Many respondents across all groups had some experience with AI, but adoption rates varied by role. Approximately three-quarters of teachers reported that they had adopted or used AI tools in their professional activities, the highest rate among the groups. Nearly as high, about 70% of administrators indicated they use AI in their work. In contrast, student adoption was somewhat lower—an estimated 60% of students acknowledged using AI tools in their studies. Notably, a significant portion of students lacked clear awareness: roughly 25% of student respondents answered “not sure” when asked if they had adopted AI, a higher uncertainty rate than that of teachers or administrators (about 20% in those groups). This suggests that a subset of students may be using AI-driven features (e.g., built-in functions in apps or websites) without realizing that they count as “AI,” pointing to a gap in AI literacy among learners. Only a small minority of respondents outright said they have never used AI (e.g., about 6% of teachers explicitly reported not adopting any AI, and around 10% of administrators, often those in roles not yet touched by AI). In sum, teachers led in AI adoption (roughly three out of four using AI), with administrators only slightly behind, while students lagged somewhat, with many unsure about AI usage.

Table 1 Respondent demographics, AI adoption, and AI training exposure (survey, n = 1,085). Values are percentages within each role and are rounded to the nearest whole number as reported by respondents.

Across all groups, AI usage was dominated by Chinese-developed platforms, reflecting the national tech environment and restrictions on foreign AI services. The most used AI systems were Baidu’s ERNIE Bot and Alibaba’s Qwen (Tongyi Qianwen). Over one-third of each group selected ERNIE Bot as one of their primary AI tools, and about a quarter reported using Qwen. Another widely used tool was iFlytek’s SparkDesk (an AI platform popular for language and education support), used by roughly 18% of respondents as a main AI application. These domestic AI models were favored due to their availability in China, Chinese language proficiency, and institutional endorsements. In contrast, the usage of foreign AI models was very limited. Fewer than 5% of respondents reported regularly using ChatGPT or similar international AI systems—for example, only a handful of teachers and students (on the order of 2%–4% of each group) indicated they had access to or frequently used ChatGPT. Interview comments confirmed that most colleges rely on approved domestic AI platforms; access to tools like ChatGPT is often blocked or discouraged due to regulatory constraints, and language barriers also make Chinese platforms more practical for daily use. A small number of tech-savvy individuals experimented with VPNs or unofficial access to foreign models, but this was the exception rather than the norm. Aside from large language models, many respondents also used AI-powered applications specific to their tasks. For instance, machine translation software (such as Youdao or Baidu Translate) was extremely common among students for language learning and completing English assignments, and automated proofreading/grammar checkers (often integrated into word processors) were used by both students and teachers. In summary, the AI adoption landscape in these colleges is characterized by heavy reliance on domestic AI tools (with ERNIE Bot leading in popularity) and a near-absence of foreign AI usage due to accessibility issues and policy blocks.

AI usage tended to be occasional rather than constant for most respondents. The survey asked how often individuals engage with AI tools, and the distributions revealed that only a minority are daily heavy users. For instance, only about 20% of students said they use AI daily (often for routine tasks like translation or homework help each day), whereas approximately another 20% of students reported using AI only “monthly or less.” The largest segment of students (around 60%) fell in between—using AI a few times a week or “occasionally” as needed. Teachers showed a somewhat higher regular usage than students: many teachers incorporate AI weekly for lesson planning or grading assistance, with a subset (~25%–30%) using AI tools daily (for example, using an AI tool every day to generate quizzes or to interact on an educational platform). Still, a considerable number of teachers (roughly one-third) said their AI use was infrequent (monthly or rare), which may include educators teaching subjects where AI tools are less applicable or who are still experimenting cautiously. Administrators had similar patterns to teachers: most administrators used AI weekly in their workflow (often for tasks like data analysis during weekly reports, scheduling, or handling student services through AI systems), and about 20% were daily users (likely those in IT or data management roles). Another subset of administrators (roughly 20%–25%) engaged with AI only rarely or not at all, possibly those whose job functions had not yet been augmented by AI. The overall impression is that regular but not pervasive use is the norm—AI is commonly used perhaps a few times per week as part of specific tasks, rather than continuously throughout the day. This indicates that while AI tools have been integrated into routines, their use remains targeted to activities rather than across all aspects of work or study.

