Student-Perceived Normalization Without Stabilization
AI Use, Readiness, Institutional Support, and Educational Concern among University of the Philippines Diliman Undergraduate Students
Abstract
Generative artificial intelligence is increasingly embedded in students’ academic work, but student adoption does not by itself indicate that universities have established the literacy, policy, assessment, and support conditions required for responsible use. This study examines student AI engagement at the University of the Philippines Diliman through a quantitative, descriptive-correlational, cross-sectional survey of 87 student respondents. It analyzes five interrelated domains: reported AI normalization, perceived AI readiness, perceived institutional stabilization, perceived legitimacy of AI-enabled academic support services, and educational concern. Findings show that students reported regular access to and use of AI tools for central academic tasks, particularly grammar checking, information search, summarization, paraphrasing, and drafting. However, this behavioral normalization coexisted with uneven perceived readiness, strong expectations for student and faculty training, only partial confidence in policy clarity and feedback mechanisms, selective support for AI-enabled academic services, and persistent concern about overreliance, fairness, and educational value. Exploratory correlation results further showed that AI use intensity was moderately associated with perceived readiness and support expectations, but was nearly unrelated to perceived institutional stabilization and educational concern. These patterns support a bounded interpretation of student-perceived normalization without stabilization: AI has become routine in student academic practice before students perceive the university’s literacy, policy, and support conditions for responsible use as fully stabilized. The study contributes a Global South public university case that links adoption, readiness, institutional support, legitimacy, and concern within a single analytical model, while avoiding unsupported claims about actual institutional governance or objective AI competence.
Keywords
artificial intelligence, higher education, AI literacy, institutional support, student perceptions, generative AI
1 Introduction
Generative artificial intelligence (AI) has moved quickly from a novel digital tool to a routine part of academic work. Students now encounter AI as a practical resource for writing, summarization, brainstorming, translation, information search, and other forms of knowledge production. Recent studies show that students recognize both the benefits and risks of generative AI, including its potential to support learning while raising concerns about ethics, plagiarism, academic integrity, and overreliance (Chan and Hu, 2023; Francis et al., 2025). The key question for higher education is therefore no longer whether students will encounter AI. It is whether universities can provide the literacy development, policy clarity, pedagogical guidance, and assessment infrastructure needed to make student AI use responsible, equitable, and educationally meaningful.
This article examines that problem through the idea of student-perceived normalization without stabilization. In this study, normalization refers to the extent to which students report regular access to and use of AI tools for central academic tasks. Stabilization refers not to the verified institutional state of AI governance, but to students’ perception that the university provides adequate literacy training, policy guidance, feedback channels, faculty preparation, and support mechanisms for responsible AI engagement. The argument advanced here is deliberately bounded: AI may be behaviorally normalized in students’ academic routines before students perceive the institutional conditions for responsible use as stable.
The empirical problem is especially important in Global South public university contexts, where institutions may face uneven resources, rapidly changing policy demands, and pressure to respond to technologies shaped largely outside local governance systems. UP Diliman, as the flagship campus of the national university and a large public research university, provides a relevant setting for examining this issue.
This study asks:
- What forms and frequencies of AI use do students report in academic work at UP Diliman?
- What levels of perceived AI readiness do students report, and how does perceived readiness differ between immediate academic use and future workplace preparedness?
- How do students evaluate current and desired institutional support for AI use, including training, policy clarity, feedback mechanisms, and participation in decision-making?
- Which AI-enabled academic support initiatives do students perceive as educationally useful, and which receive weaker legitimacy?
- What concerns do students report regarding overreliance, fairness, and educational value, and how do these concerns coexist with reported AI use?
- What exploratory descriptive variation appears in AI use, readiness, institutional expectations, and concern across academic level, year level, or academic cluster?
The study contributes by operationalizing student-perceived normalization without stabilization as a measurable, multi-dimensional condition; by providing a student-level account of AI engagement in a Philippine public university; and by identifying areas where student adoption appears to outpace perceived literacy, policy, and support infrastructure.
2 Literature Review
Research on AI in higher education has expanded rapidly since the public release of widely accessible generative AI systems. One stream documents student perceptions and uses. Chan and Hu (2023) found that students viewed generative AI as potentially useful for learning while also identifying challenges related to ethics, plagiarism, accuracy, and overreliance. A large cross-national study similarly shows that higher education students use ChatGPT for study-related tasks while raising questions about regulation, skills, and career preparation (Ravšelj et al., 2025). Recent work also emphasizes that AI is increasingly embedded in higher education practice and that institutional responses must balance innovation with academic integrity (Francis et al., 2025; Kofinas et al., 2025).
