AI in STEM education: The relationship between teacher perceptions and ChatGPT use

University of Education (psychology / digital media in education)

Freiburg, Germany

Researchers administering questionnaires and SEM/path analysis

• In-school (K–12)

Cross-sectional online questionnaire to STEM teachers

Single-wave survey study

n = 102 STEM teachers

ChatGPT / GPT-class tools as referent technology

• Self-report and cross-sectional design
• STEM-only sample may differ from humanities teachers
• Rapid ChatGPT versioning since 2024
• Causality limited without longitudinal behavioral traces

Secondary students indirectly (teacher-reported classroom integration)

Uneven student use; paper notes high German student AI use in HE/engineering tracks

STEM teachers may have stronger digital media self-efficacy than non-STEM peers

• Model relationships among perceived benefits, risks, competence, ChatGPT use, and future intentions
• Test whether perceived risks suppress perceived usefulness of ChatGPT in class
• Connect to teaching quality change perceptions

• Inform PD targeting competence and benefit framing
• Discuss ethical and pedagogical stakes of GenAI in STEM

• Not student learning outcome measurement
• Not classroom observation study
• Not multi-country comparison

ChatGPT (generative LLM) as primary AI referent

• Co-creator
• Automation tool

German secondary education context

• Teachers consider GenAI for preparation, feedback, and in-class support
• Affect-based judgments link benefits and risks negatively

• Academic integrity policies for STEM written work
• Transparency when AI assists grading or feedback
• Mitigate over-reliance despite low measured risk→use suppression in this sample
• Equity of access to paid vs free models

• Instrument development from researcher and policy classifications
• Collect responses from 102 STEM teachers
• Estimate path model for hypotheses
• Interpret affect heuristic and teaching quality pathways

• Teachers judge classroom usefulness; models summarize association patterns only

• Target PD on competence-building and realistic benefit scenarios
• Pair tool training with ethics and assessment design

• Risk perceptions may not deter usage if benefits and competence are high
• Need differentiated guidance for in-class vs out-of-class GenAI tasks
• STEM teachers may lead adoption—policy should include all subjects

• Path analytic framing beyond descriptive adoption surveys
• Explicit linkage to political/educational AI classification schemes in instruments

• Future usage expectations exceed current usage—room for growth

• Competence and benefits positively predict use and intention; risks negatively correlate with benefits

Suggests teachers may adopt GenAI while holding concerns; implications for quality-focused PD and policy.

• Anonymous survey handling and institutional ethics
• Avoid stigmatizing teachers who experiment with GenAI
• Student data not primary here but classroom GenAI raises FERPA-like issues when adopted
• Monitor vendor terms for school use of ChatGPT

• Post assessment
• Activity documentation
• Practitioner observation

• Longitudinal adoption with classroom observation
• Student achievement and integrity outcomes under STEM GenAI policies

• STEM teacher education
• Educational psychology of technology adoption
• GenAI in secondary schools

• Completed

Secondary (teacher lens)

Germany

ChatGPT integration

Survey / path model

Medium

Low–Medium