Respondents reported using AI for a range of purposes aligned with their roles. Among students, the top uses of AI were for language learning and coursework assistance. Nearly all students (an overwhelming ~90%) indicated they use AI-based tools for language-related tasks—for example, using translation apps, AI chatbots for practicing English conversation, or intelligent tutoring systems for studying language courses. Many students also use AI informally to aid their studies: common examples (from open-ended responses) included using AI to summarize articles or generate ideas for assignments, employing math-solving apps or code auto-completion tools for technical subjects, or consulting AI assistants to get quick answers or explanations. However, students used AI less in collaborative or project-based contexts—only a small fraction had experience with AI in group projects or creative work, indicating that AI’s role in fostering collaboration or higher-order projects was still nascent.

For teachers, the dominant uses of AI are centered on instructional support and assessment tasks. Teachers commonly leveraged AI to generate teaching materials (over half the teachers mentioned using AI to find new examples, create practice questions or worksheets, or produce lecture summaries). Many teachers (around 60%–70%) also reported using AI for grading and assessment support—for instance, using automated grading tools or having AI help draft quiz questions and evaluate student responses. Indeed, teachers gave moderately high ratings (averaging around 3.0 on a 4-point scale) to statements like “AI helps in creating and grading assessments,” reflecting that they find AI valuable in reducing routine workload [this aligns with global reports of educators using AI to streamline repetitive tasks and focus more on teaching (Tomar and Verma, 2021)]. AI was also used by some teachers for personalizing learning, such as identifying resources for different student needs, though this was less developed: on average, teachers rated AI’s role in tailoring instruction to individual students only between “small” to “moderate” extent. This indicates that adaptive learning technologies, while available, were not yet widely or effectively used to differentiate instruction in these colleges—a finding echoing that personalized AI-driven teaching remains an area of untapped potential. Additionally, only a minority of teachers had tried AI tools for promoting student collaboration (e.g., AI-supported group work platforms), suggesting that AI’s use to enhance interactive or collaborative learning was still limited.

Administrators primarily utilized AI for institutional and student support functions. Notably, an overwhelming 90% of administrator respondents said they use AI in student services—this was the highest among any reported purpose. This implies that AI tools like campus chatbots, automated enrollment systems, or AI-driven library and counseling services have become nearly universal in how administrators interact with students outside the classroom. Examples provided by administrators included AI Chabot’s handling common inquiries (e.g., answering FAQs on enrollment, dormitory information, or schedules), AI systems for monitoring student attendance and alerting staff of issues, and algorithm-driven systems matching students to resources (like recommending scholarships or internships). A significant proportion of administrators also use AI for data management and analysis: about 36% reported using AI for institutional data analysis (such as analyzing enrollment trends, processing survey feedback, or generating reports). While that percentage indicates that advanced analytics tools are not yet used by all administrators, it reflects a growing adoption of AI in decision-support roles. Fewer administrators (around 39%) used AI for content creation, since content generation is less central to their duties than it is for teachers or students; however, some did use AI to draft documents, emails, or marketing materials for their college. In summary, for administrators, AI has become integral in delivering and managing student services, and it is also emerging as a tool for improving administrative decision-making and efficiency (e.g., automating scheduling, predicting student needs).

One of the stark findings across all groups was the lack of formal training in AI. Only a small minority of participants had ever received structured training or professional development on using AI tools. For instance, merely 9% of teachers said they had attended any formal training program on AI integration. A virtually identical proportion of administrators (~9%) had formal training. Students reported slightly lower formal training exposure; in the student group, fewer than 10% had participated in any workshop or course specifically on AI usage (aside from their regular curriculum). While some individuals engaged in self-learning (e.g., watching online tutorials or learning from colleagues), these were also in the minority—only a slightly larger fraction of respondents (perhaps 15%–20% in each group) indicated they learned to use AI on their own or from peers. The overwhelming majority, particularly over half of the students, reported no training at all in how to use AI tools. This lack of capacity-building was reflected in qualitative comments: many teachers admitted they were self-taught in using AI, often learning through trial and error or informal peer guidance. Administrators, too, noted that any expertise they had with AI was acquired on the job without formal instruction. The absence of training contributed to uneven or suboptimal usage; for example, some teachers might only use very basic features of an AI tool or avoid more advanced applications because they are not comfortable or aware of them (as one teacher put it, without training, “we use only a fraction of an AI tool’s features”). Moreover, respondents felt that insufficient training and support were one reason AI had not yet deeply transformed pedagogical practices—without knowing how to integrate AI fully into lesson design or research, users stuck to surface-level uses (like using AI for grammar checks or slide generation but not for interactive pedagogy or data analysis). These findings indicate a critical professional development gap: the colleges had not systematically prepared their staff or students to leverage AI, thereby limiting the depth of AI integration. It also highlights a need for institutional support (e.g., IT support staff, workshops) to build user confidence and skills in using AI effectively.