A second stream focuses on AI literacy. AI literacy is not merely tool familiarity. Long and Magerko (2020) define it through competencies that allow people to critically evaluate, communicate with, and use AI. Ng et al. (2021) similarly conceptualize AI literacy as a multidimensional capacity involving knowledge, use, evaluation, and ethical awareness. In higher education, this distinction matters because frequent use of generative AI does not necessarily mean that students can judge outputs, identify limitations, understand data and bias issues, or use AI in ways consistent with disciplinary and ethical norms. Delcker et al. (2024) also show that AI competence is relevant to intended and actual AI-tool use in learning processes.
A third stream concerns institutional support and governance. Technology acceptance theory has long argued that technology use is shaped by facilitating conditions as well as individual perceptions of usefulness and effort (Venkatesh et al., 2003). In the generative AI context, facilitating conditions include training, assessment guidance, policy clarity, faculty preparation, and mechanisms for student feedback. Chan (2023) argues that higher education needs comprehensive AI policy education frameworks that address pedagogical, governance, and ethical dimensions rather than treating AI as a narrow technical issue.
A fourth stream addresses academic integrity, assessment, and educational concern. Scholars increasingly caution that a detector-centered response to generative AI is insufficient. Ardito (2024) argues for robust assessment approaches rather than reliance on AI detection tools. Kofinas et al. (2025) links generative AI to vulnerabilities in authentic assessment and highlights the need to balance educational benefits against risks to authorship and integrity. Francis et al. (2025) similarly frames generative AI as a dual-edged technology that can support personalized learning while raising concerns about authenticity, equity, and essential cognitive skills.
Digital inequality and institutional capacity further complicate AI adoption in Global South and resource-constrained higher education settings. Research on higher education digital divides in developing-country contexts shows that access, motivation, skills, and actual use remain uneven and are shaped by infrastructure, socioeconomic conditions, and institutional support (Niyigena et al., 2020; Soomro et al., 2020). This matters for AI governance because generative AI tools may diffuse quickly through student practice even where localized policy, training, and support systems develop more slowly. The present study therefore treats the Philippine public university context not as a simple geographic label, but as a setting where student adoption, institutional capacity, and responsible-use guidance must be examined together.
Despite these advances, many studies examine use, literacy, institutional support, or concern separately. The present study responds by integrating student-reported use, readiness, perceived institutional stabilization, support legitimacy, and concern in one descriptive-correlational analysis.
2.1 Conceptual Framework / Analytical Model
The conceptual framework treats student AI engagement as a multi-dimensional condition rather than a single attitude or behavior. It uses four analytically defined constructs.
Reported AI normalization is the behavioral dimension. It refers to regular student access to AI tools, frequency of use, range of tools used, and use for central academic tasks such as writing, research, and summarization.
Perceived AI readiness is the competence dimension. It refers to students’ self-rated knowledge, skills, confidence, critical use capacity, and perceived preparedness for future study or work involving AI. Because the study uses self-report, readiness is interpreted as perceived readiness rather than objective competence.
Perceived institutional stabilization is the institutional dimension. It refers to students’ perception that the university provides adequate AI literacy training, policy clarity, feedback and participation mechanisms, faculty preparation, and support structures. The study does not directly audit university policy or faculty practice.
Educational concern is the evaluative dimension. It refers to students’ perceived risks related to overreliance, fairness in evaluation, academic integrity, and possible reduction of educational value.
The model in Figure 1 organizes these constructs around a central interpretive claim: student AI engagement may reflect behavioral normalization under conditions of uneven readiness, incomplete perceived institutional stabilization, and persistent educational concern. Because the dataset is small and cross-sectional, the study uses analytic propositions rather than causal hypotheses:
P1. High reported AI use may coexist with uneven perceived readiness.
P2. Strong institutional expectations may indicate a perceived stabilization gap rather than simple enthusiasm for AI.
P3. Educational concern may persist even among students who report regular AI use.
P4. Perceived legitimacy of AI-enabled support is likely stronger when initiatives are aligned with immediate academic tasks.
