Quantifying Perception-Based Student Success with Generative AI: An Exploratory Monte Carlo Simulation

Quantifying Student Success with Generative AI: A Monte Carlo Simulation Informed by Systematic Review Seyma Yaman Kayadibi Victoria University [email protected] Abstract The exponential development of generative artificial intelligence (GenAI) technologies like ChatGPT has raised increasing curiosity about their use in higher education, specifically with respect to how students view them, make use of them, and the implications for learning outcomes. This paper employs a hybrid methodological approach involving a systematic literature review and simulationbased modeling to explore student perceptions of GenAI use in the context of higher education. A total of nineteen empirical articles from 2023 through 2025 were selected from the PRISMA-based search targeting the Scopus database. Synthesis of emerging patterns from the literature was achieved by thematic categorization. Six of these had enough quantitative information, i.e., itemlevel means and standard deviations, to permit probabilistic modeling. One dataset, from the resulting subset, was itself selected as a representative case with which to illustrate inverse-variance weighting by Monte Carlo simulation by virtue of its well-designed Likert-scale format and thematic alignment with the use of computing systems by the researcher. The simulation provided a composite “Success Score” forecasting the strength of the relationship between student perceptions and learning achievements. Findings reveal that attitude factors concerned with usability and realworld usefulness are significantly better predictors of positive learning achievement than are affective or trust-based factors. Such an interdisciplinary perspective provides a unique means of linking thematic results with predictive modelling, resonating with longstanding controversies about the proper use of GenAI tools within the university. 1. Introduction Artificial intelligence (AI) is playing an increasingly pivotal role in higher education, influencing not only how students engage academically but also how they navigate everyday life (Stöhr et al., 2024). Among recent advancements, conversational AI tools powered by natural language processing and machine learning, most notably ChatGPT, released by OpenAI in November 2022, have attracted global attention. By enabling human-like dialogue and complex text-based interactions, ChatGPT has been described as a paradigm-shifting technology in education. As noted by Stöhr et al. (2024), within two months of release, ChatGPT was reported to have reached roughly 100 million users, making it, at the time, the fastest-growing consumer application. Its widespread adoption has prompted urgent pedagogical, ethical, and institutional debates, as universities and students grapple with new norms, blurred boundaries, and evolving expectations around academic integrity and tool usage. Moreover, students are not only using GenAI tools like ChatGPT in unprecedented numbers but are also reinterpreting the ethical landscape surrounding their use. A recent study among Australian undergraduates found that more than one-third had used ChatGPT for assessmentrelated tasks, such as writing or revising assignments. Yet, the majority did not perceive this as academic dishonesty, suggesting a shifting understanding of institutional integrity in the age of AI (Gruenhagen et al., 2024). Such findings reflect how institutional integrity and perceptions of ethical boundaries are being reshaped by GenAI’s widespread availability. A growing number of survey-based studies have explored how university students across different cultures and disciplines perceive and use ChatGPT. Abbas et al. (2024) found that students in Pakistan experiencing high academic workload and time pressure were more likely to adopt ChatGPT, whereas reward-sensitive students avoided it due to fear of academic penalties. In Iran, Rahimi et al. (2025) showed that personalized engagement with ChatGPT enhanced learners’ time management and self-regulation in language education. Meanwhile, Alghazo et al. (2025) revealed that students across Pakistani universities predominantly used ChatGPT for research purposes but expressed concern over its potential to undermine deep thinking and academic authenticity. Additional research has examined the emotional, attitudinal, and ethical aspects of GenAI adoption. Acosta Enriquez et al. (2024) demonstrated that university students’ behavioral intentions were significantly influenced by both cognitive and affective factors. In Spain, Azcárate (2024) combined survey data with speculative fiction workshops, revealing that while students trusted AI outputs, they emphasized the continuing need for human-centered judgment. Gruenhagen et al. (2024) reported that although Australian students widely adopted GenAI for efficiency and assessment support, many expressed concerns about academic integrity, digital equity, and the reliability of information. Cross-disciplinary comparisons have also revealed that academic background and digital exposure significantly affect students’ perceptions. Dolenc and Brumen (2024) found that computer science students in Slovenia were more optimistic about GenAI’s integration in foreign language education compared to their social science peers. In a large-scale study from Sweden, Stöhr et al. (2024) identified notable differences in ChatGPT familiarity across genders, academic levels, and fields of study, with engineering students showing the highest exposure and health sciences students the least. Regional and infrastructural disparities further shape GenAI adoption. Oyelere and Aruleba (2025), in a study of 322 students from Kenya, Nigeria, and South Africa, emphasized that while learners