Universities risk leaving students underprepared as AI literacy gaps persist

University students are entering an AI-shaped academic and professional world with moderate confidence in their ability to access, understand and use artificial intelligence, but their actual use remains cautious, uneven and heavily shaped by classroom rules, according to a new study published in Trends in Higher Education. 

The study, titled AI Literacy: University Students’ Perceptions and Practices, is based on survey responses from 130 undergraduate and graduate students at a large university in the southeastern United States. It examines how students perceive their own AI literacy across access, understanding, critical thinking, application and ethics, offering a detailed view of how higher education is responding to the rapid spread of generative AI tools such as ChatGPT.

Students use AI cautiously as ChatGPT dominates tool choice

The survey found that students reported overall agreement across all five AI literacy subscales, but their strongest scores came in ethics, while their lowest scores appeared in understanding. On a five-point scale, students averaged 3.48 for access, 3.27 for understanding, 3.63 for critical thinking, 3.34 for application and 3.85 for ethics. The pattern suggests that students are more confident in recognizing responsible use issues than in explaining how AI systems work.

The gap is crucial because AI literacy is no longer limited to technical knowledge. The study defines it as a multidimensional skill set that includes access to AI tools, understanding of AI functions, critical assessment of outputs, practical application and ethical responsibility. In higher education, that combination affects academic integrity, learning quality, data privacy and workforce readiness.

ChatGPT was by far the most commonly reported AI tool. Among students who named externally accessed tools, 67.7 percent mentioned ChatGPT. Grammarly followed at 13.3 percent, while Gemini, Perplexity, MagicSchool AI and Microsoft Copilot were each reported by smaller shares. In another item asking students to list the AI tools they use most often, 52.3 percent named ChatGPT, while 39.2 percent listed no tool.

The dominance of ChatGPT is notable because the surveyed university provided other AI tools to students, but only a small share of respondents reported using institution-approved options. The finding raises a practical issue for universities: students may gravitate toward familiar, widely known tools even when campuses officially support different platforms.

The study also found that 105 of the 130 participants had access to AI tools beyond those provided by the university. That suggests AI access is widespread, but access did not translate into frequent use. Only 5.4 percent reported using AI five or more times a day, 16.9 percent used it two to four times a day and 13.1 percent used it once a day. Another 20.8 percent used AI every once in a while, 30.8 percent used it infrequently, 2.3 percent said they had no regular access and 10.8 percent chose not to use AI.

Open-ended responses helped explain the caution. Some students used AI for brainstorming, grammar checks, study guides, summaries or help getting started on assignments. Others avoided it because they worried it might be considered cheating, did not trust AI outputs or did not know how to use the tools effectively. Several students appeared to rely on instructor permission before using AI in coursework, pointing to the central role faculty play in shaping student behavior.

The researchers found that students often learned about AI informally. Many relied on self-directed online searches, videos, reading or trial and error. Others learned from instructors, peers or seminars. A smaller group said they did little or nothing to learn about AI, reinforcing the need for structured instruction rather than assuming students will build AI literacy on their own.

AI literacy varies by field, level and frequency of use

The study found several important differences across student groups. Education majors reported significantly lower understanding of AI than non-education majors. That result is especially important because many education students are preparing to become teachers, a profession now facing urgent questions about AI use in classrooms, student writing, assessment and academic integrity.

The authors suggest education students may be more cautious about AI because their field places heavy emphasis on learning processes, student development and ethical teaching. If future teachers have lower confidence in understanding AI, teacher preparation programs may need to introduce AI literacy more clearly and earlier.

The study also found that undergraduates reported higher ethics scores than graduate students. This finding ran against the researchers’ expectation that graduate students would report stronger understanding of ethics. One possible explanation is that undergraduates may have had more recent exposure to AI discussions in general education or early coursework. Another is that graduate students may be more aware of the complexity of ethical issues, making them less likely to rate their own knowledge highly.

