Finding Users Next Door-AI Adoption among Graduate Students
A primary survey-based study into how analytics graduate students are using AI and what it means for program administrators, educators and product builders.
Recently being a teaching assistant for a 45-student MBA class at USC Marshall, I found myself in a unique position: the intersection of academic policy and student reality. While faculty meetings focused on AI detection tools and academic integrity policies, I watched … nay, lived student life navigating this new landscape in real-time.
This perspective presented few questions such as what are the sub-use cases under classroom, work where AI was used, how do we actually measure and develop AI preparedness? More importantly, what can we learn in-real time from the students who are living this transformation daily?
After four months of survey design, literature review, and data analysis across my *very supportive* 200-people cohort, the findings challenge several assumptions about AI adoption in higher education and reveal interesting patterns that help educators and LLM product teams.
The Adoption is Wide and Democratized
The most striking finding from regression analysis was what didn’t matter: traditional markers of professional readiness such as years of work experience, or industry background/ future interest showed no significant correlation with AI preparedness scores.
Unlike previous technology adoption curves that favored early-career professionals or recent graduates, AI tools appear to have democratized access to advanced capabilities. A mid-career executive pivoting into tech showed similar preparedness potential as a fresh undergraduate, provided they engaged meaningfully with the tools.
From curriculum development standpoint, rather than assuming certain student populations need AI support, programs should focus on creating pathways that meet learners wherever they are as preparedness is being shaped by engagement, not background.
Multi-platform Reality
ChatGPT predictably dominates usage patterns. Perplexity emerged with a small but rather engaged user base with ~ 25% of student share. Users show meaningful integration across productivity and academic contexts.
More telling was the pattern of multi-LLM usage. High-preparedness students weren’t just using one tool more frequently; they are multi-homing platforms, perhaps strategically? One student described using Claude for analytical writing, ChatGPT for brainstorming, and Perplexity for research synthesis, each tool serving distinct cognitive functions within larger workflows.
This indicates preparedness isn’t about mastering a single platform but developing platform agnosticism and the ability to match tool capabilities to specific task requirements.
Popular Sub Use Cases
Students reported using AI across an average of 3–5 distinct categories,
General Applications: Personal usage across trip planning, exercise, and diet recommendations
Career Development: Resume optimization and interview preparation were top uses while career planning and job search were less common use cases.
Academic Integration: Despite policy uncertainties, students reported significant usage for assignment drafting and learning support within the allowed limit.
Breadth over Depth Drives Confidence
When we analyzed what impacts AI preparedness (self-rated), two factors stood out with statistical significance (95% level):
- Number of different LLMs used (correlation coefficient: 0.68)
- Breadth of use cases across life contexts (correlation coefficient: 0.71)
The chart shows high preparedness (self-rated) lies in the top right corner, sweet spot of using different llms across use cases. In fact, variety of LLMs tried is significantly (p<0.05) correlated to preparedness level.
This learning curve is a contrast to the SaaS skilling curve where preparedness comes from deep expertise with a single tool. Instead, students who demonstrated highest preparedness were digital polyglots, comfortably switching between platforms and applying AI thinking across multiple domains of their lives.
The implication is counterintuitive: rather than teaching students to become power users of ChatGPT, we should learn more about connecting AI systems and different tools to become thoughtful consumers of the broader ecosystem.
Get Aakanksha Sangwan’s stories in your inbox
Join Medium for free to get updates from this writer.
Tiered Learning Spectrum
Perhaps the most actionable question and directly relevant insights emerged from analyzing what and how do they want to learn? Workshops were the highest chosen option, followed by courses. The areas of interest are as follows, presented by preference across self-rated preparedness levels,
Beginners- students showed strong interest in foundational concepts:
- Prompt engineering fundamentals
- Understanding AI’s strategic business impact
- Basic workflow integration
Advanced-preparedness students gravitated toward advanced applications:
- Specialized tools (Tableau GPT, LangChain integration)
- Productivity and Lowcode/Nocode development
Interestingly AI agents was high across both the groups. This natural tiering suggests educational programs should pivot from one-size-fits-all approaches. Instead, institutions should develop parallel tracks that allow foundational learners to build confidence while providing advanced learners with the sophisticated challenges that are needed for growth.
Workplace Discussions Matter
The survey revealed an interesting effect: students who regularly discussed AI applications in workplace settings showed significantly higher preparedness scores (p = 0.05) compared to those whose AI conversations were limited to academic contexts.
Workplace AI discussions appeared to provide something classroom conversations miss, a real-world application context, immediate feedback loops, and exposure to diverse use cases across different industries and functions.
Interestingly, while formal academic AI discussions showed marginal influence (p = 0.06), the effect wasn’t statistically significant. This suggests that academic institutions may need to rethink how they facilitate AI literacy perhaps moving beyond policy-focused discussions toward applied, case-based learning that mirrors workplace integration.
Some Things to Think About
For Academic Program Administrators
Build Discussion Culture, Not Just Curriculum: The data suggests that vibrant, ongoing AI conversations matter more than formal coursework. Consider creating AI learning communities, peer mentorship programs, and regular showcase sessions where students share their latest AI applications. There are already events with AI LA and PyLadies that the program does undertake.
Democratize Access Through Scaffolding: Since preparedness isn’t predicted by background, focus on creating multiple entry points. Provide prompt libraries, project templates, and guided demonstrations that help any student begin their AI journey, regardless of starting point.
Adopting AI Together: The technology is new for everyone from students, instructors, educators. Channeling the student enthusiasm for learning and their AI competency through formal institutional programs and collaborative projects with other schools and centers such as USC Centre for AI in Society can develop community enthusiasm and boost discussions around AI in the program.
For LLM Product Teams
Working and Designing in the Multi-Platform Reality: Students aren’t looking for a single AI solution. Designing for interoperability and making it easy for users to incorporate your tool into broader workflow is better than being their only platform. This could mean owning a vertical, a workflow that is core to student life functions.
Context-Specific Paths: The data shows students use AI differently across life contexts. I am curious if companies have tried developing use-case-specific onboarding and value delivery flows such as “AI for Career Advancement,” “ for Academic Success,” that help users quickly identify relevant applications.
Having Something for Everyone: The user base includes both prompt engineering novices and students building multi-LLM workflows. What is the range of capabilities that your product can offer? Or is it losing advanced users to players with better developer ecosystem? Key might be creating product experiences that grow with users rather than overwhelm beginners or bore power users.
Execution in an accessible and competitive landscape
The technology is a level playing field, for maintaining competitive edge ensuring AI literacy of students, staff and educators seems non-negotiable day by day. Some fantastic, randomized control trials are underway studying impact of AI in education and will hopefully provide positive learnings to look forward to. These insights are supportive for administrators as the question isn’t whether students will become AI-literate, but whether institutions will be able to evolve quickly enough to embody and guide that transformation thoughtfully and proactively.
For survey methodology, detailed statistical analysis, and raw data visualizations please reach out to me via email. Special thanks to MSBA 2025–26 class who participated and shared their journeys and Professor Dawn Porter, Academic Director for guiding this initiative.