Overall, AI’s integration into teaching and classroom practices was rated as moderate. Teachers and administrators gave mean ratings typically around 2.8 to 3.0 on the 4-point scale for the extent to which AI is used in routine teaching tasks. This suggests that AI has begun to make inroads in certain teaching activities, particularly those involving automation and efficiency, but it has not revolutionized day-to-day instruction. Specifically, AI is effectively handling some routine instructional tasks: tasks such as generating quizzes, grading objective tests, or providing additional practice materials are commonly aided by AI, and educators acknowledged these contributions. For instance, many teachers agreed that AI significantly aids in assessment creation and grading, freeing them from some repetitive work. The data showed high utilization of AI for assessment—one of the highest-rated teaching sub-areas was AI-assisted exam generation and automated grading, which teachers rated near the top of the moderate range (e.g., average around 3.0). This aligns with global trends that assessment and feedback are among the first teaching areas where AI is adopted (Ocen et al., 2025).

However, the impact of AI on more dynamic or student-centered teaching methods was reported to be limited so far. Personalized instruction, where AI would tailor learning materials to individual student needs, had the lowest ratings among teaching-related items. Across respondent groups, the weighted mean for statements like “AI is used to adapt teaching to each student’s learning level” was only about 2.8 (just at the lower end of “moderate extent” or even “small extent”). This indicates that while adaptive learning technologies exist, they were not widely implemented in these colleges. Teachers might not have the tools or training to truly personalize learning with AI beyond one-size-fits-all content. Collaborative learning enhancements through AI were also not strongly evident: the use of AI to facilitate student collaboration or interactive learning (for example, through AI-moderated forums or group project tools) was rated around the midpoint (neither low nor high). This suggests that core pedagogical approaches (like group discussions, project-based learning, etc.) had not yet been significantly transformed by AI in the classroom setting. In essence, teaching integration of AI remains focused on efficiency gains (automation of tasks) rather than fundamental pedagogical change. Many teachers still rely primarily on traditional methods for instruction delivery, bringing in AI tools selectively for convenience and support.

Among the three domains, language learning stood out as the most AI-integrated area. Respondents across all groups indicated a relatively great extent of AI utilization in language education activities. Students and teachers gave higher ratings for AI use in language learning tasks than for teaching or research tasks. This is reflected in the pervasive use of translation apps, language AI tutors, and speech recognition tools for language practice. For example, almost every student reported using some AI-based language tool, and teachers of English or other languages frequently assign or recommend such tools. The average extent rating for AI in language learning was in the upper-moderate range (approaching “to a great extent”). Many students agreed that AI tools (like intelligent tutoring systems for language practice) form a regular part of their language study routine. Teachers similarly noted that AI had become a valuable supplement for language instruction—for instance, AI-driven language learning platforms that provide adaptive vocabulary and grammar exercises, pronunciation feedback via AI speech analysis, or chatbots for conversational practice. These tools allow students to get instant corrections and practice beyond classroom hours, which was cited as a major benefit. One concrete finding was that adaptive learning systems in language education are widespread: many of both students and teachers indicated that AI provides individualized support in language learning to at least a moderate or high extent. This might include systems that adjust the difficulty of exercises based on student performance or recommend focused practice where a student is struggling. Such adaptive features help address varying proficiency levels, which is crucial in vocational colleges where students’ language backgrounds can differ. The high integration in language learning is likely because the available AI applications for language are mature and user-friendly, and language learning has well-defined tasks (translation, vocabulary practice, etc.) that AI can assist with effectively. Furthermore, external factors contribute to the popularity of plenty of mobile apps and online platforms for language learning (in both Chinese and English), which are popular and easy to adopt, often without requiring institutional provisioning. Therefore, students organically gravitate to these AI tools to help with their English classes or learning Mandarin (for non-native speakers). The data confirm that language learning is the area where AI’s educational impact is currently the most visible in Sichuan’s vocational colleges.

