Research using control groups to evaluate approaches like robot-assisted learning or predictive analytics systems would help establish whether particular strategies actually cause the improvements claimed for them. The perception implementation gap identified in this review calls for mixed-methods research that can capture both measurable attitudes through TAM metrics and the messier realities of barriers and enablers that qualitative work reveals. Combining quantitative and qualitative approaches would provide richer insight than either method alone. Action research warrants more attention in this field than the current literature affords it. VET already operates through close industry partnership and work-integrated learning, giving the sector an established culture of knowledge generated through practice. That orientation makes VET well-suited to participatory designs in which teachers and administrators help shape research questions and trial interventions within their own programmes, addressing the shortage of practitioner-centred inquiry that both Holmes et al. (2019) and Luckin et al. (2016) identified in educational technology research.

This systematic review illuminates the transformative potential of AI integration as well as implementation challenges in vocational education. The emergence of two principal themes – beliefs and attitudes towards AI, and innovative integration practices – provides a valuable framework for understanding how AI can reshape VET’s pedagogical approaches and market positioning in an increasingly technology-driven educational landscape. The economic imperative for this technological transformation, which is central to the projection from McKinsey’s that AI adoption could contribute an additional $13 trillion to global economic activity by 2030 (Bughin et al., 2018), highlights the critical need for VET institutions to align their educational offer with this technological revolution.

Our analysis identified a critical disconnect between positive AI perceptions and limited practical engagement, representing what we term the perception–implementation gap. Research by Ridzuan and Junaidi (2023) revealed that whilst participants expressed appreciation for AI's potential benefits, their day-to-day utilisation remained significantly lower than expected, given these positive attitudes. This discordance echoes what Wu (2021) described as the “implementation gap” prevalent across educational technology adoption, a phenomenon that becomes particularly evident with sophisticated technologies like AI.

The enthusiasm–utilisation disparity suggests a more complex adoption pathway than previously understood. Positive sentiment towards AI appears insufficient to catalyse meaningful integration: vocational education institutions require implementation frameworks that systematically address practical obstacles to engagement. Simultaneously, these frameworks must build on the existing goodwill towards AI technologies as a foundation upon which more sustained adoption can be constructed. The significant variability in students' comfort with AI applications further highlights the need for differentiated approaches to technological integration. As Davis (1989) established, perceived ease of use constitutes a critical determinant of technology acceptance. This variability indicates that successful AI implementation in VET requires scaffolded learning experiences that accommodate diverse technological proficiencies whilst gradually building competence and confidence. Part of the explanation lies in conditions that individual attitudes alone cannot change. A teacher who recognises AI’s potential still needs protected time in the teaching week to learn a new tool, a workload structure that does not penalise experimentation, and clear institutional guidance on what responsible use means within a competency-based programme. The evidence across the reviewed studies suggests that where these conditions were present, even in modest form, uptake was more consistent, whereas where they were absent, expressed enthusiasm rarely translated into changed practice. This points to a reallocation of effort, in that interventions directed at attitude change are likely to produce limited returns if the organisational conditions that allow attitudes to become actions remain unaddressed.

This gap raises questions about how well the TAM fits vocational settings. While TAM has been useful for explaining attitudes in many educational contexts, the VET environment is unusual: highly structured, mapped to industry expectations and governed by competency frameworks that sometimes leave little room for experimentation. Our findings extend traditional TAM understanding by suggesting that “perceived usefulness” and “ease of use” may not always drive sustained technology adoption in the VET context. Relevance to industry, the fit with assessment practices, competency alignment or even the emotional dimension of teacher confidence may matter just as much, if not more. These factors may also mediate the relationship between attitude and behaviour, roles that are not yet captured by the traditional TAM constructs. The urgency of addressing this gap is amplified by the World Economic Forum's (2020) prediction that 50% of all employees will need reskilling by 2025 due to the adoption of technology, positioning VET institutions as critical facilitators of this workforce transformation.