appreciated AI’s utility in programming education, they were wary of its limitations in promoting social belonging and equitable access. Meniado et al. (2024) found that EFL students in Vietnam and Thailand mainly used ChatGPT for idea generation and grammar correction, though some participants noted limitations in the accuracy and transparency of its feedback during editing. In Spain, Baltà-Salvador et al. (2025) conducted an experimental study with industrial design engineering students, showing that while overall creativity outcomes did not differ significantly between AI-assisted and unaided groups, prior experience with ChatGPT positively influenced students’ evaluations and perceptions of AI-supported ideation. Sun et al. (2024) compared ChatGPT-supported programming with traditional self-guided learning in a Chinese university setting, finding marginal performance improvements and high perceived usability, though no significant gains in overall success. Song et al. (2025) demonstrated that GenAI chatbots fostered more knowledge-based and elaborated dialogue among doctoral students during creative problemsolving tasks, leading to significantly better performance. In Bulgaria, Valova et al. (2024) found that students valued ChatGPT’s clarity for research and writing but often failed to apply critical evaluation. Meanwhile, Veras et al. (2024), using a mixed-methods randomized controlled trial, found that ChatGPT-3.5 scored slightly higher in usability than traditional tools, though qualitative findings highlighted student concerns about misinformation and the need for clearer classroom guidelines. Several studies have explicitly called for more systematic, comparative, and statistically grounded analyses. Chellappa and Luximon (2024), in a survey of Indian design students, noted that ChatGPT’s effect on creativity remains inconclusive and suggested that further methodological refinement is needed. Zhang et al. (2025), working with over 1,200 American students, called for longitudinal and international comparisons to explore how variables such as socioeconomic status and race shape AI perceptions. Wang et al. (2024), in a pre-experimental study of college students’ use of ChatGPT in writing tasks, acknowledged design limitations and called for future research using experimental and advanced statistical methods to better capture evolving student attitudes and performance. Despite these calls, no existing study has aggregated perception data into a standardized success metric, nor applied simulation-based techniques like Monte Carlo modeling to synthesize GenAI impact across educational contexts. To address this critical gap, the present study introduces a novel Monte Carlo simulation framework that consolidates findings from 19 peer-reviewed survey-based studies. Using reverse-coded Likert data, inverse-variance weighting, and standardized thematic indicators, the model estimates composite success scores across three key perception domains: usability, system efficiency, and integration complexity. In doing so, this research offers a replicable, statistically grounded model to evaluate how students perceive GenAI’s role in higher education. The study further contributes a simulation-based success metric that incorporates disciplinary diversity, cultural variation, and methodological rigor, providing both a theoretical advancement and a practical foundation for future policy and curriculum design. 2. Background and Rationale The emergence of generative AI tools such as ChatGPT has prompted widespread academic curiosity, with studies exploring patterns of use, learning performance, ethical use, and emotional responses. Nonetheless, extant research often is fragmented, limited by sample numbers, localized conditions, and descriptive research approaches. One study found that students facing high academic workloads and time pressure were more likely to rely on ChatGPT, which in turn contributed to procrastination and memory-related difficulties, while reward-sensitive students used it less frequently due to fear of detection (Abbas et al., 2024). Rahimi et al. (2025) showed that personalized motivational systems enhanced learners’ self-regulation in ChatGPT-assisted language learning, using a hybrid PLS-SEM and ANN approach. While the study advanced methodological rigor, future research could also consider other techniques (e.g., decision trees or structural causal inference) to further capture complex relationships. Students in Pakistan primarily used ChatGPT for information retrieval rather than for writing or language tasks, with many agreeing that it could reduce deep thinking (Alghazo et al., 2025). A tripartite model of attitudes revealed that cognitive and emotional factors jointly shaped students’ behavioral intentions toward ChatGPT, though the study’s non-probability sampling approach limited generalizability (Acosta Enriquez et al., 2024). In Spain, students viewed ChatGPT as a helpful brainstorming tool but emphasized the importance of contextualizing AI-generated information and maintaining human authorship in final outputs (Azcárate, 2024). In a study of Australian university students, AI-powered chatbots were commonly used in assessments and were not widely perceived as violating academic integrity (Gruenhagen et al., 2024). Disciplinary differences were found between social science and computer science students, with the latter group more supportive of AI-enhanced language learning, though the small sample size and reliance on self-reporting posed methodological constraints (Dolenc & Brumen, 2024). Students in Hong Kong expressed greater familiarity and confidence with ChatGPT, but also raised concerns over transparency and the lack of institutional policy guidance (Chan & Hu, 2023). In Sweden, most students lacked formal instruction on AI