In contrast to language learning, AI utilization in research was the least developed area among those studied. Respondents across all groups indicated that using AI for research purposes was still relatively limited. On average, the extent of AI integration into research-related tasks was rated in the lower-moderate range. For instance, students and teachers gave modest ratings to items like “AI is used to assist in conducting research or projects.” Many students in vocational colleges do not engage in traditional research extensively (as their programs are skill-focused), which partly explains the lower use. Nonetheless, for those involved in research or project work (such as thesis projects or faculty research), AI has begun to play some role. The most common way AI was used in research was for information gathering and literature review. Faculty and even students noted that AI tools (like intelligent literature search engines or summarization tools) helped them find relevant academic articles or summarize large amounts of information—essentially accelerating the initial research phase. This is consistent with observations in higher education that AI can efficiently sift through databases and generate literature overviews. Indeed, respondents appreciated AI’s ability to reduce information overload by quickly providing summaries or identifying key sources, and this was one area of research support that got relatively positive feedback. However, more advanced research applications of AI (such as using AI for data analysis, simulations, or writing drafts) were not common. Only a minority of faculty had tried AI-based data analysis for research projects, and students seldom used AI beyond searching for information or checking grammar in their papers. The limited use in research is likely due to both lack of awareness and the nature of vocational programs—research is not a primary focus for most vocational students, and faculty may not have access to specialized AI research tools. Furthermore, some respondents expressed caution about relying on AI for research, due to concerns about accuracy or academic integrity (e.g., plagiarism risks if AI writes text). Overall, the findings highlight an untapped potential: while AI could greatly aid research (for example, through machine learning analysis of experimental data, or generating insights from big datasets), such applications were largely underutilized in this context as of the study period.

This research provides one of the first in-depth looks at how AI is being adopted and used in the context of Chinese higher vocational education. The findings paint a picture of moderate but growing integration of AI within Sichuan’s vocational colleges. In interpreting these results, it is useful to consider both the local context and broader trends in AI-in-education research.

Firstly, the moderate overall adoption of AI across these institutions indicates that AI technologies have moved beyond the pilot stage and into regular use, but not yet into fully transformative use. Most administrators, teachers, and students have at least some exposures to AI tools, which is an encouraging sign of diffusion. Teachers and administrators have embraced AI for routine tasks—a result consistent with innovation adoption theory that suggests early uses of a new technology often focus on improving efficiency in existing practices (Rogers, 2003). Our data showed teachers using AI to streamline lesson preparation and grading, and administrators using AI to automate student services and data management. These uses align with findings from other settings: for example, educators worldwide often initially use AI to handle burdensome tasks like grading or FAQ answering, thereby freeing time for more complex work (Crompton and Burke, 2023; Ocen et al., 2025). The fact that Sichuan’s vocational college faculty are doing the same suggests that they are keeping pace with global educational technology trends, at least in the domain of automating routine activities.

Where AI’s impact is most pronounced is in language education. The high usage of AI for language learning in our study reflects the abundance and maturity of AI language tools. This result is in line with prior research noting that language learning applications (intelligent tutors, translation apps, etc.) are among the most widespread AI applications in education (Crompton and Burke, 2023). In China, AI-driven language apps (for English learning especially) have been popular and well-supported, so it is logical that vocational students and teachers lean heavily on them. The implication is that when AI tools are readily accessible, aligned with user needs, and user-friendly, integration can happen quite organically. Language learning presented a clear use-case where AI adds value—personalized practice and instant feedback—, so both teachers and students eagerly utilized these tools. This success in the language domain could serve as a model for other subject areas: if similarly user-friendly AI tools (with clear pedagogical value) are developed for other disciplines (say, AI lab simulators for engineering trades or AI coaching for nursing procedures), we might expect uptake to be similarly enthusiastic.