P5. Variation in AI engagement by academic cluster and level should be treated as exploratory.
An optional hypothesis-testing extension is provided in Appendix A for future studies with larger samples.
3 Methodology
This study used a quantitative, descriptive-correlational, cross-sectional survey design. The design was appropriate because the study aimed to describe student-reported AI use and examine how use, perceived readiness, perceived institutional stabilization, support legitimacy, and concern coexist within a single institutional setting. The design does not support causal inference, direct measurement of institutional governance, or claims about actual AI competence.
The study was conducted at UP Diliman, the flagship campus of the University of the Philippines. The target population was currently enrolled undergraduate and graduate students. The final analytic sample contained 87 respondents. Because the sample was small and non-probabilistic, it is treated as a context-specific basis for descriptive and exploratory analysis rather than as a statistically representative sample of the UP Diliman student population.
3.1 Sampling and Recruitment
Participants were recruited through purposive, voluntary-response online recruitment. Eligibility was limited to currently enrolled UP Diliman undergraduate and graduate students during the Second Semester of Academic Year 2025-2026 who had used or encountered AI tools in an academic setting. The survey link was disseminated through formal email requests to Offices of College Secretaries, academic departments, student organizations, and student councils across UP Diliman, and through university-specific communities on Facebook and Reddit. Data were collected through Google Forms over a four-week period. The broader cleaned study dataset contained 108 valid student responses; this manuscript retained 87 respondents with complete data on the core variables required to construct the cross-domain indices. Descriptive and correlation analyses were then conducted on available valid responses for each item or index. Because the survey link was distributed through multiple institutional and social media channels rather than a closed sampling frame, a response rate could not be calculated. Because recruitment used purposive and voluntary-response sampling, the sample should not be interpreted as statistically representative of the UP Diliman student population. This limitation is especially relevant for online educational surveys, where response rates and sample composition can be shaped by access, motivation, and recruitment channels (Saleh and Bista, 2017).
The instrument was an online structured questionnaire. It measured reported AI use, perceived AI readiness, institutional support and expectations, perceived usefulness of AI-enabled academic services, and educational concern, as shown in Table 1.
| Construct | Definition | Indicators |
|---|---|---|
| Reported AI normalization | Regular access to and use of AI for academic work | Access, use frequency, tools used, writing, research, summarization, drafting |
| Perceived AI readiness | Self-reported capacity to use AI effectively and prepare for AI-shaped study or work | Knowledge and skills, confidence, workforce preparedness |
| Perceived institutional stabilization | Student perception that the university provides support conditions for responsible AI use | AI integration, policy awareness, guidelines, training, faculty preparation, feedback, participation |
| Support legitimacy | Perceived educational usefulness of proposed AI-enabled academic services | Research tools, tutoring chatbots, writing assistants, personalized learning, career tools, wellbeing tools |
| Educational concern | Perceived risks of AI use for learning, fairness, and educational value | Overreliance, evaluation fairness, reduced educational value |
Data analysis proceeded in five steps: descriptive statistics for categorical and multiple-response variables; descriptive summaries of Likert-type items; composite indices for readiness, institutional stabilization, support legitimacy, and concern; exploratory internal-consistency checks for multi-item indices; and Spearman correlations to examine cross-domain relationships. After restricting the dataset to respondents with complete core variables for the composite indices, available-case analysis was used for item-level descriptive summaries and pairwise correlations where minor item nonresponse remained. Subgroup patterns by academic cluster and level were visualized but not treated as confirmatory tests.
The study was conducted with prior ethics approval from the Research Ethics Board of the University of the Philippines Diliman. Participation was voluntary, informed consent was obtained electronically before survey access, and respondents were informed that they could discontinue participation at any time before submission. No directly identifying information was collected. Data were stored securely and accessed only by the researchers.
4 Results
The results show that AI had already become an ordinary feature of academic life for the respondents, but that its normalization was not matched by equally stable levels of perceived readiness, institutional support, or trust. Across the dataset, students reported substantial access to AI-capable technologies and regular use of AI tools for core academic tasks. At the same time, their self-reported readiness remained uneven, their expectations for university support were strong, and their concerns about overreliance, fairness, and educational value persisted. Taken together, these findings support a bounded interpretation: AI was already behaviorally normalized in student practice, but students did not perceive the literacy, policy, and support conditions for responsible use as fully stabilized.
