Abstract:

We tested and deployed AI tutors to support learning in higher education in Germany. These tutors were implemented as Custom GPTs. Their usage was evaluated using both quantitative and qualitative methods. The findings show that only about half of the students actively used the AI tutor; however, nearly all students made use of AI tools more generally to support their learning. In this sense, the AI tutor seems to be “nice to have” rather than “need to have.”

Across the courses, students primarily used AI for learning support, including assistance with structuring and writing. Most students perceived the use of AI as beneficial. The underlying mechanism appears to be that AI lowers barriers to learning, creates experiences of progress, and helps keep learning processes “in motion.” The results also suggest that higher-performing students may gain more from integrating AI into their learning, an observation consistent with the “digital divide” hypothesis.

At the same time, the qualitative interviews indicate that students have a largely realistic understanding of potential risks: they recognize the danger of becoming overly reliant on AI, as well as the risks of inaccurate outputs, outsourcing work, or reduced peer interaction. Overall, students expressed a desire for broader and more transparent integration of AI in teaching, supported by clear guidelines and subject-specific tools.

Keywords:

Higher Education, AI, Tutoring.