Large language models (LLMs) offer a scalable alternative to labour-intensive human norming studies in psycholinguistic research. This article introduces chatRater, an R package for rating text, image, and audio stimuli via multiple LLM providers, and demonstrates its use with a case study validating GPT-4o ratings of English idioms against human norms. We extend the probability-weighted scoring method of Brysbaert et al. (2025) to three idiom-specific dimensions: familiarity, literal plausibility, and decomposability. GPT-4o ratings of 45 English idioms were compared against two independent human norming sources (Bulkes & Tanner, 2017; Libben & Titone, 2008). Results show moderate-to-strong convergent validity: Pearson correlations ranged from r = 0.38 to r = 0.68 across raw-score comparisons, with familiarity vs. LT2008 showing negligible systematic bias (d = −0.18, p =.239). When converting to rank scores to eliminate distributional differences, correlations remained stable (r = 0.40–0.77), and all paired t-tests became non-significant, confirming that rank orders are well preserved. The alignment() function in chatRater provides a complete validation pipeline with an optional rank = TRUE parameter for automatic rank-score conversion.