use despite widespread familiarity, and engineering students demonstrated the highest engagement levels, particularly at the postgraduate level (Stöhr et al., 2024). Students across Kenya, Nigeria, and South Africa generally perceived ChatGPT as a helpful aid for programming education, but also expressed concerns about cultural bias, inclusion, and social cohesion, particularly as the education level increased (Oyelere & Aruleba, 2025). A mixed-method study on creativity revealed no statistically significant performance difference between AI-assisted and non-assisted groups, although prior experience with ChatGPT was linked to higher-quality outcomes (Baltà-Salvador et al., 2025). Students in Vietnam reported more positive perceptions of ChatGPT in second-language writing than their Thai counterparts, with key challenges emerging around teacher guidance and access to infrastructure (Meniado et al., 2024). Although ChatGPTsupported programming students in China showed more frequent debugging behavior and slightly higher performance, the difference was not significant, and affective attachment remained limited despite functional gains (Sun et al., 2024). Song et al. (2025) found that doctoral students who interacted with a generative AI chatbot (Dou Bao) demonstrated more knowledge-based and elaborated dialogue, reported higher perceived usefulness and intention to use, and achieved significantly better creative problem-solving performance compared to those working with peers. While Bulgarian students widely used ChatGPT for assignments and projects, many lacked critical engagement with AI-generated content, exposing gaps in digital literacy (Valova et al., 2024). Compared to traditional digital tools, ChatGPT was rated somewhat higher in usability by Canadian health sciences students, though focus groups pointed out issues with misinformation and unclear academic integrity boundaries (Veras et al., 2024). Design students in India expressed mixed views on creativity: juniors were impressed by the tool’s novelty, while seniors raised concerns over motivation and ethical risks (Chellappa & Luximon, 2024). Socioeconomic status and gender emerged as important factors in shaping attitudes toward ChatGPT in a U.S. university, with higherSES and female students showing more favorable views, but access inequalities persisted (Zhang et al., 2025). While most students developed a more positive attitude after using ChatGPT to write personal statements, they also noted the tool’s limitations in tone, emotional resonance, and critical depth (Wang et al., 2024). Collectively, these studies highlight fragmented findings that often lack modeling power, representative sampling, or longitudinal insight. To address these limitations, the current study systematically reviews 19 survey-based articles, identifying six with sufficient itemlevel statistics suitable for simulation. A Monte Carlo model was then implemented using one representative dataset (Veras et al., 2024), selected for its thematic alignment and structured Likertscale design. By constructing a multidimensional student success score through inverse-variance weighted statistics across themes such as ease of use, system efficiency, and integration complexity, this work offers a replicable framework to enhance policy relevance and model-based generalizability. Research Question: How can the Monte Carlo simulation of systematic literature review findings be used to quantify student success with Generative AI in higher education? Purpose Statement: This study aims to develop a probabilistic model of student success by synthesizing empirical literature on Generative AI through Monte Carlo simulation, thereby offering a quantitative framework for assessing the educational effectiveness of AI-assisted learning tools. 3. Methodology 3.1. Overview and Rationale The methodology in this study addresses a key gap in the literature: while many studies examine student attitudes toward Generative AI (GenAI), few attempt to model how such perceptions relate to academic outcomes. To bridge this gap, a hybrid design is employed, combining a systematic literature review (SLR) with a Monte Carlo simulation framework. The SLR synthesizes peer-reviewed articles published between 2023 and 2025, focusing on student perceptions of GenAI tools across various domains of higher education. It is important to note that these perceptions are not confined solely to learning processes; instead, they span broader affective, functional, and ethical dimensions such as usability, trust, emotional alignment, and institutional fit. In alignment with PRISMA guidelines, the review was conducted using a single, high-quality database, Scopus. This decision was methodologically strategic: Monte Carlo modeling relies on standardized descriptive statistics (means, standard deviations), which are consistently reported in Scopus-indexed research. This decision was methodologically strategic: the selection focused exclusively on survey-based studies, and only at a later stage were standardized descriptive statistics, such as means and standard deviations considered for determining suitability for Monte Carlo simulation. The second stage applies a Monte Carlo simulation to bridge the perceptual data collected through SLR with probabilistic estimations of academic outcomes. Drawing on item-level descriptive statistics from eligible studies, the model simulates 10,000 synthetic student scores for each thematic domain. This allows for the construction of weighted “success scores” representing likely learning outcomes, based on perception-driven variance. Themes with lower standard error, indicating stronger consensus and greater measurement stability, receive greater weight through inversevariance weighting. This methodological design offers several key