4.1 Respondent Profile
The analytic sample consisted of 87 respondents from the University of the Philippines Diliman. The respondent pool was heterogeneous, drawing from multiple academic clusters, year levels, and degree programs. This matters because the study aimed to capture AI engagement across a diverse academic environment rather than within a single disciplinary or programmatic context. Recent higher education research similarly emphasizes the need to document variation in AI use, attitudes, and literacy across student groups in order to understand how universities can support equitable engagement (Brown et al., 2025).
The composition of the sample showed that undergraduate students formed the majority of respondents, with a smaller group of graduate students also represented. The academic-cluster distribution included students from Science and Technology, Management and Economics, and Social Sciences and Law, alongside a small number of respondents whose entries did not map cleanly to a predefined cluster category. This pattern suggests that the study captured AI engagement in a university setting where disciplinary expectations and learning demands were not uniform.
As shown in Table 2, the respondent pool was distributed across academic levels, year levels, and academic clusters, providing a heterogeneous basis for descriptive analysis.
| Domain | Category | n | Percent |
|---|---|---|---|
| Academic level | Undergraduate | 45 | 51.7% |
| Academic level | Graduate | 42 | 48.3% |
| Year level | 1st | 12 | 13.8% |
| Year level | 2nd | 15 | 17.2% |
| Year level | 3rd | 12 | 13.8% |
| Year level | 4th | 16 | 18.4% |
| Year level | 5th | 3 | 3.4% |
| Year level | Not applicable / not stated | 29 | 33.3% |
| Academic cluster | Science and Technology | 34 | 39.1% |
| Academic cluster | Management and Economics | 27 | 31.0% |
| Academic cluster | Arts and Letters | 14 | 16.1% |
| Academic cluster | Social Sciences and Law | 11 | 12.6% |
| Academic cluster | Unclassified | 1 | 1.1% |
The distribution provides a heterogeneous descriptive basis for examining student AI engagement, while remaining limited by the study’s non-probabilistic recruitment strategy.
4.2 Access to and Use of AI
The data showed that access to AI-capable technology was widespread among respondents. Most students reported that they had the technological means necessary to use AI tools, indicating that AI engagement in this sample was not constrained primarily by basic access barriers. AI use was also substantial rather than incidental. Respondents reported using AI with meaningful frequency, suggesting that AI had moved beyond novelty and into routine academic practice.
As shown in Figure 2, AI use had already moved into routine academic practice for many respondents.
The figure makes visible the extent to which AI was already part of ordinary study routines rather than an occasional or experimental technology.
The most commonly used tool was ChatGPT, followed by other widely available AI-assisted applications. This concentration around a small set of dominant tools indicates that students were relying primarily on accessible, general-purpose systems rather than highly specialized platforms.
Table 3 shows the concentration of tool use around a relatively small set of accessible, general-purpose systems.
| AI tool | n | Percent of respondents |
|---|---|---|
| ChatGPT | 67 | 77.0% |
| Google Gemini | 59 | 67.8% |
| Grammarly | 41 | 47.1% |
| Microsoft Copilot | 13 | 14.9% |
| Perplexity | 11 | 12.6% |
| Google NotebookLM | 4 | 4.6% |
| DeepSeek | 3 | 3.4% |
This distribution indicates that adoption was concentrated in platforms that are easy to access and readily applicable to writing- and research-related work.
The purposes of use were equally revealing. Respondents most often used AI for writing assistance, summarization, brainstorming, and research-related support. These are not peripheral activities; they sit close to the core of academic knowledge work.
As shown in Table 4, AI use was concentrated in writing- and research-related tasks rather than in peripheral or highly specialized activities.
| Purpose of use | n | Percent of respondents |
|---|---|---|
| Check grammar | 67 | 77.0% |
| Search for and explain information | 66 | 75.9% |
| Summarize documents | 48 | 55.2% |
| Paraphrase a document | 32 | 36.8% |
| Create a first draft | 29 | 33.3% |
| For flow, coherence, and choice of words | 1 | 1.1% |
The table shows that students most often used AI to support the production, refinement, and organization of academic work, reinforcing the argument that AI had become embedded in core knowledge tasks.
