What question did this study set out to answer?

This research aims to assess how reliably large language models judge empathic communication compared to human experts and crowdworkers.

February 12, 2026Open Access

When large language models are reliable for judging empathic communication

Key Points

This research aims to assess how reliably large language models judge empathic communication compared to human experts and crowdworkers.
Analyzed 200 real-world conversations where one speaker shares a personal problem and the other provides support.
Compared annotations from 3,150 experts, 2,844 crowdworkers, and 3,150 LLMs across four evaluative frameworks.
Evaluated interrater reliability among the three groups using expert scores as a benchmark.
Expert agreement on annotations is high but varies across subcomponents depending on clarity and complexity.
LLMs performed comparably to experts, often exceeding the reliability of crowdworker annotations.
Findings indicate that LLMs can be effectively validated for specific tasks, enhancing their use in sensitive emotional contexts.

Abstract

Abstract Large language models (LLMs) excel at generating empathic responses in text-based conversations. But, how reliably do they judge the nuances of empathic communication? Here we investigate this question by comparing how experts, crowdworkers and LLMs annotate empathic communication across four evaluative frameworks drawn from psychology, natural language processing and communications applied to 200 real-world conversations where one speaker shares a personal problem and the other offers support. Drawing on 3,150 expert annotations, 2,844 crowd annotations and 3,150 LLM annotations, we assess interrater reliability between these three annotator groups. We find that expert agreement is high but varies across the frameworks’ subcomponents depending on their clarity, complexity and subjectivity. We show that expert agreement offers a more informative benchmark for contextualizing LLM performance than standard classification metrics. Across all four frameworks, LLMs consistently approach this expert level benchmark and exceed the reliability of crowdworkers. These results demonstrate how LLMs, when validated on specific tasks with appropriate benchmarks, can support transparency and oversight in emotionally sensitive applications including their use as conversational companions.

Bookmark

View Full Paper