advantages. First, by simulating full distributions rather than relying on single mean estimates, the model captures the uncertainty and variability inherent in human perceptions. Second, the simulation framework allows for themelevel generalization across disparate empirical contexts, an advantage that traditional meta-analyses or narrative syntheses cannot fully realize. Third, Monte Carlo simulation has long been recognized as a robust computational tool for decision modeling under uncertainty, particularly in complex systems such as education. Torres et al. (2018) demonstrated the use of Monte Carlo simulation in educational settings to model graduation timelines based on course pass rates and curriculum bottlenecks. Although their model focused on structural efficiency in degree progression, it supports the broader application of probabilistic modeling to simulate academic outcomes under real-world constraints. Several of the reviewed studies explicitly acknowledge methodological shortcomings that limit the generalizability and predictive power of their findings. Wang et al. (2024) highlight the need for more advanced statistical frameworks beyond descriptive analyses. Zhang et al. (2025) call for longitudinal and cross-national studies to capture evolving attitudes toward GenAI across time and cultural contexts. Stöhr et al. (2024) emphasizes the limitations of single-item measures and suggests the development of multidimensional scales to better assess AI perceptions. Dolenc and Brumen (2024) point to small sample sizes and urge future research to adopt broader sampling strategies. Rahimi et al. (2025) propose combining PLS-SEM with artificial neural networks (ANN) to enhance predictive power in modeling self-regulation in ChatGPT-assisted language learning. AcostaEnríquez et al. (2024) recommend probabilistic sampling methods to enhance representativeness. Song et al. (2025) highlight the value of mixed-methods approaches in capturing dialogic dynamics and student perceptions, while Oyelere and Aruleba (2025) emphasize contextual and equity-related factors shaping AI adoption in African higher education. Taken together, these recommendations underscore the current gap in methodological rigor and the absence of scalable modeling frameworks. The present study addresses this gap by integrating a systematic literature review with a Monte Carlo simulation model, thus combining thematic synthesis with probabilistic inference to estimate student success scores based on perception data under uncertainty. This framework builds upon a previously developed simulation model that was designed to link student perception data with learning outcome projections through inverse-variance weighting and synthetic score generation. 3.2. Systematic Literature Review (SLR) This systematic literature review (SLR) compiled empirical evidence on how higher education students perceive, engage with, and respond to generative artificial intelligence (GenAI) tools. The review was conducted in line with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 guidelines (Page et al., 2021), which ensured methodological transparency, consistency, and replicability throughout the search and selection process. The search was conducted using the Scopus database due to its multidisciplinary coverage and high-quality metadata. The initial query was designed to capture a broad range of relevant articles using the following syntax: TITLE-ABS-KEY ("ChatGPT" OR "Generative AI" OR "Generative Artificial Intelligence") AND TITLEABS-KEY ("students" OR "student") AND ("perception" OR "performance" OR "trust" OR "learning" OR "success") This yielded a total of 2,191 full-text, English-language, open-access articles. To focus specifically on empirical research, a secondary, more restrictive query was applied: ABS ("ChatGPT" OR "Generative AI" OR "Generative Artificial Intelligence") AND ABS ("students" OR "university students" OR "higher education") AND ABS ("perception" OR "attitude" OR "academic performance" OR "learning outcome") AND ABS ("survey" OR "experiment" OR "empirical" OR "questionnaire") This step reduced the results to 274 articles based on abstract screening. In the final stage, conceptual relevance was ensured through manual filtering using Scopus’s built-in classification tools, including subject area, keyword tags, and publication type. Studies not directly related to student use of GenAI in higher education were excluded. This refinement yielded 49 core studies. All 49 articles were reviewed in full. Methodological diversity was preserved by including both quantitative and mixed-methods studies, provided they investigated student perceptions of generative AI tools in higher education contexts. Of the 19 studies included, six were identified as reporting sufficiently detailed item-level Likert-scale statistics, means, and standard deviations necessary for quantitative simulation. These six studies were shortlisted based on their alignment with the statistical prerequisites for modeling theme-level usability constructs (Chan & Hu, 2023; Chellappa & Luximon, 2024; Dolenc & Brumen, 2024; Gruenhagen et al., 2024; Oyelere & Aruleba, 2025; Veras et al., 2024). Among them, one representative study, Veras et al. (2024), was selected for detailed simulation modeling. This study employed a 10-item questionnaire with a clear thematic structure and provided the descriptive statistics required for inverse-variance weighting. Importantly, the objective of this systematic review was not to perform a thematic synthesis across all included studies, but rather to identify a subset of articles with sufficient quantitative detail to support a Monte Carlo simulation. While 19 studies met the inclusion criteria for student perceptions of GenAI, only six reported item-level Likert statistics (means and standard deviations) necessary for simulation modeling. As such, the SLR served primarily as a methodological filtering mechanism to extract statistically viable sources for parameterized simulation, rather than as a basis for broad thematic generalization. This approach ensures alignment between the review’s outcome and the simulation framework that follows. Figure 1 PRISMA 2020 flow diagram illustrating article screening and inclusion process based on a systematic review through the Scopus database. 