4.3 AI Readiness
While AI use was widespread, self-reported readiness was more uneven. Respondents generally perceived themselves as having at least moderate knowledge and skills in using AI effectively, but the distribution of responses suggested variability rather than uniformly high confidence. This distinction is important because it shows that routine use and stable competence are not identical.
Table 5 summarizes the distribution of readiness indicators and shows that confidence was present, though not uniformly strong across domains.
| Statement | Mean | SD | Agree (4-5) | Neutral (3) | Disagree (1-2) |
|---|---|---|---|---|---|
| I have sufficient knowledge and skills to use AI effectively | 3.98 | 0.96 | 72.4% | 19.5% | 8.0% |
| I feel prepared for a future workplace that uses AI | 3.39 | 1.12 | 48.3% | 28.7% | 23.0% |
The pattern suggests that respondents had moved beyond mere exposure. Yet readiness remained uneven enough to show that routine use should not be treated as evidence of stable competence across the sample.
4.4 Institutional Support and Expectations
The strongest pattern in the dataset was the extent to which respondents expected the university to play an active role in shaping AI use. Students did not appear to treat AI as a purely personal or self-managed issue. Instead, they expressed clear expectations regarding AI literacy courses, student training, faculty training, policy guidance, feedback mechanisms, and participation in AI-related institutional decision-making.
As shown in Table 6, expectations for training, policy, guidance, and participation were consistently strong across institutional support domains.
| Statement | Mean | SD | Agree (4-5) | Neutral (3) | Disagree (1-2) |
|---|---|---|---|---|---|
| The university is integrating AI effectively | 3.17 | 1.06 | 34.5% | 44.8% | 20.7% |
| The university should increase AI use | 3.23 | 1.28 | 47.1% | 21.8% | 31.0% |
| More AI literacy courses are needed | 4.02 | 1.08 | 74.7% | 14.9% | 10.3% |
| Students need training on AI | 4.23 | 1.06 | 82.8% | 9.2% | 8.0% |
| Faculty need training on AI | 4.26 | 1.12 | 81.6% | 10.3% | 8.0% |
| Students should be involved in AI decisions | 4.14 | 1.10 | 78.2% | 12.6% | 9.2% |
| The university seeks student feedback on AI | 2.75 | 1.10 | 19.5% | 42.5% | 37.9% |
| Clear AI guidelines are in place | 3.24 | 1.15 | 36.8% | 39.1% | 24.1% |
| I am aware of AI-related policies | 3.38 | 1.30 | 48.3% | 29.9% | 21.8% |
The table highlights a recurring pattern in the dataset: students did not treat AI as a self-managed issue, but expected structured institutional scaffolding around its use.
Responses regarding policy and guidance were especially important. Students appeared to value the existence of clear rules and guidance, yet policy awareness was not uniformly strong. This points to a condition of partial institutionalization: AI was present in practice, but the structures meant to govern or clarify its use were not yet equally visible or settled.
4.5 Perceived AI-Enabled Academic Support Initiatives
Table 7 shows that support was strongest for AI-enabled initiatives aligned with immediate academic practices, especially research assistance, writing support, and tutoring-like help.
| Statement | Mean | SD | Agree (4-5) | Neutral (3) | Disagree (1-2) |
|---|---|---|---|---|---|
| AI research tools | 3.85 | 1.19 | 74.7% | 9.2% | 16.1% |
| AI tutoring chatbots | 3.28 | 1.25 | 52.9% | 19.5% | 27.6% |
| AI writing assistants | 3.37 | 1.19 | 51.7% | 25.3% | 23.0% |
| AI personalized learning paths | 3.22 | 1.13 | 44.8% | 28.7% | 26.4% |
| AI career recommendation systems | 3.25 | 1.26 | 50.6% | 23.0% | 26.4% |
| AI mock interview avatars | 2.98 | 1.33 | 41.4% | 25.3% | 33.3% |
| AI wellbeing support systems | 3.14 | 1.37 | 44.8% | 25.3% | 29.9% |
This pattern suggests that respondents were selective rather than indiscriminately enthusiastic, favoring services with clear pedagogical relevance and immediate academic payoff.
4.6 Concerns Regarding AI
The normalization of AI use did not eliminate concern. On the contrary, the dataset showed that concerns about AI formed a central dimension of student engagement. The most prominent concerns involved overreliance on AI, fairness in evaluation, and the possible reduction of educational value.
