Subject Teachers’ Behavioral Intention to Use Generative Artificial Intelligence in Rural Serbian Primary Schools (Grades 5–8): An Extended UTAUT Mixed-Methods Study

Generative artificial intelligence (GenAI) is increasingly available to schools, but evidence on GenAI acceptance and intention to use it among subject teachers in rural primary education remains limited. This study examined behavioral intention to use GenAI among 202 subject teachers teaching Grades 5–8 in rural Serbian primary schools through an extended Unified Theory of Acceptance and Use of Technology (UTAUT) framework and semi-structured interviews with 20 survey participants. The initial composite-score regression explained 10.5% of Behavioral Intention variance (R2 = 0.105). Performance Expectancy was marginally positive (β = 0.136, p = 0.049) and Facilitating Conditions showed a nominal, unadjusted negative association (β = −0.187, p = 0.009); however, the Performance Expectancy coefficient was no longer nominally significant with HC3 robust standard errors (p = 0.054), and neither coefficient survived Holm correction applied to the HC3 hypothesis-test p-values. Prior GenAI use (83.7%) did not materially change the model. A post hoc PLS-SEM sensitivity analysis, using theory-guided indicator refinement and 10,000 bootstrap resamples, showed composite reliability and convergent-validity indices of CR = 0.820–0.928 and AVE = 0.560–0.865, with low HTMT values (maximum = 0.482). Unadjusted bootstrap intervals suggested a positive Effort Expectancy path and a negative Facilitating Conditions path, but none of the eight hypothesized paths survived family-wise multiplicity correction. Facilitating Conditions also showed severe floor restriction (M = 1.34, maximum = 2.50/5; 52.0% at the minimum), while interviews described infrastructure as a barrier—a qualitative pattern that diverged from the negative coefficient. The findings therefore indicate limited and measurement-sensitive support for transferring an extended UTAUT model to this context and emphasize the need for stronger prospective measurement validation, context-specific constructs, and longitudinal research.