Moreover, TAM has been viewed as limited in its ability to explain individual technology adoption decisions, which may not necessarily translate into institutional adoption or a sector-wide uptake of AI solutions that the model might predict. Whether an institution will integrate AI solutions into its teaching, assessment and governance practices may depend on a complex set of factors that are not necessarily the direct determinants of an individual's perceived usefulness or ease of use, which may be beyond the scope of TAM. Hence, future theoretical work would benefit from treating VET’s structural features as central rather than contextual, moving beyond TAM's individual perception to the influence of actual “facilitating conditions” and distinguishing between surface-level tool adoption and the deeper pedagogical shifts that genuine integration of AI into vocational learning would require.

Our analysis indicates that despite growing awareness of AI’s workforce implications, students’ career decisions remain primarily influenced by traditional factors such as subject interest and job satisfaction (Moreno and Petko, 2024). This finding contradicts what might be expected given Mandala Partners’ (2023) projection that AI will transform workplace requirements for 7.2 million Australian workers. The disconnect between the awareness of AI’s transformative capabilities and its minimal impact on career trajectory planning constitutes both a significant challenge and a strategic opportunity for VET institutions attempting to reposition themselves as conduits to AI-augmented professional pathways.

This observed incongruity becomes particularly concerning when considered alongside evidence from Verma and Srivastava (2021), who demonstrated that AI integration in business communication is creating new skill requirements across industries, fundamentally altering workplace competencies and career trajectories. By demonstrating how vocational qualifications equip learners with competencies requisite for technologically reconfigured workplace environments, such institutions can enhance their competitive market positioning whilst facilitating greater alignment between student career expectations and the evolving requirements of contemporary industry sectors.

Teacher readiness emerged as a critical factor in successful AI integration. The strong demand for ongoing professional development (Nyaaba and Zhai, 2024) reflects an emerging theoretical consensus that technological integration requires more than hardware and software investments – it necessitates sustained capability development among educational practitioners. This finding aligns with what Holmes et al. (2019) and Luckin et al. (2016) demonstrated in earlier research on digital tool implementation: educator competence constitutes the foundation for meaningful technological integration.

Seufert’s (2024) findings highlight that vocational educators are maintaining a delicate balance, acknowledging AI’s promising benefits whilst expressing substantial concerns about system-wide preparedness with educators, further suggesting that professional development initiatives should address technical skills and contextual understandings. The complexity of this challenge grows with the rapid pace of AI advancement across multiple business sectors, as documented by Verma and Srivastava (2021), who highlight how AI applications in business communication continue to evolve, requiring continuous upskilling of educational practitioners to remain current with industry developments. Comprehensive training programmes would ideally encompass technical competencies, data literacy, computational thinking and ethical frameworks to prepare educators for multifaceted AI implementation.

The emotions matter as well. If teachers feel unsettled or exposed, they are less likely to experiment or to trust new tools. Several studies suggested that the training available to teachers was piecemeal; for example, it consisted of short workshops, optional sessions or one-off introductions. What is needed is something longer-term, tied into everyday teaching rather than delivered on the side.

Teachers consistently expressed significant ethical concerns centred on academic integrity, data security and algorithmic fairness (Bekiaridis and Attwell, 2024; Nyaaba and Zhai, 2024). These concerns underscore the need for robust governance structures to guide the responsible integration of AI into learning environments. Recurring across multiple studies, these issues point to broader questions of responsible technology use, requiring institutional policies and sector-wide standards to ensure AI supports transformation without compromising ethical principles. These concerns gain further weight when viewed through Schwab's (2016) lens of Industry 4.0, where rapid technological change introduces complex ethical challenges across all areas of life.

The World Economic Forum's (2020) analysis of emerging job categories and skill requirements further emphasises the need for ethical frameworks that balance innovation with protection, particularly as AI reshapes traditional employment patterns and creates new forms of work that require different competencies and safeguards. Striking the right balance by enabling meaningful AI access while preventing misuse is critical. For VET institutions, this means establishing clear protocols for responsible use without placing undue restrictions that stifle innovation or hinder skill development. This is particularly important in vocational contexts, where practical competence with evolving technologies is a core educational goal.