As summarized in Table 8, concern levels were substantial across all major dimensions included in the study.
| Statement | Mean | SD | Agree (4-5) | Neutral (3) | Disagree (1-2) |
|---|---|---|---|---|---|
| Over-reliance on AI may harm academic performance | 4.32 | 1.16 | 81.6% | 6.9% | 11.5% |
| Over-reliance on AI may reduce educational value | 4.05 | 1.20 | 73.6% | 13.8% | 12.6% |
| AI may affect fairness in evaluation | 4.29 | 1.08 | 80.5% | 9.2% | 10.3% |
The table makes clear that normalization of AI use did not erase evaluative caution; concern remained central to how students understood the technology’s academic role.
This coexistence of use and concern is one of the most significant findings of the study. It shows that student AI engagement was not uncritical. Respondents appeared willing to use AI while simultaneously questioning its implications for learning and fairness.
4.7 Cross-Domain Relationships
To evaluate the central proposition more directly, the analysis created exploratory composite indices for readiness, perceived institutional stabilization, institutional support expectations, support legitimacy, and educational concern. These indices are not treated as fully validated scales; they are descriptive summaries used to examine whether the manuscript’s core constructs move together in the expected directions.
Table 9 reports internal-consistency estimates for each multi-item index. The estimates should be interpreted cautiously because the sample is small and several domains contain only a limited number of items.
| Index | Items | Cronbach alpha | Two-item rho | Spearman-Brown |
|---|---|---|---|---|
| Perceived readiness | 2 | 0.71 | 0.56 | 0.72 |
| Perceived institutional stabilization | 4 | 0.79 | ||
| Institutional support expectations | 5 | 0.89 | ||
| Support legitimacy | 7 | 0.87 | ||
| Educational concern | 3 | 0.83 |
Note. For the two-item perceived readiness index, the inter-item Spearman correlation and Spearman-Brown coefficient are reported alongside alpha. Blank cells indicate that the statistic was not applicable.
Table 10 summarizes Spearman correlations among AI use intensity and the composite indices. The purpose is not to infer causal effects, but to test whether the descriptive argument is empirically coherent across domains.
| Domain | AI use intensity | Perceived readiness | Perceived institutional stabilization | Support expectations | Support legitimacy | Educational concern |
|---|---|---|---|---|---|---|
| AI use intensity | 1.00 | 0.37 | 0.04 | 0.35 | 0.33 | 0.00 |
| Perceived readiness | 0.37 | 1.00 | 0.41 | 0.53 | 0.31 | 0.10 |
| Perceived institutional stabilization | 0.04 | 0.41 | 1.00 | 0.31 | 0.25 | 0.14 |
| Support expectations | 0.35 | 0.53 | 0.31 | 1.00 | 0.58 | 0.07 |
| Support legitimacy | 0.33 | 0.31 | 0.25 | 0.58 | 1.00 | -0.12 |
| Educational concern | 0.00 | 0.10 | 0.14 | 0.07 | -0.12 | 1.00 |
The correlation pattern supports a bounded interpretation of student-perceived normalization without stabilization. AI use intensity was moderately associated with perceived readiness (rho = .37) and support expectations (rho = .35), suggesting that students who used AI more frequently also tended to feel more capable and to desire stronger institutional scaffolding. However, AI use intensity was almost unrelated to perceived institutional stabilization (rho = .04) and educational concern (rho = .00). Frequent use therefore did not necessarily correspond to stronger perceptions of institutional readiness, nor did it reduce concern about overreliance, fairness, or educational value. The strongest observed association was between support expectations and support legitimacy (rho = .58), indicating that students who wanted stronger institutional support were also more likely to view AI-enabled academic services as useful. Support legitimacy and educational concern showed only a weak negative association (rho = -.12), suggesting that concern did not strongly suppress perceived usefulness. Taken together, these patterns show that student AI adoption, perceived readiness, institutional support demands, and concern coexist without collapsing into a single pro-AI or anti-AI attitude.
4.8 Descriptive Patterns Across Student Groups
Descriptive subgroup patterns suggested that AI engagement was not perfectly uniform across academic clusters, levels, and year groups. Some groups appeared to integrate AI more intensively into routine academic work, while others seemed more cautious or less fully normalized in their use.
Figure 3 illustrates variation in AI use across academic clusters and shows that normalization was not experienced in identical ways throughout the sample.