In contrast, the limited use of AI for research and collaborative learning reveals the current boundaries of AI adoption in vocational colleges. Vocational institutions traditionally place less emphasis on research than academic universities, focusing more on practical training and employability. Our findings reflect this: neither faculty nor students are extensively using AI in research contexts, likely because research activities themselves are limited in these settings. Additionally, using AI for research often requires advanced tools and skills (like knowledge of data analysis software or research databases), which many vocational college staff and students may lack. This highlights an important consideration: contextual relevance. AI adoption will flourish where it aligns with the core activities and immediate needs of the institution. In vocational colleges, teaching and skill practice are core, whereas formal research is peripheral; hence, AI is naturally more integrated into teaching-related tasks than research tasks. Another factor could be trust and reliability—academic research demands high accuracy and credibility, and both faculty and students may be cautious about relying on AI for literature reviews or analysis due to concerns about getting flawed or non-validated information (a sentiment some interviewees expressed). This caution echoes warnings by scholars about the risks of unchecked AI outputs in academic work. As AI tools improve and as research becomes more data-driven even in vocational fields, we might see growth in this area, but for now, AI’s role in vocational college research remains supplementary at best.

One of the notable outcomes was that there were no significant discrepancies between the different stakeholder groups in how much AI they use. This uniformity can be interpreted positively. It suggests that the introduction of AI in these colleges has been relatively inclusive and well-communicated. Often with new technologies, one sees early adopters and laggards; here, however, administrators, faculty, and students all converged on a moderate level of use. This could imply strong institutional leadership and culture around AI. Possibly, administrative initiatives provided the push, teachers responded by integrating AI into their curriculum, and students followed suit as part of their learning process—a cohesive rollout. Literature on educational change supports the idea that synchronized adoption leads to smoother implementation (Law, 2010). From a practical standpoint, this parity means there is not a misalignment, such as teachers wanting to use AI but students are unprepared (or vice versa). Everyone is somewhat on the same page, which is beneficial for moving forward: any scaling up of AI integration can target the whole institutional community without needing to remediate a specific group.

However, we should nuance this finding: teachers and administrators did have slightly higher engagement and took more initiative with AI compared to students. In our context, that is expected—educators were often the ones driving the use of AI tools in their classes or work. The fact that students reported slightly lower adoption might indicate that students use AI when it is embedded in their coursework or recommended by teachers, but less so independently. This interpretation aligns with a study by Li et al. (2025), which found that undergraduate students’ perceptions of AI were significantly influenced by instructor guidance and institutional signals. In our study, many students were unaware they were “using AI” unless it was explicitly pointed out (e.g., they might use a grammar-checker without labeling it as AI). This calls for efforts to boost student AI literacy and agency—making students more conscious of AI tools and how to proactively use them for learning, beyond just what is assigned. Encouraging a culture where students experiment with AI for self-study or creative projects could elevate their engagement to match that of their instructors.

The positive correlation between training and AI usage is a critical insight for policy. It essentially validates that “if you train them, they will use it.” Faculty who underwent even minimal AI training integrated those skills into their teaching significantly more. This resonates with other findings in educational technology adoption that professional development is a key enabler of technology integration. In our context, where only a small fraction had training, the difference in usage is stark—those few with training were pioneers using AI to a greater extent, implying that lack of training is holding many others back from doing the same. It is a reminder that introducing AI tools alone is not enough; human capacity building is necessary to unlock the potential of those tools. Notably, even non-technical training like workshops on AI pedagogy or sharing best practices could demystify AI for less confident teachers and inspire new applications.

It is also worth noting that institutional support and clear policies can affect AI usage patterns. Our study period preceded the widespread issuance of formal AI-use policies within these colleges. In the absence of clear guidelines, some educators were treading carefully around issues like generative AI use in assignments. The development of a comprehensive AI policy (which we undertook as an outcome of this study) is timely. Such a policy can provide clarity – for example, explicitly permitting the use of AI tools for learning with proper attribution, or outlining how AI can be used in assessments—which in turn can embolden teachers and students to use AI constructively without fear of “doing something wrong.” Other research has highlighted that when institutions provide guidelines and support structures (like AI ethics guidelines, data privacy rules, and resource provisioning), it fosters a safer and more confident environment for technology adoption. Our recommendations later in this discussion incorporate these aspects.