3.3. Representative Study Selection and Simulation Preparation 𝑋theme         =∑𝑤𝑖⋅𝑋𝑖 ∑𝑤𝑖=(1.7778⋅3.71)+(2.2957⋅4.21)+(3.1888⋅4.33)+(1.4516⋅4.00) 8.7139 𝑋theme         ≈4.1169 To ensure an unbiased estimate of theme-level variance (due to the use of unnormalized weights), a Bessel-corrected weighted variance formula was applied: 𝑠weighted 2=∑𝑤𝑖(𝑋𝑖−𝑋)2 (𝑀−1 𝑀)∑𝑤𝑖 where M = 4 is the number of items. The numerator: ∑𝑤𝑖(𝑋𝑖−𝑋)2≈ 0.4739 After calculating all terms and applying the denominator adjustment (3 4)⋅∑𝑤𝑖, The result was: 6.5354 Weighted Standard Deviation (SD): ≈ 0.2707 Summary of Composite Metrics for Theme 1: Weighted Mean: ≈ 4.1169 Weighted Standard Deviation: ≈ 0.2707 These summary statistics were used in the Monte Carlo simulation to generate 10,000 synthetic Theme 1 scores, enabling robust modeling of perceived success under empirical variance conditions. 3.5.2. Theme 2: System Efficiency & Learning Burden Theme 2 focused on evaluating students’ perceptions of the generative AI system’s efficiency and the cognitive effort required to use it. This theme comprised three negatively worded items (Q2, Q4, Q10), each of which was reverse-coded to ensure directional alignment of scoring. Reverse coding was performed using the transformation: X’ = 6 - X This approach ensured that higher scores consistently represented more favorable perceptions, thereby preserving internal consistency when aggregating items into a theme-level metric. For example, although Q2 (“This system was unnecessarily complex”) had a low raw mean of 1.92, suggesting positive perceptions, it was still reverse-coded to align with the positively oriented interpretation of other items. This step is crucial in statistical modeling procedures such as inversevariance weighted means, regression, or simulation, where input variables must be directionally coherent. The purpose of reverse coding in this context is not interpretive, but operational: it ensures that all items contribute comparably to the composite metric. The reverse-coded statistics for each item are presented below: Item 𝑹𝒆𝒗𝒆𝒓𝒔𝒆𝒅𝑴𝒆𝒂𝒏(𝑿𝒊    ) 𝑺𝑫(𝛔𝒊) 𝑽𝒂𝒓𝒊𝒂𝒏𝒄𝒆(𝛔𝒊𝟐) 𝑰𝒏𝒗𝒆𝒓𝒔𝒆 𝑾𝒆𝒊𝒈𝒉𝒕(𝒘𝒊) Q2 4.08 0.58 0.3364 2.9727 Q4 4.25 0.79 0.6241 1.6023 Q10 4.08 0.78 0.6084 1.6437 ∑𝑤𝑖=2.9727+1.6023+1.6437=6.2187 The weighted theme score was calculated as: 𝑋Theme 2           =∑𝑤𝑖𝑋𝑖 ∑𝑤𝑖=25.6647 6.2187 ≈4.1238 To compute the weighted standard deviation, the following Bessel-corrected formula was applied: 𝑠weighted 2=∑𝑤𝑖(𝑋𝑖−𝑋)2 (𝑀−1 𝑀)⋅∑𝑤𝑖 Where M = 3 is the number of items. The weighted variance was calculated as: Numerator: ∑𝑤𝑖(𝑋𝑖−𝑋)2≈0.0344 Denominator:2 3⋅6.2187≈4.1457 𝑠weighted =√0.0344 4.1457≈√0.0083≈0.0911 Thus, the theme-level statistics for Theme 2 were: Weighted Mean: 𝑋 ≈ 4.1238 Weighted Standard Deviation: SD ≈ 0.0911 These values were used as direct parameters in the Monte Carlo simulation, enabling a precisionweighted estimation of user satisfaction for the System Efficiency dimension. 3.5.3. Theme 3: Perceived Complexity & Integration Theme 3 assessed students’ perceptions of the system’s internal coherence, functional integration, and experienced complexity. This theme comprised three items: Q5, Q6, and Q8. Among them, Q6 and Q8 were negatively worded and thus reverse-coded before analysis using the standard transformation: X’ = 6 - X This step ensured directional consistency, so that higher values uniformly represented more favorable student evaluations. Maintaining consistent directional interpretation across all items is essential for constructing reliable composite metrics, particularly when aggregating via inversevariance weights. The processed descriptive statistics for each item (after reverse coding) are presented below: Item 𝑹𝒆𝒗𝒆𝒓𝒔𝒆𝒅 𝑴𝒆𝒂𝒏(𝑿𝒊    ) 𝑺𝑫(𝛔𝒊) 𝑽𝒂𝒓𝒊𝒂𝒏𝒄𝒆(𝛔𝒊𝟐) 𝑰𝒏𝒗𝒆𝒓𝒔𝒆 𝑾𝒆𝒊𝒈𝒉𝒕(𝒘𝒊) Q5 3.80 0.61 0.3721 2.6894 Q6 3.46 0.88 0.7744 1.2913 Q8 3.62 0.82 0.6724 1.4872 ∑𝑤𝑖=2.6894+1.2913+1.4872=5.4659 The weighted mean was calculated as: 𝑋Theme 3           =∑𝑤𝑖𝑋𝑖 ∑𝑤𝑖=10.4291+4.4646+5.3837 5.4659 ≈20.2774 5.4659 ≈3.7100 The weighted standard deviation was estimated using the Bessel-corrected formula: 𝑠weighted 2=∑𝑤𝑖(𝑋𝑖−𝑋)2 (𝑀−1 𝑀)⋅∑𝑤𝑖 With M = 3, the calculation was: Numerator:∑𝑤𝑖(𝑋𝑖−𝑋)2≈0.1704 Denominator:2 3⋅5.4659≈3.6439 𝑠weighted =√0.1704 3.6439≈√0.0468≈0.2163 Thus, the final metrics for Theme 3 were: Weighted Mean: 𝑋 ≈ 3.7100 Weighted Standard Deviation: SD ≈ 0.2163 These parameters were used to model perceived integration and system consistency in the simulation framework, based on empirical variance-weighted scores. 3.6. Simulated Success Score To estimate students’ perceived academic success when interacting with generative AI tools in higher education, a Monte Carlo simulation was implemented in Python. The simulation model relied on empirically validated, theme-level survey data drawn from Likert-scale instruments, where three core usability dimensions, Ease of Use & Learnability, System Efficiency & Learning Burden, and Perceived Complexity & Integration served as predictors for a composite Success Score. Each thematic dimension was weighted using the inverse-variance method, which assigns greater influence to components with lower variability and higher measurement precision. This approach aligns with best practices in fixed-effect meta-analytic modeling, where inverse-variance weighting is used to minimize estimation error (Borenstein et al., 2009). To ensure robust population-level estimates and to reduce the impact of random sampling error, a synthetic sample of 10,000 virtual student respondents was generated. 