The Impact of Generative AI on Academic Performance, Skill Retention & Student Well-Being: A Large-Scale Educational Analytics Study

By abhishek.verma75000 · May 20, 2026

Datastam conducted a comprehensive educational analytics study examining the impact of Generative AI usage on academic performance, skill retention,…

The analysis produced a heatmap and a bar chart showing which combinations of Prompt Engineering Skill level and Primary Use Case yield the highest average Skill Retention Scores. The heatmap gives a full matrix view across all skill levels and use cases, while the bar chart highlights the top 10 highest-performing combinations at a glance.

STEM is the standout major category, leading in both metrics. It has the largest average GPA improvement of 0.217 (from 3.15 pre-semester to 3.36 post-semester) and the highest average Skill Retention Score of 76.80. Three bar charts and a scatter plot were generated to visualize these comparisons across all five major categories.

The optimal ratio of Traditional Study Hours to Weekly GenAI Hours is between 2 and 4 (meaning students study traditionally about 2–4 times more than they use GenAI tools). This ratio range yields the highest average Post-Semester GPA of 3.376, based on data from 9,316 students. The median ratio within this top-performing group is 2.77. A bar chart has been generated showing average GPA across all ratio bins, making it easy to see how performance peaks in the 2–4 range.

The analysis shows that Weekly GenAI Hours has very little impact on GPA change from Pre-Semester to Post-Semester GPA. The correlation is -0.0465, indicating an extremely weak negative relationship — meaning more GenAI hours are very slightly associated with smaller GPA gains, but the effect is negligible. Two charts and supporting data tables were generated to visualize this relationship.

Institutional Policy type has a modest but measurable influence on student outcomes. Two charts and supporting data tables were generated to show how Post-Semester GPA, Skill Retention Score, and Burnout Risk Level vary across policy types. The 'Actively Encouraged' policy is associated with the best outcomes across all three metrics, while 'Strict Ban' consistently ranks lowest.

Both Perceived AI Dependency and Anxiety Level During Exams are strongly linked to Burnout Risk Level. Students with the lowest AI dependency have only a 9.9% rate of high burnout, while those with the highest AI dependency face a striking 94.7% high burnout rate. Similarly, low exam anxiety is associated with 17.7% high burnout, compared to 70.2% for those with the highest anxiety. Three visualizations were generated: a heatmap showing the percentage of high-burnout students across every combination of AI dependency and anxiety level, plus stacked bar charts breaking down burnout risk by each factor individually.

The analysis successfully produced two visualizations and supporting data tables comparing AI tool usage and paid subscription rates across different major categories. A grouped bar chart shows the average Tool Diversity (with standard deviation error bars) for each major, and a stacked bar chart breaks down the proportion of students with paid vs. free subscriptions by major.

The analysis successfully produced box plots and data tables highlighting outliers in both Weekly GenAI Hours and Traditional Study Hours using the IQR method. Two side-by-side box plots were generated, each visually highlighting outlier points beyond the whiskers (Q1 - 1.5×IQR and Q3 + 1.5×IQR bounds). The charts clearly show the distribution of study hours and flag students whose values fall outside the expected range. Additional data tables provide further detail on the outlier groups and their associated GPA and Skill Retention Scores.

A full correlation matrix heatmap has been generated for all 8 numeric variables in the dataset: Pre Semester GPA, Weekly GenAI Hours, Tool Diversity, Traditional Study Hours, Perceived AI Dependency, Anxiety Level During Exams, Post Semester GPA, and Skill Retention Score. The heatmap uses a Red-Blue color scale with annotated correlation coefficients, making it easy to spot strong positive (blue) and negative (red) relationships at a glance.

The analysis successfully compared academic performance trajectories across all four year groups (Freshman, Sophomore, Junior, Senior). Two visualizations were generated: a horizontal bar chart showing average GPA delta per year group ranked from highest to lowest improvement, and a bar chart displaying average weekly GenAI usage hours by year group. Data tables were also produced summarizing the key metrics for each cohort.