Empowering teachers to use AI for differentiation: changes in teachers’ acceptance of AI-based technologies following participation in a short professional development training session

Synthesis: Mesenhöller and Böhme (2026) report a pilot pre-post evaluation of INSIGHT (Information on Schooling with Generative and Helpful AI Tools), a three-hour teacher professional development session on AI-based technologies in schools, with a focus on large language models. Drawing on the Technology Acceptance Model, the authors measured social acceptance as perceived usefulness, perceived ease of use, and behavioral intention among 100 German teachers from primary and secondary education in one federal state, using a pre-post design without a control group. Paired-samples t-tests showed significant pre-to-post increases in perceived usefulness and perceived ease of use, but no significant change in behavioral intention to use AI-based technologies for differentiation. Teachers also rated the training as highly helpful and broadly relevant. Taken together, the findings suggest that a short, practice-oriented training can shift how teachers judge AI tools without, on its own, shifting their stated intention to adopt them in everyday instructional practice.

Key Findings

• Perceived usefulness rose after the three-hour training. A paired-samples t-test on 100 teachers showed a significant pre-to-post increase, t(99) = -3.24, p = .002, mean difference = -0.14, a small effect (d = .32).
• Perceived ease of use also rose, with a significant increase reported after participation and a small effect size (d = .25), indicating teachers found AI-based technologies somewhat easier to use.
• Behavioral intention did not change. The intention to actually use AI-based technologies for differentiation showed no significant pre-post difference, despite the gains in usefulness and ease of use.
• Baselines were already favorable. Before the training, perceived usefulness averaged 2.89 (SD = .49), perceived ease of use 2.85 (SD = .72), and behavioral intention 3.07 (SD = .63) on a four-point Likert scale, suggesting a possible ceiling effect.
• The training was rated highly helpful and relevant across subjects, school types, and grade levels, although ratings for participants' own future lesson planning and instructional design were somewhat lower than for general relevance.

Study Design & Method

INSIGHT was designed as a pilot intervention to be refined before scaling into a larger format. The three-hour session combines theoretical input, hands-on experience, and guided reflection across four parts: an introduction to AI, training an AI-based system, LLM-supported individualized instruction, and teacher-guided support of students' AI competencies. The evaluation collected pre-post self-report data from N = 100 teachers in primary and secondary education in a single German federal state, using the SAELKIS instrument, which operationalizes the Technology Acceptance Model through perceived usefulness, perceived ease of use, and behavioral intention. Quantitative analysis relied on descriptive statistics and paired-samples t-tests. Two research questions guided the work: whether acceptance constructs changed after participation (H1a to H1c), and whether teachers judged the training helpful and relevant (H2).

What the pre-post changes showed

Perceived usefulness and perceived ease of use both increased significantly, supporting H1a and H1b, while behavioral intention showed no significant change, leaving H1c unsupported. The authors note that the changes were statistically significant but small, and that the pre-training means already exceeded the theoretical midpoint of the four-point scale. That pattern is consistent with earlier survey work on teachers' AI acceptance conducted before LLMs became widely available. For behavioral intention, the high baseline (M = 3.07) suggests a ceiling effect with limited room for improvement. Beyond that methodological reading, the authors argue that teachers may regard AI-based technologies as useful and easy to use while still perceiving contextual barriers to using them. Because those contextual and organizational factors were not measured here, the interpretation remains speculative, but it raises the possibility that acceptance depends on more than the original model's constructs.

What this means for practice

• Trainers and faculty developers. A three-hour session can measurably improve how teachers judge AI tools, so short formats are worth building where teachers cannot commit to lengthy programs. Do not expect a short session, on its own, to change stated intention to adopt.
• School leaders. Pair any training with the conditions that make adoption possible, since the study's authors point to time, digital infrastructure, and unclear institutional rules as likely constraints on turning favorable perceptions into practice.
• Designers of teacher AI competency programs. Anchor content in concrete instructional demands such as differentiation and individualized materials, which is how INSIGHT connected AI capabilities to teachers' daily planning work.
• Policymakers. Treat regulation and guidance as part of the adoption equation rather than as background, because participants evaluated the training before national recommendations on educational AI use were available.

Limitations

• The study is a pilot with a pre-post design and no control group, which limits causal claims about the observed changes; the authors present it as feasibility evidence rather than definitive proof of effectiveness.
• The sample was drawn from a limited number of schools in one German federal state, and the final analytic sample was small, restricting generalizability to German teachers at large and to other systems.
• Social conditions matter here too: classrooms are heterogeneous, so access to such training and the infrastructure that supports AI use are unevenly distributed, and the findings say nothing about whether the benefits reach schools with fewer resources.
• Finally, the SAELKIS items refer to AI-based technologies rather than explicitly to LLMs, so it remains unclear whether participants answered with LLM applications or broader AI in mind. Subsequent work should use larger, more diverse samples, comparison groups, longer follow-up, and measures of actual use.

Connected Concepts

• Technology Adoption Models — the framework structuring the perceived usefulness, ease of use, and intention measures
• Workplace Learning
• Teacher AI Competency
• AI Literacy
• Large Language Models (LLMs)
• Generative AI
• Personalized Learning
• Equity

Connected Articles

• Transforming Curriculum Design with Generative AI: A Model for Assessing Teacher Digital Competence — TAM-based assessment of teacher digital competence with GenAI
• Activity theory as a lens on teachers' adoption of AI technologies: A structural equation modeling — Activity theory and structural equation modeling of teacher AI adoption
• From Proficiency to Pedagogy: A Mixed-Methods Study of In-Service Teachers' TPACK-GenAI and the Mediating Role of Pedagogical Knowledge — In-service teachers' TPACK-GenAI and pedagogical knowledge

Citation

Mesenhöller, J., & Böhme, K. (2026). Empowering teachers to use AI for differentiation.