3.6.1. Simulation Model and Implementation The simulation process was implemented as follows: 1. Initialization: Theme-level empirical means and Bessel-corrected standard deviations from Section 3.5 were used as distributional parameters for each of the three usability dimensions. 2. Random Sampling: For each synthetic student respondent j = 1, 2, …, 10,000, a random score was drawn from a normal distribution defined by the empirical parameters of each theme: 𝑇𝑘∼𝒩(μ𝑘,σ𝑘), for 𝑘=1,2,3 3. Inverse-Variance Weighting: Using the inverse of each theme’s variance 𝑤𝑘=1 σ𝑘 2, a composite success score was calculated for each respondent using the following weighted formula: Success𝑗=𝑤1𝑇1𝑗+𝑤2𝑇2𝑗+𝑤3𝑇3𝑗 𝑤1+𝑤2+𝑤3+ε𝑗 𝑤ℎ𝑒𝑟𝑒 ε𝑗∼𝒩(0,0.05) adds small, normally distributed noise to account for unobserved factors such as attention, effort, or prior familiarity. 4. Boundary Clipping: The resulting success scores were bounded within the original 1–5 Likert scale using truncation (or clipping), ensuring simulation fidelity to the survey response structure. This simulation enabled the construction of a statistically principled, perception-based estimate of generative AI’s impact on student success. It also provided a scalable methodological framework for evaluating educational technology interventions using summary-level data. The simulation was implemented in Python using NumPy-based random sampling and inverse-variance weighting. Figure 2 shows the Monte Carlo simulation in Python. 3.6.2. Descriptive Statistics of Simulated Success Score The Monte Carlo simulation yielded a synthetic dataset of 10,000 student responses, each representing a composite Success Score calculated as a weighted average of three usability themes: Ease of Use & Learnability, System Efficiency & Learning Burden, and Perceived Complexity & Integration. The weighting was based on the inverse of the Bessel-corrected variance associated with each theme, ensuring that more precisely measured dimensions contributed more strongly to the final score (Borenstein et al., 2009). Random variation (ε𝑗∼𝒩(0,0.05)) was added to each simulated respondent’s score to represent unmeasured influences such as attention, motivation, or cognitive variability. After computation, all scores were clipped between 1 and 5 to ensure consistency with the original Likert scale. The descriptive statistics of the simulated data are summarized in the following table: Statistic Value Interpretation Count 10,000 Number of simulated student responses Mean 4.0666 High perceived success across the sample Standard Deviation 0.0956 Low variability; strong consensus among respondents Minimum 3.7294 Lowest simulated score remained moderately favorable 25th Percentile 4.0032 Lower quartile boundary Median (50%) 4.0667 Central tendency closely aligned with the mean 75th Percentile 4.1308 Upper quartile boundary Maximum 4.4172 Highest observed success score 3.6.2.1. Figure Inclusion and Caption Insert this just below the descriptive statistics section: The histogram displays the distribution of the simulated Success Scores, showing a near-normal shape centered around a mean of approximately 4.07. The tight clustering of values between 3.9 and 4.2 indicates that most synthetic students perceived generative AI tools as beneficial, with minimal dispersion in the sample. Figure 3 shows a normal distribution of Success Scores, reflecting consistent perceptions. As shown in Figure Success Score, the distribution of simulated Success Scores is approximately normal and highly concentrated around the mean. This reinforces the consistency of positive student perceptions observed in the simulation. 3.6.3. Inter-Theme Comparison To further explore the differential impact of the three usability dimensions on perceived academic success, a multiple linear regression model was employed. This model estimates the relative contribution of each theme to the simulated Success Score. 3.6.3.1. Linear Regression Model The general form of the multiple linear regression model is as follows: 𝑌𝑗=β0+β1𝑋1𝑗+β2𝑋2𝑗+β3𝑋3𝑗+ε𝑗 Where: 𝑌𝑗 : Simulated Success Score for respondent j (dependent variable) 𝑋1: Theme 1 — Ease of Use & Learnability 𝑋2: Theme 2 — System Efficiency & Learning Burden 𝑋3: Theme 3 — Perceived Complexity & Integration 𝛽0: Intercept (constant term) 𝜀𝑗: Error term capturing unexplained variation Estimated Regression Equation: Success𝑗=0.2885+0.0856⋅𝑋1𝑗+0.7823⋅𝑋2𝑗+0.1381⋅𝑋3𝑗+ε𝑗 This equation demonstrates that Theme 2 (System Efficiency & Learning Burden) has the largest coefficient, suggesting it is the most influential predictor of perceived academic success in the simulated data. Themes 1 and 3 also have positive coefficients, indicating meaningful, though comparatively smaller contributions. 3.6.3.2. Coefficient Interpretation Coefficient (β) Interpretation Intercept (0.2885) Represents the expected Success Score when all three theme scores are zero. Not statistically significant (p = 0.328) and has no practical interpretive value. Theme 1 (0.0856) A one-unit increase in Ease of Use & Learnability is associated with a 0.0856-point increase in Success Score on average (p < 0.001). Theme 2 (0.7823) A one-unit increase in System Efficiency & Learning Burden results in a 0.7823point increase in Success Score, the strongest predictor (p < 0.001). Theme 3 (0.1381) A one-unit increase in Perceived Complexity & Integration yields a 0.1381-point increase (p < 0.001). 