The Artificial Intelligence Acceptance Model (AIAM-21): Split-Sample Psychometric Evaluation of a Five-Factor Hungarian Measure

Artificial-intelligence acceptance is frequently assessed with broad attitude scores or ad hoc items that do not distinguish trust/reliability, usefulness, ease of use, intended use, and willingness to contribute feedback. This study reports the refinement and initial split-sample psychometric validation of the 21-item Artificial Intelligence Acceptance Model (AIAM-21), a Hungarian, structure-derived operationalization of AI-TAM. The de-identified respondent dataset comprised 819 eligible survey submissions, divided into a calibration sample of 405 and a held-out validation sample of 414. A polychoric maximum-likelihood exploratory factor analysis with oblimin rotation was broadly consistent with the intended five-factor solution after the removal of a weak, negatively worded candidate, although the collaborative-intention cluster showed localized calibration-sample instability. In the held-out sample, the fixed five-factor CFA showed good approximate fit (CFI = 0.977, TLI = 0.972, RMSEA = 0.076, SRMR = 0.042) and fit significantly better than a four-factor alternative that merged usefulness and behavioral intention. Composite omega coefficients ranged from 0.769 to 0.923, AVE from 0.613 to 0.846, and all polychoric HTMT2 values were ≤0.821. The final structural model reproduced across both subsamples. AI-use frequency provided concurrent construct-related evidence, whereas the recommendation item showed only weak associations and conflicting direction cues. Because the validation subsample contained only 104 men, multigroup ordinal gender invariance was treated as provisional and is deferred to a larger independent sample. AIAM-21 should be interpreted as five related subscales rather than a total score.