The four key areas of innovative AI implementation identified in our analysis – AI qualifications and frameworks, predictive achievement models, robot-assisted learning and the wisdom teaching model – demonstrate AI’s potential to transform vocational education while reinforcing its distinctive educational identity. These applications align remarkably well with the Future Skills Organisation (2024a) framework, Building an AI-Empowered Workforce, which calls for significant changes in the VET sector to address emerging workforce requirements. The strategic importance of these innovations is underscored by the McKinsey Global Institute’s analysis (Bughin et al., 2018), which identified education and training as one of the sectors most likely to benefit from AI’s productivity enhancements, suggesting that VET institutions that successfully integrate these technologies will gain significant competitive advantages.

What the literature did not reveal was a clear sense of how institutions plan to sustain or scale these experiments. Very few studies looked at implementation over time. There was almost no exploration of leadership perspectives, even though leaders typically decide which tools are adopted, which are funded, and which are quietly shelved. Similarly, no study compared adoption across different shapes of VET provision – public, private, apprenticeship-based or industry-led – despite the likelihood that these settings experience AI very differently.

Taken together, the findings suggest a sector that is cautiously interested in AI but is still early in its journey. The enthusiasm is genuine, but so are the hesitations. The technology appears to hold real potential, yet meaningful integration will rely on clearer governance, stronger professional development and a closer alignment between AI tools and the underlying logic of vocational learning.

The studies also point to something more hopeful. AI might help VET reposition itself not as a secondary pathway but as a sector connected to innovation and to the future of work. For this to happen, though, the adoption of AI needs to be shaped by the realities of vocational teaching, and not by general narratives drawn from higher education. The field is ready for deeper, more sustained research that involves teachers and leaders directly and that follows the process of adoption over months and years rather than days and weeks.

Across the studies, AI was used in ways that mostly fitted around what vocational teachers were already doing. In the example from Souza et al. (2022), students simply had another route into the task rather than a new method altogether. A similar pattern came through in Chang and Hwang's (2024) work, where robots were added into sessions that looked very much like usual teaching. In Kong et al.’s (2024) study, even though the focus was on encouraging creativity, the learning still relied quite heavily on practical involvement. So, across these different settings, AI seemed to sit next to normal activity rather than reshaping it, and quite often just made space for additional practice or repetition.

Our analysis confirms the dominance of the TAM and its derivatives in studies of AI adoption in VET (e.g. Seufert, 2024). While Davis's (1989) framework remains influential, its widespread use reveals the need for more context-sensitive models that reflect the unique characteristics of vocational education. The connection between our two core themes – beliefs and attitudes towards AI, and innovative integration and TAM constructs – affirms its ongoing relevance. Yet, the persistent gap between perceived usefulness and actual implementation highlights its limitations. As Marikyan and Papagiannidis (2024) note, TAM's value lies in capturing intention rather than guiding application.

This shortcoming becomes more noticeable when considering the projected AI-leveraged economic shifts described by authors such as Bughin et al. (2018). It has been suggested that traditional models may not capture the complexity in leveraging AI across diverse educational contexts. The application of AI in vocational settings depends on factors that TAM does not always capture (e.g. competency-based approaches, workshop structures and employer expectations). These everyday features matter, and their absence from TAM could explain why positive attitudes may not always translate into consistent use, especially in VET.

Future theoretical development should extend TAM to account for VET's distinctive features: its practice-based pedagogy, industry alignment and competency-driven goals. Developing more robust, sector-specific conceptual models would not only support effective AI implementation but could also contribute to rebalancing the perceived status of vocational education. As Donovan (2019) argued, trust is key to legitimising VET and trust grows when innovation is grounded in credible, contextually relevant frameworks.

Across the studies, a number of practical issues kept resurfacing with implications for a clear regulatory framework. For example, ethical concerns were mentioned frequently, including questions about privacy, fairness and how work produced with AI should be treated. These concerns appeared in Bekiaridis and Attwell (2024) and again in Nyaaba and Zhai (2024). In Seufert’s (2024) findings, teachers were interested in AI but not always confident that organisational systems were ready for it. This reflects broader arguments from the World Economic Forum (2020), which emphasise the need for planned reskilling initiatives that will be crucial for managing the transition to an AI-augmented workforce, positioning VET institutions as key partners in national economic competitiveness strategies.