China’s restrictions on many foreign AI services (e.g., limited access to ChatGPT without workarounds) shape not only tool choice but also the innovation trajectory of colleges. The heavy reliance on domestic platforms (ERNIE Bot, Qwen, Spark Desk) can be advantageous for Chinese-language performance, data localization, and alignment with national governance requirements. At the same time, reliance on a primarily domestic ecosystem may constrain pedagogical experimentation when certain features, plugins, or international teaching resources are unavailable or when interoperability across platforms is limited. For private vocational colleges with constrained budgets, policy-driven concentration in a few domestic vendors can reduce search costs and simplify compliance, but it can also amplify vendor lock-in and limit exposure to global benchmarking practices. To mitigate these risks, institutions can (a) prioritize platform-agnostic pedagogical designs (e.g., prompt literacy, evaluation rubrics, and disclosure practices that transfer across tools), (b) document tool capabilities and limitations systematically for internal benchmarking, and (c) participate in permitted international collaborations (e.g., joint evaluation datasets or published benchmarks) to maintain comparability even when direct access to some foreign tools is restricted.

The findings of this study carry several implications for educational practice and institutional policy in the vocational education sector (and potentially beyond). To harness AI’s benefits while addressing the challenges identified, a multi-pronged strategy is needed:

Capacity building through training: There is a clear need to invest in professional development for both faculty and students regarding AI. With only ~9% of teachers having formal training, colleges should implement comprehensive AI literacy programs. This could involve regular workshops that introduce teachers to available AI tools (e.g., demonstration sessions on using AI for creating assessments, or how to use adaptive learning software), as well as ongoing technical support or coaching for lesson integration. Training should not just cover tool usage, but also pedagogical strategies for AI (for instance, how to incorporate AI tutoring systems into homework assignments effectively, or how to supervise student use of AI so that it enhances rather than replaces learning). Administrators, too, would benefit from targeted training on AI for education management (e.g., using analytics dashboards or privacy management when deploying AI). For students, integrating AI literacy into the curriculum or orientation programs is recommended. Students entering these colleges could receive an orientation module on “Effective and Ethical Use of AI for Learning,” where they learn about the AI resources at their disposal (such as library chatbots, learning apps, etc.), how to use them to complement studying, and what constitutes misuse (to address integrity concerns). Building confidence and skills among all users will likely increase the extent and quality of AI utilization. Notably, similar calls for AI training in education have been made by policy bodies and researchers, emphasizing that without developing human capital, the technology cannot be used to its full potential (Office of Educational Technology, U.S. Department of Education, 2023).

Improving technological infrastructure: The infrastructural constraints must be addressed at the institutional and possibly provincial level. Ensuring that all vocational colleges have adequate internet bandwidth, modern computers or device access for students, and server capacity for AI applications is foundational. Investment in infrastructure upgrades—such as setting up AI computer labs or providing grants for colleges to acquire licensed AI software—would reduce barriers to adoption. In Sichuan, where some colleges are in less urban areas, it may require government or public-private initiatives to fund these improvements. Additionally, maintaining up-to-date software and providing IT support staff who can assist with AI tools will ensure the technology is usable day-to-day. Some colleges might explore sharing AI resources; for example, a cluster of colleges could jointly host an AI platform if individual funding is low. The implication here is that policymakers should view AI integration not just as a pedagogical issue but also as an infrastructure one—it requires tangible resource allocation to technology.

Developing clear AI usage policies: With the rapid rise of generative AI, educational institutions are grappling with policy questions. Our study underlines the importance of having a formal AI policy at the college level. This policy should cover aspects such as: acceptable use (e.g., guidelines on students using AI for assignments, requiring disclosure of AI assistance to prevent plagiarism), data privacy and security (ensuring compliance with regulations when using AI platforms that handle student data), and ethical standards (for example, forbidding the use of AI in ways that could reinforce bias or replacing human judgment where it’s crucial). By formally articulating these rules, colleges can provide a framework within which innovation can happen responsibly. During this research, we drafted a proposed Academic AI Policy for vocational colleges, which aligns with national directives and international best practices. It calls for things like establishing an AI ethics committee at the college, regular reviews of AI tool impacts, and integration of AI ethics into training sessions. Implementing such policies will help mitigate concerns expressed by participants about misuse or ethical pitfalls. Importantly, policies should be communicated and implemented through workshops so that everyone from top leadership to students understands them and why they matter (emphasize balancing innovation with caution, which is exactly what good policy enables).

