Pragmatic users and skeptical nonusers: A qualitative typology of ChatGPT adoption in physics education

Synthesis: Becker, Bauer, Schrader, Bitzenbauer & Veith (2026) analyze 1,189 survey responses from physics students using qualitative content analysis and latent class analysis, identifying two distinct user profiles: 70% are "Pragmatic Users" who use ChatGPT for Scaffolding despite awareness of inaccuracies, and 30% are "Skeptical Non-Users" who avoid it over overreliance concerns. Both groups make calculated risk-utility trade-offs — challenging one-size-fits-all AI policies and calling for differentiated pedagogy.

Summary

This study develops a data-driven typology of physics students' ChatGPT use through qualitative content analysis of 1,189 survey responses and latent class analysis (LCA). Two profiles emerge: Pragmatic Users (70%) who are aware of AI's inaccuracies but use it for conceptual clarification and Problem Solving scaffolds, and Skeptical Non-Users (30%) who avoid the tool over concerns about overreliance and its potential to hinder independent problem-solving. Both groups demonstrate calculated risk-utility trade-offs rather than uncritical adoption or blanket rejection — challenging institutional one-size-fits-all AI policies.

Key Findings

• Qualitative content analysis of 1,189 responses + latent class analysis
• Two profiles: 70% Pragmatic Users, 30% Skeptical Non-Users
• Pragmatic Users use ChatGPT for conceptual clarification, finding starting points, scaffolds — despite awareness of inaccuracies
• Skeptical Non-Users avoid ChatGPT over overreliance/independent problem-solving concerns
• Both groups make calculated trade-offs; neither is uncritical
• Calls for differentiated pedagogy respecting both profiles

What this means for practice

• Learners. Decide which ChatGPT use you are endorsing. The 70% Pragmatic User profile uses the tool for conceptual clarification, finding a starting point, and Scaffolding — keep the answer itself out of the loop, since only 15% of this sample reported using it for direct problem solving.
• Learners. Target the failure modes this sample actually named: incorrect calculations and difficulty with complex physical problems, not generic AI inaccuracy.
• Instructors. Require students to critically evaluate and justify the reasoning behind AI output, and co-create prompting templates with them; with 70% already using the tool, guided reflective use is what converts adoption into AI Literacy.
• Instructors. Respect the refusal. Declining generative AI remains a valid stance for the 30% Skeptical Non-User profile who prioritize independent problem solving and productive struggle; offer that group exposure only in low-risk, bounded exploratory or feedback tasks.
• Administrators. Reject one-size-fits-all AI policy. Both profiles make calculated risk-utility trade-offs, so a mandate misreads the skeptical group and a ban misreads the pragmatic majority.

Limitations

• Cross-sectional online questionnaire. 1,189 German physics students (N = 1,019 after excluding response categories named by fewer than 5% of participants), with self-report bias, sampling bias, and no temporal follow-up.
• Coding was largely automated. The qualitative content analysis was carried out mainly by a generative language model with only 10% human verification of both codebook development and annotation, so coding errors and overlooked nuance remain possible.
• Zero-inflation in the latent class analysis. Categories were converted to indicator variables, so a topic a participant never mentioned is coded as "no endorsement"; the dataframe fed to the LCA is likely overpopulated with zeros, and filtering sparse variables mitigates but does not eliminate the distortion.
• One tool, one discipline. Findings are specific to ChatGPT in physics study programs at German universities and should not be read as generalizable to generative AI as a whole.

Citation

Becker, E., Bauer, A., Schrader, J., Bitzenbauer, P., & Veith, J. (2026). Pragmatic users and skeptical nonusers: A qualitative typology of ChatGPT adoption in physics education.