Longitudinal Insights into AI in Education: Usage, Ethics, and Policy Development in Higher Education

Synthesis: Parker, Loper, Carter, Hayes, and Karakas (2026) analyze four semesters of survey data from more than 300 students in a Midwestern US Teacher Education program to trace the evolving use and ethical perceptions of generative AI after ChatGPT's launch. GenAI use grew substantially, especially for assessment preparation (57% to 83%) and studying (44% to 76%), while students showed growing uncertainty rather than fixed stances about ethical use, and ethical perceptions differed significantly between AI users and non-users.

Key Findings

• Across four semesters, student use of GenAI grew substantially: for assessment preparation from 57% to 83% and for studying from 44% to 76%.
• Ethical perceptions showed ongoing uncertainty, with 'unsure' responses steadily increasing across survey iterations.
• There was a statistically significant difference in ethical perceptions between AI users and non-users.
• Students continued using ChatGPT for coursework even as institutions considered or introduced AI-detection software, underscoring tensions between policy and practice.

What this means for practice

• Administrators. Replace the assumption that policy alone governs behavior with formal, ethics-inclusive AI training inside Teacher Education, because ethical uncertainty persisted even as use rose.
• Instructors. Make acceptable AI use explicit in each course's rules and assignments, since students continued using ChatGPT for coursework while institutions adopted detection software.
• Administrators. Treat AI detection as one weak signal rather than a governance tool; rising detection rates may reflect rising prevalence rather than better detectors.
• Instructors. Schedule explicit ethical discussion, because in every wave students who used AI rated its academic use as more ethical than students who did not.

Limitations

• The study ran at a single public research-intensive university in the Midwestern United States, using purposive and convenience sampling from one teacher education program, which the authors say limits generalizability.
• All measures come from an 11-question self-report survey vulnerable to social desirability bias, and some of the research team also taught the surveyed classes.
• Of 423 eligible students, 319 completed the survey (about 75%), leaving only 65, 60, 96, and 98 responses across the four waves — small counts for comparing subgroups.
• Some survey items changed across the four waves, and external factors such as specific course requirements and disciplinary norms were not accounted for.

Citation

Parker, L., Loper, A. J., Carter, C. W., Hayes, J., & Karakas, A. (2026). Longitudinal insights into AI in education: Usage, ethics, and policy development in higher education. Computers and Education Open, 10, 100329.