This investigation draws on survey responses collected from university students on their use of generative AI platforms — ChatGPT, Gemini, Claude, Copilot, and DeepSeek. Students reported their weekly AI usage hours, the primary task they used AI for (study, writing, coding, or research), a satisfaction rating from 1 to 5, and a self-assessed academic performance score out of 100.
The dataset was sourced from Kaggle and contained 230 raw entries. After removing non-numeric hour values, performance descriptors that could not be converted to a numeric score, and unrecognised tool or purpose labels, 196 usable records were retained. All cleaning, transformation, and visualisation was performed in R.
I didn’t realise how deeply generative AI had become part of my learning until I tried completing an assignment without it. What started as a quick tool for explanations and coding help slowly turned into something more constant — a writing partner, a debugging assistant, and sometimes even a substitute for thinking through problems independently.
This experience is no longer unusual. Across university campuses, generative AI tools like ChatGPT, Gemini, Claude, Copilot, and DeepSeek have become embedded in everyday study practices.
But this shift raises an important question: are these tools genuinely improving how students learn, or are they quietly changing what it means to learn in the first place?
This story uses student survey data to explore how generative AI is being used, what purposes it serves, and whether heavier use actually translates into better academic performance. The findings challenge a common assumption — that more AI usage automatically leads to better outcomes — and instead reveal a more complex relationship between convenience, learning, and dependency.
Students are not waiting for institutional guidance on which platform to use. They are making their own choices, and the market is far more fragmented than headlines about ChatGPT suggest.
ChatGPT accounts for approximately 30.6% of all reported AI tool use — a commanding lead, but far from a monopoly. Gemini, Claude, Copilot, and DeepSeek all hold meaningful shares. Rather than converging on one platform, students appear to be assembling personal toolkits. This distributed adoption reflects just how deeply generative AI has embedded itself into everyday university life.
Tool popularity alone does not explain how students actually use generative AI in their academic work. To understand this better, it is important to look at how different tools are associated with different types of tasks.
ChatGPT is the dominant choice for study-oriented tasks, while Gemini shows a strong presence among writing users. Claude and Copilot appear more often in coding contexts, and DeepSeek draws a notable cluster of research-focused students. These patterns are not random — they reflect informal reputations built through peer recommendations and direct experience. Students, it seems, have quietly become informed AI consumers.
The assumption that heavier AI use leads to better academic results is intuitive. The data suggests it deserves scrutiny.
However, usage patterns and preferences do not necessarily translate into academic improvement. The next step is to examine whether increased AI usage is actually associated with higher performance.
A formal correlation test confirms what the chart shows visually: the relationship between weekly AI usage hours and academic performance is weak (r = 0.17, p = 0.026). High-scoring students appear across the full range of usage levels — including those using AI for just a couple of hours a week. Frequency of use is not a reliable lever for academic improvement. What appears to matter more is how students engage with these tools, not how often.
Not all uses of AI feel equally worthwhile to students. When satisfaction is broken down by task type, clear differences emerge — along with a tension worth examining.
Beyond academic performance, student satisfaction provides another perspective on how effective these tools feel in practice.
Writing-related AI use generates the highest average satisfaction of any task category. Students clearly value the assistance AI provides with drafting and editing. Yet writing users also invest the most hours per week. This pairing raises a pointed question: if AI is handling an increasing share of the compositional thinking — generating structure, proposing arguments, smoothing transitions — are students still developing the writing skills that come from genuinely struggling with a blank page? High satisfaction can, in some cases, be a signal of convenience rather than genuine learning.
Finally, combining satisfaction, performance, and time investment allows us to evaluate which tools may offer the most efficient learning experience.
Viewing satisfaction, performance, and time investment together reveals that no single tool wins on every measure. Some platforms combine solid satisfaction and performance scores while requiring comparatively fewer hours — indicating a more efficient learning relationship with the tool. Others demand a higher weekly time commitment but return only modest gains. For students who are deliberate about how they manage their study time, tool selection is a genuine academic decision, not just a matter of personal preference.
Generative AI has clearly become a standard part of university learning, but its role is more nuanced than simply improving academic outcomes. Students are not just using AI more — they are using it differently, developing distinct preferences across tools and tasks.
However, the most important finding is not how much AI students use, but how weakly usage is linked to academic performance. The data suggests that higher engagement alone does not guarantee better results. Instead, the way students interact with these tools appears to matter far more than frequency of use.
This creates a challenge for universities. Restricting AI use is unlikely to be effective. Instead, the focus may need to shift toward designing learning environments where AI supports critical thinking rather than replacing it.
Ultimately, generative AI is not simply a tool students use — it is becoming part of how they think, learn, and produce knowledge. The question now is not whether it should be used, but how it can be used in ways that strengthen, rather than weaken, learning.
Baloch, Z. (2026). Student AI tool usage dataset [Data set]. Kaggle. https://www.kaggle.com/datasets/zohairbaloch/student-ai-tool-usage-dataset
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), 1–18. https://doi.org/10.1186/s41239-023-00411-8
Holmes, W., Bialik, M., & Fadel, C. (2019). Artificial intelligence in education: Promises and implications for teaching and learning. Center for Curriculum Redesign.
Lodge, J. M., Thompson, K., & Corrin, L. (2023). Mapping out a research agenda for generative artificial intelligence in tertiary education. Journal of Applied Learning and Teaching, 6(1), 1–6. https://doi.org/10.37074/jalt.2023.6.1.34
OECD. (2023). Generative artificial intelligence in education: Chatbots and beyond. Organisation for Economic Co-operation and Development. https://www.oecd.org/education/
Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002
TEQSA. (2023). Guidance note: Artificial intelligence. Tertiary Education Quality and Standards Agency. https://www.teqsa.gov.au/guides-resources/guidance-note-artificial-intelligence
UNESCO. (2023). Guidance for generative AI in education and research. United Nations Educational, Scientific and Cultural Organization. https://doi.org/10.54675/PCNF9592
Yan, L., Sha, L., Zhao, L., Li, Y., Martinez-Maldonado, R., Chen, G., Li, X., Jin, Y., & Gašević, D. (2024). Practical and ethical challenges of large language models in education: A systematic scoping review. British Journal of Educational Technology, 55(1), 90–112. https://doi.org/10.1111/bjet.13370
Generative AI tools (ChatGPT and Claude) were used in this assignment to support brainstorming, improve clarity of writing, and assist with debugging R code. All data cleaning, analytical decisions, interpretation of results, and final visualisation design were independently developed and critically reviewed by own. The final narrative, insights, and conclusions reflect the my own understanding and interpretation of the dataset.