Although these subgroup patterns remain descriptive, they suggest that institutional responses should not assume a single or uniform student experience of AI.
Taken together, the subgroup comparisons reinforce the broader conclusion of the results chapter: AI was widely present across the student population, but its integration, interpretation, and perceived implications were uneven.
5 Discussion
The findings support the article’s central claim, but in a deliberately bounded form: among respondents, AI appears behaviorally normalized before the perceived institutional conditions for responsible use are fully stabilized. This does not prove that UP Diliman lacks AI governance. It shows that students’ reported use, readiness, support expectations, and concerns do not align in a way that would justify treating adoption as settled integration.
First, AI use has become embedded in core academic work. The frequency, tool-use, and purpose-of-use results show that respondents used AI for writing, summarization, information search, drafting, and language support. These are not peripheral tasks; they are central to how students produce and organize academic work. This finding aligns with student-perception studies showing that generative AI is valued for learning support but also experienced through mixed judgments about benefits and risks (Chan and Hu, 2023).
Second, behavioral normalization does not equal readiness. The readiness results show that students may report functional confidence while still showing uneven preparedness for future AI-shaped academic and workplace contexts. This distinction is central to AI literacy scholarship, which frames literacy as critical, evaluative, and ethical competence rather than mere tool exposure (Long and Magerko, 2020; Ng et al., 2021). The implication is that universities should not infer competence from use.
Third, students perceive institutional stabilization as incomplete. Strong expectations for student training, faculty training, policy clarity, feedback mechanisms, and participation indicate that students do not regard AI as a purely individual responsibility. In technology-acceptance terms, this is a facilitating-conditions problem (Venkatesh et al., 2003). In policy terms, it supports the need for AI education frameworks that clarify how AI should be used, disclosed, assessed, and governed in teaching and learning (Chan, 2023).
Fourth, support for AI-enabled services is selective rather than unconditional. Respondents were more receptive to services aligned with immediate academic tasks, such as research support, writing support, and tutoring-like assistance, than to initiatives with less direct academic fit. This pattern suggests that legitimacy depends on educational purpose, not simply on the presence of AI.
Fifth, concern is not resistance. Students can use AI regularly while remaining concerned about overreliance, fairness, and educational value. This finding is consistent with academic integrity and assessment literature warning that universities need assessment redesign and transparent guidance rather than simple detection-based responses (Ardito, 2024; Kofinas et al., 2025). The coexistence of use and concern is therefore one of the study’s strongest contributions: it shows students as active evaluators of AI, not merely adopters.
Finally, subgroup patterns should be read as exploratory signals. The sample is too small for strong claims by cluster or academic level, but the descriptive variation suggests that institutional AI responses should not assume a single student experience. A public research university may need both university-wide policy clarity and discipline-sensitive guidance.
5.1 Conclusion and Recommendations
This study examined AI engagement among UP Diliman students across reported use, perceived readiness, perceived institutional stabilization, support legitimacy, and educational concern. The findings show a condition of student-perceived normalization without stabilization: AI is already used for central academic tasks, but students still report uneven readiness, strong demand for guidance, selective legitimacy judgments, and persistent concern.
Six recommendations follow. First, the university should develop tiered AI literacy provision that separates functional tool use, critical evaluation, ethical judgment, and discipline-specific practice. Second, it should create transparent assessment and disclosure policies that distinguish prohibited use, permitted support, required disclosure, and discipline-specific expectations. Third, faculty development should precede large-scale AI integration, especially in assessment redesign, feedback practices, and equitable access. Fourth, student participation should be built into AI governance because respondents strongly endorsed involvement in AI-related decisions. Fifth, AI-enabled services should be prioritized where students perceive clear academic fit, especially research, writing, and tutoring support. Sixth, libraries and learning support units should be considered part of AI literacy infrastructure, while acknowledging that these units were not directly measured in this study.
5.2 Limitations
The findings should be interpreted with caution. The sample was small, single-institutional, and non-probabilistic, which limits generalizability and makes subgroup analysis exploratory. The data were self-reported, so they may reflect recall error, social desirability, common-method bias, and perceived rather than actual competence. The cross-sectional design prevents causal inference and cannot show how student engagement changes over time. The study did not directly measure institutional policies, faculty practice, library services, learning outcomes, or objective AI literacy. Future research should use larger stratified samples, mixed methods, policy-document analysis, faculty and staff perspectives, longitudinal designs, and objective or performance-based measures of AI literacy.