Frequency of use emerged as one of the strongest dividing lines. Students who used AI at least once a day reported significantly higher scores in understanding, critical thinking and application than students who used AI less often. The largest difference appeared in application, indicating that students who use AI more regularly are more likely to feel capable of applying it in academic contexts.

The pattern suggests that practice may build confidence and perceived literacy. However, it also raises a concern for universities. If students learn AI mainly by experimenting on their own, those who are cautious, uncertain or discouraged by unclear policies may fall behind peers who use the tools more often. That could create uneven readiness for AI-enabled workplaces.

No significant differences were found in AI literacy by gender, first-generation student status or financial aid status. This is an important finding because earlier research on digital access has often shown gaps linked to socioeconomic background. In this sample, students receiving financial assistance and first-generation students did not report lower AI literacy than their peers.

The lack of difference may reflect campus access to free internet, university devices, library technology and mobile tools. But the authors caution that self-reported literacy is not the same as measured competence. Students may feel confident without fully understanding AI systems, or they may underreport use because they are unsure whether certain embedded tools count as AI.

The survey’s qualitative responses also showed that students make AI-use decisions through a mix of personal rules, instructor policies, task type and ethical concern. Some used AI only for personal matters. Others used it as a starting point, for clarification, organization, summaries or grammar correction. Some followed school or teacher policies closely. Others avoided AI because they believed assignments should represent their own learning.

These patterns show that students are not approaching AI in a single, uniform way. They are negotiating tool access, classroom expectations, academic integrity concerns and personal judgment in real time. That makes clear university guidance more urgent.

Universities face pressure to teach AI literacy across every program

Higher education institutions cannot treat AI literacy as optional or assume it belongs only in computer science. AI is now part of writing, communication, education, research, workplace preparation and everyday digital life. Students need direct instruction in how AI systems work, when they are useful, where they fail and what ethical responsibilities come with their use.

The researchers argue that universities should adopt clear AI literacy frameworks and embed them across curricula. Existing frameworks often share common elements, including understanding AI functions, access, ethical practice, critical evaluation, security, discourse and practical use. But the study warns that frameworks must be translated into actual student learning experiences.

A key challenge is inconsistency across courses and disciplines. Students may be encouraged to use AI in one class and prohibited from using it in another. Without clear communication, students may become risk-averse or confused. This can limit learning and deepen uncertainty about academic integrity.

The findings also show that faculty remain key to AI literacy. Students often reported using tools introduced by instructors or avoiding AI when policies were unclear. That means faculty development is just as important as student training. Universities cannot expect students to develop responsible AI practices if instructors themselves lack shared guidance, confidence or institutional support.

Data privacy and bias also require stronger attention. Few students raised privacy and security concerns in their responses, even though AI systems can involve sensitive data use. The study highlights the need for explicit instruction about how AI tools collect, process and use information. Students also need to understand that AI outputs can reflect bias, reproduce errors and generate false or misleading information.

The issue is not simply whether students can use AI to complete assignments faster. The deeper question is whether students can evaluate outputs critically, protect data, maintain academic integrity and decide when AI use supports learning rather than replacing it.

The study calls for early AI literacy instruction in college programs, ideally before students encounter high-stakes coursework where rules and expectations may vary. It also points to the need for interdisciplinary planning, with universities bringing together faculty, administrators, technology teams and students to define what AI readiness should mean across fields.

The findings also reshape workforce priorities. Employers are increasingly expecting graduates to understand and use AI tools. If universities do not provide structured AI literacy training, some graduates may leave better prepared than others depending on their major, instructor exposure or personal experimentation. That uneven preparation could create new forms of career-readiness inequality.

The authors acknowledge a few limitations in their work. The sample came from one university and was concentrated in social sciences and humanities rather than hard sciences. The survey relied on self-reported data, which may not reflect actual AI competence. Future research should compare student perceptions with behavioral measures, such as whether students can identify AI-generated content, evaluate output quality or apply AI responsibly in real academic tasks.