Encouraging deeper integration into curriculum: Educators should be encouraged and supported to move beyond superficial use of AI (just for routine tasks) towards holistic curricular integration. This means rethinking some teaching methods, considering AI capabilities. For example, in teaching practice, instructors could incorporate AI-driven projects—such as having students use an AI tool to analyze a dataset or simulate a scenario relevant to their field—thereby directly engaging students with AI as part of learning. In language classes, since AI is already heavily used, teachers could formalize it by designing assignments that require using an AI translator, then critically evaluating its output, which both leverages and develops students’ higher-order thinking. The colleges could also revise curricula to include topics on AI basics and applications relevant to each vocational field (highlight successful cases of embedding digital and AI literacy in university programs). This will prepare students for the AI-infused workplaces they will enter. Faculty should be brought together (perhaps through communities of practice) to share best practices on using AI pedagogically. When teachers see concrete examples of improved learning outcomes from AI-enhanced methods (like adaptive quizzing raising student performance), they may be more motivated to adopt similar approaches.

Bridging the student engagement gap: Since the data suggests students were less engaged with AI unless guided, colleges should try to actively engage students with AI. This could involve creating more opportunities for students to experiment with AI in a safe learning environment. For instance, colleges might host innovation challenges or hackathons where students can create small projects using AI tools relevant to their trade (e.g., an IT student building a chat bot service for the campus, or design students using an AI image generator to aid a project, within ethical bounds). Such activities can demystify AI and spur interest. Additionally, promoting student-led AI clubs or interest groups could give students a peer-supported way to learn about new AI technologies and share usage tips. The goal is to foster a culture where students view AI not just as something for completing assignments faster, but as a tool for creativity, exploration, and self-improvement in their field. Over time, this could lead to students more independently integrating AI into their learning, which would maximize the benefits they get from these tools.

Ongoing support and governance: institutions must establish structures for sustained support and governance of AI integration. One recommendation from the study is for colleges to form AI steering committees or task forces that include representatives from administration, faculty, IT staff, and students (mirroring recommendations by Jobin et al., 2019 for inclusive AI governance). These committees can oversee the rollout of AI initiatives, ensure that tools are evaluated for effectiveness and fairness, and stay updated with emerging technologies or regulations. They would also be responsible for revising policies as needed and championing resources (budget for tools, training, etc.). Having a formal body signals the college’s commitment and provides a centralized place to address issues that arise (for example, if a teacher reports a problem with an AI system, the committee can investigate and liaise with vendors or authorities).

Addressing ethical and integrity issues: From a practice standpoint, the challenge of maintaining academic integrity in the age of AI requires strategic action. Teachers should design assessments and learning activities in ways that encourage ethical use of AI—for instance, assignments can be structured to require personal reflection or in-class components that AI cannot generate, or students can be asked to show how they used AI as a tool (thus making AI a part of the learning process rather than a shortcut around it). Educating students about the pitfalls of over-reliance on AI (such as potential inaccuracies or the loss of skill practice) is equally important. On the institutional side, deploying AI plagiarism detection or AI-output detection tools might be necessary as a deterrent (though these are evolving). An ethical implication is also ensuring AI tools used by the college are themselves fair and accessible—for example, making sure that an AI platform is usable by students with disabilities, or that it does not disadvantage any group of students. This might involve testing tools for bias and being transparent about how AI decisions (like automated grading) are made, aligning with principles of trustworthy AI in education (European Commission 2019; UNESCO, 2021).

This study had several limitations. First, the sample was limited to private higher vocational colleges in Sichuan Province, which may restrict generalizability to public institutions or other regions. Future studies should include broader or international samples to compare AI adoption patterns. Second, the cross-sectional design captured a single time point, despite the rapid evolution of AI technologies. Longitudinal research is needed to trace how adoption changes over time and impacts educational outcomes. Third, the reliance on self-reported data may introduce bias; incorporating objective usage metrics or direct classroom observations would improve reliability. Fourth, the qualitative component was modest and may not fully reflect stakeholder perspectives. In-depth qualitative methods like ethnography or focus groups could yield richer insights. Finally, while this study took a broad view of AI integration, more focused studies on specific domains such as AI in language learning, counseling, or administrative support are recommended to deepen understanding.