3.6.3.3. Model Evaluation The multiple linear regression model was evaluated using standard goodness-of-fit diagnostics and residual analysis: R² = 0.724: The model explains 72.4% of the total variance in the simulated Success Score, indicating a strong fit. F(3, 9996) = 8741, p < .001: The overall model is statistically significant, confirming that at least one of the usability themes contributes meaningfully to the prediction of success. 3.6.3.4. Final Regression Equation To evaluate the contribution of each usability theme to students’ perceived success, a multiple linear regression was conducted. The Success Score was modeled as a function of the three core usability themes using the following regression formula: Success𝑗=β0+β1(Theme1)𝑗+β2(Theme2)𝑗+β3(Theme3)𝑗+ε𝑗 Where: Success𝑗: Simulated Success Score for student j 𝛽0: Intercept β1,β2,β3: Coefficients for each theme 𝜀𝑗: Error term capturing individual variation The regression was implemented in Python using the statsmodels package and applied to the full set of 10,000 synthetic student records. The summary output of the regression analysis is presented below. Figure 4 summarizes the regression predicting Success Scores from usability themes. picked as a representative dataset for the purposes of demonstrating the possibility of using Monte Carlo simulation for modeling student perceptions. This was due to its thematically structured survey questionnaire articulated around a balanced Likert-scale design and direct relevance of usability elements with respect to digital usability within higher education. Thematic coding was used to group the ten pre-existing survey items into three usability constructs defined by the researchers: Ease of Use & Learnability, System Efficiency & Learning Burden, and Perceived Complexity & Integration. These formed the basis for synthesizing artificial Success Scores from 10,000 simulated student profiles sampled from inverse-variance-weighted normal distributions. Simulation used fixed-effect assumptions congruent with typical meta-analytic methodology (Borenstein et al., 2009) and included inverse-variance-weighted non-independence adjustments. Regression analysis revealed System Efficiency & Learning Burden as a prime predictor variable for perceived success (β = 0.7823, p < .001), followed by Perceived Complexity & Integration (β = 0.1381, p < .001) and Ease of Use & Learnability (β = 0.0856, p < .001). The global model showed good explanatory power (R² = 0.724), signifying these three dimensions represent a large portion of variance attributable to how students assess usefulness upon deployment of GenAI tools. The fundamental methodological contribution through this work is how simulation techniques and inverse-weighted variants can be utilized for modeling education's subjective constructs without direct access to any dataset. This presents an improvement on conventional synthesis methods through facilitating composite scoring and solution testing while imposing stringent statistical assumptions. The Success Score approach is very portable and is capable of being adapted to institutional settings where data privacy, restricted access to microdata, or lack of resources would inhibit the collection of direct empirics. It is therefore able to benefit educational leaders and instructional designers in benchmarking or pilot-testing GenAI tools in consideration of both pedagogical intentions and institutional circumstances. Other than innovating methodologically, this work also has real-world applicability for AI-informed education design. The Success Score delineated herein allows institutions to measure usability in conjunction with decreased cognitive workload and compatibility with workflows. Simulation outputs reflect a deeper shift in student expectations: efficiency-based, performance-driven design is potentially supplanting surface-level usability in defining perceived success, a signpost of increasing maturity in the student-AI alliance. Whilst any such model has its flaws. With just a single representative study (Veras et al., 2024) providing clarity for documenting method but also limiting thematic and statistical diversity, future iterations must contain multiple representative datasets per theme for improved generalizability. Adding variables such as academic discipline, pre-existing GenAI experience, or task type would also enhance contextual awareness. Higher qualitative levels, such as interviews or open-ended survey responses, could also provide depth to the scope for interpretation. Longitudinal work would also benefit assessment of whether factors for GenAI success are temporally stable as these technologies further develop. Finally, this work offers a replicable and theory-guided simulation template spanning qualitative understanding and quantitative modelling. The proposed Success Score provides not just a diagnostic tool for unmasking student experiences with GenAI within postsecondary education but also constitutes a strategic tool for institutions on policy and design. With the increasing presence of GenAI within academic experience, Simulationbased models such as this will play a key role in informing inclusive, effective, and evidence-based education futures. Funding This study did not receive external funding. Competing interests The author declares no conflicts of interest. Ethical approval Not applicable. Data availability This study uses only published literature; no new datasets were generated or analysed. REFERENCES Abbas, M., Jam, F. A., & Khan, T. I. (2024). Is it harmful or helpful? Examining the causes and consequences of generative AI usage among university students. International Journal of Educational Technology in Higher Education, 21(1), 10. https://doi.org/10.1186/s41239-024-00444-7 Acosta Enriquez, B., Arbulú Ballesteros, M., Huamani, O., Roca, C., & Tirado, K. (2024). Analysis of college students' attitudes toward the use of ChatGPT in their academic activities: effect of intent to use, verification of information and responsible use. BMC Psychology, 12, Article 255. https://doi.org/10.1186/s40359-024-01764-z Alghazo, R., Fatima, G., Malik, M., Abdelhamid, S. E., Jahanzaib, M., Nayab, D. e., & Raza, A. (2025). Exploring ChatGPT’s Role in Higher Education: Perspectives from Pakistani University Students on Academic Integrity and Ethical Challenges. Education Sciences, 15(2), 158. https://www.mdpi.com/2227-7102/15/2/158 Azcárate, A. L.