6 Appendix A
Appendix Table A1. Item composition of composite indices.
| Index | Items included |
|---|---|
| Perceived readiness | Self-rated AI knowledge and skills; preparedness for a future workplace that uses AI |
| Perceived institutional stabilization | University AI integration; clear AI guidelines; policy awareness; university seeks student feedback |
| Institutional support expectations | University should increase AI use; more AI literacy courses are needed; students need AI training; faculty need AI training; students should be involved in AI decisions |
| Support legitimacy | AI research tools; AI tutoring chatbots; AI writing assistants; AI personalized learning paths; AI career recommendation systems; AI mock interview avatars; AI wellbeing support systems |
| Educational concern | Overreliance may harm academic performance; overreliance may reduce educational value; AI may affect fairness in evaluation |
References
Ardito, C. G. (2024). Generative AI detection in higher education assessments. New Directions for Teaching and Learning, 2025(182), 11–28. https://doi.org/10.1002/tl.20624
Brown, R. D., Sillence, E., and Branley-Bell, D. (2025). AcademAI: Investigating AI usage, attitudes, and literacy in higher education and research. Journal of Educational Technology Systems, 54(1), 6–33. https://doi.org/10.1177/00472395251347304
Chan, C. K. Y. (2023). A comprehensive AI policy education framework for university teaching and learning. International Journal of Educational Technology in Higher Education, 20(1). https://doi.org/10.1186/s41239-023-00408-3
Chan, C. K. Y., and Hu, W. (2023). Students’ voices on generative AI: Perceptions, benefits, and challenges in higher education. International Journal of Educational Technology in Higher Education, 20(1). https://doi.org/10.1186/s41239-023-00411-8
Delcker, J., Heil, J., and Ifenthaler, D. (2024). First-year students AI-competence as a predictor for intended and de facto use of AI-tools for supporting learning processes in higher education. International Journal of Educational Technology in Higher Education, 21(1). https://doi.org/10.1186/s41239-024-00452-7
Francis, N., Jones, S., and Smith, D. P. (2025). Generative AI in higher education: Balancing innovation and integrity. British Journal of Biomedical Science, 81. https://doi.org/10.3389/bjbs.2024.14048
Kofinas, A., Tsay, C. H.-H., and Pike, D. A. (2025). The impact of generative AI on academic integrity of authentic assessments within a higher education context. British Journal of Educational Technology, 56(6), 2522–2549. https://doi.org/10.1111/bjet.13585
Long, D., and Magerko, B. (2020). What is AI literacy? Competencies and design considerations. Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, 1–16. https://doi.org/10.1145/3313831.3376727
Ng, D. T. K., Leung, J. K. L., Chu, S. K. W., and Qiao, M. S. (2021). Conceptualizing AI literacy: An exploratory review. Computers and Education: Artificial Intelligence, 2, 100041. https://doi.org/10.1016/j.caeai.2021.100041
Niyigena, J.-P., Jiang, Q., Ziou, D., Shaw, R. S., and Hasan, N. (2020). Modeling the measurements of the determinants of ICT fluency and evolution of digital divide among students in developing countries–east africa case study. Applied Sciences, 10(7), 2613. https://doi.org/10.3390/app10072613
Ravšelj, D., Keržič, D., and Tomaževič, N. (2025). Higher education students’ perceptions of ChatGPT: A global study of early reactions. PLOS ONE, 20(2), e0315011. https://doi.org/10.1371/journal.pone.0315011
Saleh, A., and Bista, K. (2017). Examining factors impacting online survey response rates in educational research: Perceptions of graduate students. Journal of MultiDisciplinary Evaluation, 13(29), 63–74. https://doi.org/10.56645/jmde.v13i29.487
Soomro, K. A., Kale, U., Curtis, R., Akcaoglu, M., and Bernstein, M. (2020). Digital divide among higher education faculty. International Journal of Educational Technology in Higher Education, 17(1). https://doi.org/10.1186/s41239-020-00191-5
Venkatesh, V., Morris, M. G., Davis, G. B., and Davis, F. D. (2003). User acceptance of information technology: Toward a unified view. MIS Quarterly, 27(3), 425–478. https://doi.org/10.2307/30036540