What do AI-TPACK instruments really measure? A systematic review of validity evidence and the nomological network

Artificial Intelligence (AI) has generated numerous extensions of Technological Pedagogical Content Knowledge (TPACK), yet their measurement evidence remains insufficiently evaluated. Following PRISMA 2020, this systematic review searched Semantic Scholar, ERIC, and Crossref in October 2026, identifying 175 journal articles: 25 measurement studies, 33 framework studies (Strand A), and 121 quantitative applications (Strand B), with four overlapping. Measurement quality was assessed using established testing standards and COSMIN criteria, while quantitative findings were mapped. Of 25 instruments, 22 relied on self-report, ten used non-TPACK labels, and none assessed contextual knowledge. Only four demonstrated sufficient structural validity; two examined unspecified invariance levels, and none investigated response processes or score-use consequences. Framework extensions emphasized ethics (21/33), human-AI collaboration, and generative AI, although only seven incorporated empirical, expert-consensus, or design evidence. Most quantitative studies (82.6%) were cross-sectional. AI literacy, self-efficacy, and institutional support emerged as consistent correlates, while acceptance and intention dominated outcomes. Only five of 20 interventions included comparison groups. Evidence connecting AI-TPACK to observed teaching and student outcomes remained limited, with conflicting lesson-plan findings. Future research should prioritize clearer constructs, performance-based assessments, measurement invariance, and longitudinal or experimental designs linking competencies to classroom practice and student learning outcomes.