Behind the Screens: A Data Driven Analysis of How Generative AI Shapes Student Performance, Anxiety and Skill Retention

Technology is one of the most important factors in how students learn and develop today. It changes how we study, how fast we complete assignments, and whether we are actually thinking critically and understanding concepts or just looking for shortcuts. High schools and colleges are filled with thousands of students spread across different majors with completely different workloads, technical requirements, and testing styles, and they don't all experience technology the same way. While students everywhere have started using Generative AI (GenAI) tools like ChatGPT,Claude, Google Gemini, this trend hasn't helped everyone equally. Some academic majors have figured out how to use these tools to jump ahead, while others are actually experiencing a drop in how much they remember. At the same time, the gaps between using AI actively to solve problems versus using it passively to copy answers, and between knowing how to prompt versus just typing basic questions, are growing wider.

This report looks at a dataset of 50,000 students to answer a big question: how has the impact of Generative AI evolved across different academic majors, student usage habits, and school rules? Using a series of visualizations made Tableau, this report explores major level grade patterns, usage differences, and the impact of school policies before drawing conclusions about where AI assisted education is heading.

A useful starting point is to compare student performance across different academic majors before and after they started using AI heavily during the semester, as shown in Figure 1 and Figure 2 below.

In the pre semester baseline, the average grades across all five major categories were almost exactly the same, ranging from a low of 3.141 to a high of 3.151, as illustrated in Figure 1. Students in tracks like Medical and Humanities started at practically the same point as students in Business, Arts, and STEM, which shows that everyone started the term on a level playing field.

Figure 1: Average Pre-Semester GPA Across Academic Majors

By the end of the semester, the picture had shifted noticeably, as shown in Figure 2 below. The average GPA for all categories went up, with the lowest performing major climbing to 3.336 and the highest surging to 3.363.

Figure 2: Average Post Semester GPA Across Academic Majors

This indicates that across the board, students successfully used these tools to get meaningful GPA gains over the term. Majors that require a lot of technical problem solving, like STEM and Medical, showed some of the most significant improvements, climbing to final averages of 3.363 and 3.353. Overall, this general upward trend shows that technology naturally helps raise baseline grades, even though small gaps between technical and non technical fields start to show up by the end of the year.

How many hours a week a student spends using AI is one of the clearest dividing lines in learning outcomes. Figure 3 below compares final grades, exam anxiety, and skill retention across low, moderate, and heavy weekly usage tiers.

Figure 3: Weekly GenAI Hours vs Student Outcomes

As shown in Figure 3, a consistent pattern emerges: moderate AI use gives students the best balance, while over using it comes with a major penalty. In the low usage bracket (0 to 5 hours a week), student exam anxiety was quite low at 3.85 out of 10, and their skill retention stayed strong at 75.98. In the moderate tier (5 to 15 hours a week), students hit a "sweet spot" -> their post-semester GPA peaked at 3.37 and their skill retention reached its highest point at 76.99. However, once students crossed into the heavy usage tier (over 15 hours a week), their metrics got much worse: average exam anxiety spiked to 5.33, and their skill retention dropped to 72.68. This suggests that while using AI in moderation makes a great study buddy, relying on it too much creates a bad dependency. Students end up panicking and getting high anxiety when they have to take closed book tests without their technological tools.

Looking at this problem through school rules and technical skills gives us even more insight. Figure 4 and Figure 5 below show how school policies and a student's prompting skills interact with how much AI they use. Red shades represent high stress outcomes and blue shades represent stable skill retention.

Figure 4: Institutional Policies vs Exam Anxiety for Heavy Users

In schools with highly restrictive rules, the anxiety penalty for heavy users is massive, as seen in Figure 4. Under a "Strict Ban," heavy AI users experienced a huge exam anxiety peak of 7.76. On the other hand, heavy users in schools where AI was "Actively Encouraged" maintained a much lower anxiety average of 4.79.

Figure 5: Prompt Engineering Skill vs Skill Retention

By contrast, as shown in Figure 5, having advanced prompt engineering skills acts as a safety shield against learning loss. Advanced prompt users kept high skill retention scores across the board, peaking at 83.36 under moderate use and only dropping to 78.95 under heavy use. Beginner prompt users, though, saw their retention fall apart, crashing to a low of 67.70 when they used AI heavily. This major difference highlights that AI risks aren't the same for every student; they depend heavily on school rules and how good the student is at writing prompts.

Beyond just weekly hours, the split between how students use AI and their technical skill level gives us another great lens for looking at the data. Figure 6 below breaks down skill retention scores by both the student's primary use case and their prompt engineering level.

Figure 6: Primary Use Case vs Prompt Engineering Skill Matrix

As shown in Figure 6, the data reveals a clear and persistent hierarchy: active study methods always beat passive shortcuts, and advanced prompt skills always beat beginner habits within every category. In active troubleshooting and debugging usecase, beginner prompt users scored 70.95 in skill retention, while advanced users achieved the highest mark in the matrix at 88.42. In passive applications like direct answer generation, the numbers were way lower across the board, with beginners dropping to the absolute bottom of the matrix at 69.78 and advanced users capping out at 78.35. Interestingly, advanced users doing basic copywriting (79.95) scored almost the same in retention as intermediate users doing complex coding work (75.86). This shows how much of an advantage prompt literacy gives you, no matter the task. The mental effort required to write structured prompts gives advanced students a huge learning advantage over passive users.

The data examined in this project paints a picture of steady but uneven progress in digital learning. At the macro level, student GPAs generally improved, and performance across different majors went up, showing that access to technology helps. However, when you break the data down by individual study habits and school rules, deeper inequalities show up. Students who use AI for passive answer shortcuts continue to lag far behind active debuggers in remembering what they learned, and students stuck under strict school bans suffer from much higher exam anxiety. While advanced prompt skills clearly protect students from learning loss, basic copy paste habits leave them at a disadvantage. Taken together, these findings suggest that future school policies should move away from broad, ineffective bans. Instead, they should focus on teaching advanced prompt literacy and active problem solving, making sure technology acts as a learning tool rather than an academic crutch.