Reproducibility is not construct validity: LLM measurement of institutionally situated communication

LLM-inferred measures of institutional communication exhibit high reproducibility with intraclass correlations exceeding 0.99, yet fail to converge with ground-truth survey data. In European Commission AI Act consultations, business associations showed a +1.0 divergence in risk concern between text and surveys, while public authorities exhibited smaller or negative divergences.