When rankings disagree: a robustness analysis of artificial intelligence readiness composite indicators

Global composite indicators of artificial intelligence readiness have proliferated as authoritative benchmarks that rank nation states along scales of technological preparedness, although the measurement properties of these instruments have received little empirical scrutiny. This paper examines the internal structure, cross-index concordance, rank stability, and patterns of data absence across nine major international indicator systems through a merged country-year panel covering more than two hundred countries from 2000 to 2025. Principal component analysis estimated under multiple imputation shows that the dominant first component captures variance shared with national wealth, while parallel analysis retains a multidimensional factor structure that complete-case estimation obscures. A substantial share of high-correlation variable pairs span different indicator systems, so cross-system measurement redundancy is more pervasive than conventional threshold analyses suggest. Rank positions shift considerably across alternative weighting and variable-inclusion specifications, with the greatest instability concentrated among upper-middle-income countries. Missingness clusters along income and regional lines, rendering low-income countries structurally absent from the instruments designed to guide their development. The findings argue that AI readiness indicators are measurement instruments whose properties deserve methodological scrutiny rather than uncritical adoption in policy and governance contexts. The paper develops this position further into an operational eight-item methodological checklist for indicator producers.