Several papers also noted resource challenges. Smaller VET providers may struggle to introduce new systems without targeted support. Rott et al. (2022) highlighted the importance of keeping strong links with industry when developing AI-related qualifications, particularly to ensure alignment with workplace needs. When set alongside the economic projections offered by Bughin et al. (2018), investment in the digital capability of VET becomes more of a structural requirement than an optional enhancement.

Some studies pointed to the value of starting with narrower, more contained applications of AI. Hall et al. (2023), for example, used predictive analytics to identify students needing support early on, while Ma (2022) trialled AI-based personalised learning pathways to help learners choose appropriate courses. These examples suggest that small-scale uses may help reduce the gap between attitudes and behaviour (perception-implementation gap) described by Ridzuan and Junaidi (2023).

Our findings prominently featured teacher development as well. Educators wanted guidance on technical matters but also clarity around ethical use and appropriate integration into teaching. This was shown in studies by Nyaaba and Zhai (2024) and by Bekiaridis and Attwell (2024). Verma and Srivastava (2021) added that AI-driven communication tools within the industry change quickly, reinforcing the need for ongoing training rather than one-off sessions.

For educators, several studies described classroom use of AI to assist with problem-solving or creative work. Examples include the wisdom teaching model discussed by Kong et al. (2024) or robot-assisted activities examined by Chang and Hwang (2024) and Souza et al. (2022). Despite the potential, concerns were raised about data practices, and, in some cases, students felt more confident with the tools than staff. Both Bekiaridis and Attwell (2024) and Nyaaba and Zhai (2024) stressed the need for clear expectations around responsible use. Given the concerns about shifting power dynamics identified by Nyaaba and Zhai, educators should strive for a balanced approach to AI integration that maintains pedagogical authority while leveraging AI's educational benefits.

Several practical directions follow from this. Data governance is an area where institutions can act without waiting for sector-wide policy. Bekiaridis and Attwell (2024) found that uncertainty about how student data are stored and used was among the strongest deterrents to AI adoption, and institutions that address this through clear, visible policies rather than buried documentation are likely to encounter less resistance from both staff and students. The other issue that institutions can address relatively quickly is the structural mismatch between how AI professional development is currently delivered and what teachers have said they actually need. The existing literature points to a shortage not of willingness but of time and relevant guidance, and reshaping how professional development is resourced and scheduled within VET workloads is, in principle, an administrative decision rather than a funding one. At the policy level, Rott et al. (2022) made the case for sustained investment in AI-related qualifications that keep pace with industry change. The same argument extends to the professional and governance infrastructure within providers. Sector-level guidance on data privacy, academic integrity in AI-assisted assessment, and minimum standards for algorithmic transparency would reduce the burden currently falling on individual institutions to resolve these questions without reference to any shared standard.

This systematic review uses 12 carefully selected studies to explore the transformative potential of AI in the VET sector. Twelve studies are not much, but that scarcity tells its own story about where research into AI and vocational education currently sits. What this review found was less a transformation underway and more a sector grappling with possibilities it has not yet worked out how to act on. Our findings show that people generally seem interested in learning and using AI. Students across the studies recognised that AI might be useful and did not find it particularly difficult to use. However, interest did not translate into regular engagement. Ridzuan and Junaidi (2023) called this an implementation gap, and Wu (2021) noticed something similar in broader technology adoption contexts.

Teachers felt the same tension. Seufert (2024) found that Swiss vocational educators could see potential in AI but doubted whether their institutions were equipped to make proper use of it. Professional development kept surfacing as something people wanted but were not getting enough of, particularly around automated assessment, lesson planning and research work (Nyaaba and Zhai, 2024). Ethics worried people too, with questions about academic integrity, privacy and whether algorithmic bias might creep into systems (Bekiaridis and Attwell, 2024; Nyaaba and Zhaı, 2024). These are not side concerns: they sit at the heart of teachers’ trust in the systems being introduced. Clearer frameworks, transparent processes and sector-specific guidance will be essential if AI is to support rather than unsettle vocational learning environments.