-V. (2024). Foresight Methodologies in Responsible GenAI Education: Insights from the Intermedia-Lab at Complutense University Madrid. Education Sciences, 14(8), 834. https://www.mdpi.com/2227-7102/14/8/834 Baltà-Salvador, R., El Madafri, I., Brasó-Vives, E., & Peña, M. (2025). Empowering Engineering Students Through Artificial Intelligence (AI): Blended Human–AI Creative Ideation Processes With ChatGPT. Computer Applications in Engineering Education, 33, e22817. https://doi.org/10.1002/cae.22817 Bangor, A., Kortum, P. T., & Miller, J. T. (2008). An Empirical Evaluation of the System Usability Scale. International Journal of Human–Computer Interaction, 24(6), 574-594. https://doi.org/10.1080/10447310802205776 Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R. (2009). Introduction to metaanalysis. John Wiley & Sons. https://doi.org/10.1002/9780470743386 Brooke, J. (1996). A quick and dirty usability scale. In P. W. Jordan, B. Thomas, B. A. Weerdmeester, & I. L. McClelland (Eds.), Usability evaluation in industry (pp. 189– 194). London: Taylor & Francis. Chan, C. K. Y., & 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), 43. https://doi.org/10.1186/s41239-023-00411-8 Chellappa, V., & Luximon, Y. (2024). Understanding the perception of design students towards ChatGPT. Computers and Education: Artificial Intelligence, 7, 100281. https://doi.org/10.1016/j.caeai.2024.100281 Dolenc, K., & Brumen, M. (2024). Exploring social and computer science students’ perceptions of AI integration in (foreign) language instruction. Computers and Education: Artificial Intelligence, 7. https://doi.org/10.1016/j.caeai.2024.100285 Gruenhagen, J. H., Sinclair, P. M., Carroll, J.-A., Baker, P. R. A., Wilson, A., & Demant, D. (2024). The rapid rise of generative AI and its implications for academic integrity: Students’ perceptions and use of chatbots for assistance with assessments. Computers and Education: Artificial Intelligence, 7, 100273. https://doi.org/10.1016/j.caeai.2024.100273 Meniado, J. C., Huyen, D. T. T., Panyadilokpong, N., & Lertkomolwit, P. (2024). Using ChatGPT for second language writing: Experiences and perceptions of EFL learners in Thailand and Vietnam. Computers and Education: Artificial Intelligence, 7, 100313. https://doi.org/10.1016/j.caeai.2024.100313 Oyelere, S. S., & Aruleba, K. (2025). A comparative study of student perceptions on generative AI in programming education across Sub-Saharan Africa. Computers and Education Open, 8, 100245. https://doi.org/10.1016/j.caeo.2025.100245 Page, M., McKenzie, J., Bossuyt, P., Boutron, I., Hoffmann, T., Mulrow, C., Shamseer, L., Tetzlaff, J., Akl, E., Brennan, S., Chou, R., Glanville, J., Grimshaw, J., Hróbjartsson, A., Lalu, M., Li, T., Loder, E., Mayo-Wilson, E., McDonald, S., & Moher, D. (2021). The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. BMJ, 372, n71. https://doi.org/10.1136/bmj.n71 Rahimi, A. R., Sheyhkholeslami, M., & Mahmoudi Pour, A. (2025). Uncovering personalized L2 motivation and self-regulation in ChatGPT-assisted language learning: A hybrid PLS-SEM-ANN approach. Computers in Human Behavior Reports, 17, 100539. https://doi.org/10.1016/j.chbr.2024.100539 Song, Y., Huang, L., Zheng, L., Fan, M., & Liu, Z. (2025). Interactions with generative AI chatbots: unveiling dialogic dynamics, students’ perceptions, and practical competencies in creative problem‑solving. International Journal of Educational Technology in Higher Education, 22(1), 12. https://doi.org/10.1186/s41239-02500508-2 Stöhr, C., Ou, A. W., & Malmström, H. (2024). Perceptions and usage of AI chatbots among students in higher education across genders, academic levels and fields of study. Computers and Education: Artificial Intelligence, 7, 100259. https://doi.org/10.1016/j.caeai.2024.100259 Sun, D., Boudouaia, A., Zhu, C., & Li, Y. (2024). Would ChatGPT-facilitated programming mode impact college students’ programming behaviors, performances, and perceptions? An empirical study. International Journal of Educational Technology in Higher Education, 21(1), 14. https://doi.org/10.1186/s41239-024-00446-5 Torres, D., Crichigno, J., & Sanchez, C. (2018). Assessing curriculum efficiency through Monte Carlo simulation. Journal of College Student Retention: Research, Theory & Practice, 22(4), 597–610. https://doi.org/10.1177/1521025118776618 Valova, I., Mladenova, T., & Kanev, G. (2024). Students' Perception of ChatGPT Usage in Education. International Journal of Advanced Computer Science and Applications 15(1), 466–473. https://doi.org/10.14569/IJACSA.2024.0150143 Veras, M., Dyer, J.-O., Shannon, H., Bogie, B., Ronney, M., Sekhon, H., Rutherford, D., Silva, P., & Kairy, D. (2024). A mixed methods crossover randomized controlled trial exploring the experiences, perceptions, and usability of artificial intelligence (ChatGPT) in health sciences education. DIGITAL HEALTH, 10. https://doi.org/10.1177/20552076241298485 Wang, C., Aguilar, S. J., Bankard, J. S., Bui, E., & Nye, B. (2024). Writing with AI: What College Students Learned from Utilizing ChatGPT for a Writing Assignment. Education Sciences, 14(9), 976. https://www.mdpi.com/2227-7102/14/9/976 Zhang, C., Wang, L. H., & Rice, R. E. (2025). U.S. college students’ acceptability and educational benefits of ChatGPT from a digital divide perspective. Computers and Education: Artificial Intelligence, 8, 100385. https://doi.org/10.1016/j.caeai.2025.100385