Innovative applications across the reviewed literature demonstrate the practical potential of AI in VET. These included expanded AI curricula (Rott et al., 2022), personalised learning through recommendation systems (Ma, 2022) and predictive analytics to support struggling students (Hall et al., 2023). Robot-assisted instruction improved cognitive and problem-solving skills (Souza et al., 2022; Chang and Hwang, 2024), while the wisdom teaching model fostered creative engagement (Kong et al., 2024). These developments align with urgent projected workforce trends. In Australia, 7.2 million workers – half the workforce – may need to adapt to generative AI (Mandala Partners, 2023). Unlike past automation waves, this wave affects highly skilled professions (Future Skills Organisation, 2024a), demanding a strategic response from VET providers. Programmes must be reimagined to remain relevant in a shifting landscape.

Our study contributes to educational technology theory by arguing for a recontextualised TAM framework, tailored to the unique dynamics of VET. AI adoption in this sector requires stakeholder-specific strategies. For administrators, we propose a phased implementation beginning with high-utility, low-complexity applications like predictive analytics for student retention. For educators, AI models aligned with vocational pedagogy, such as wisdom teaching or robot-assisted learning, can enhance professional practice. For policymakers, we advocate frameworks that foster innovation while addressing ethical concerns, particularly data privacy, which remains a key issue for VET staff. AI integration offers VET providers an opportunity to reshape outdated perceptions of vocational education by showcasing innovation, industry alignment and student-focused outcomes. As Donovan (2019) notes, trust is essential in legitimising VET. Our findings suggest that well-implemented AI can support this trust by enhancing educational quality and workforce relevance.

For now, however, the most significant gap is in the research itself. We lack longitudinal studies, an action research approach, comparisons across different VET systems, and detailed accounts of how leaders actually make decisions about technology adaptation. TAM has effectively evolved through longitudinal studies (e.g. Venkatesh and Davis, 2000), and action research has been very useful in higher education and technology adaptation research (e.g. Groves and Zemel, 2000). Both approaches can be innovatively combined (see Dahanayake et al., 2025) in future VET–AI studies for greater impact. Most importantly, we lack the perspectives of teachers and students living with AI over time, rather than encountering it briefly in pilot projects. Future research that takes these perspectives seriously, following institutions as they navigate implementation, uncertainty and adaptation, would offer insights far richer than the scattered snapshots currently available.

There is also a broader possibility to consider. If VET can integrate AI in ways that feel authentic to its mission, supporting practical skills, strengthening ties to industry and offering learners clearer routes into future-focused work, the sector may be able to shift the longstanding narratives that position it as secondary to academic education. AI is not a guarantee of this shift, but it may offer a catalyst for rethinking how VET is understood and valued. AI holds promise for vocational education, but that promise comes with conditions. It will require careful design, open conversation, and a willingness to learn from small-scale experiments rather than rushing towards predetermined solutions.

The 12 studies reviewed here suggest an approach that takes the perspectives of students, teachers and institutional leaders seriously, addressing ethical concerns directly rather than treating them as afterthoughts. Early evidence of innovative applications, such as predictive analytics, robot-assisted learning and the wisdom teaching model, looks promising. Leveraging AI can help VET institutions strengthen their position and better serve diverse student populations in an increasingly technology-driven world. However, further evidence is needed from future research that focuses on teachers and leaders, understanding how implementation actually works in practice rather than how it might work in theory. Bringing those voices into decisions about AI integration presents an opportunity to shift how vocational education is perceived. VET could align itself with technological innovation whilst reinforcing what makes it distinctive (i.e. its emphasis on practical, industry-relevant skill development). For now, though, VET stands at the beginning of this AI integration journey, aware of changes ahead but still working out how to take confident steps forward.