AI generates well-liked but templatic empathic responses

2026-04-09 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

Recent research indicates that more individuals are seeking emotional support from Large Language Models (LLMs), often rating LLM responses as more empathic than human-written ones. This success may stem from LLMs consistently deploying a well-liked template for expressing empathy. Researchers developed a taxonomy of 10 empathic language "tactics," such as validating feelings and paraphrasing, to characterize responses from both humans and LLMs. Across two studies involving 3,265 AI-generated responses from six models and 1,290 human-written responses, LLM outputs were found to be highly formulaic at a discourse functional level. A specific template, a structured sequence of these tactics, matched 83-90% of LLM responses (and 60-83% in a held-out sample), covering 81-92% of the response content when matched. Human-written responses, in contrast, exhibited greater diversity.

Key takeaway

For AI product managers developing emotional support features, you should recognize that while LLMs produce well-liked empathic responses, their reliance on a formulaic template could limit depth or authenticity over time. Consider integrating mechanisms to encourage greater diversity in empathic expression, potentially by fine-tuning on more varied human conversational data or by explicitly designing for less predictable response structures.

Key insights

LLMs generate highly-rated empathic responses by consistently applying a learned, formulaic template.

Principles

• LLMs excel at template-driven empathy.
• Human empathy is more diverse than AI empathy.

Method

A taxonomy of 10 empathic language tactics was developed and applied to characterize and compare human and LLM responses, identifying a consistent template in AI-generated text.

In practice

• Analyze LLM outputs for template adherence.
• Develop diverse empathy training for LLMs.

Topics

• Large Language Models
• Empathic Responses
• Emotional Support AI
• Empathic Language Taxonomy
• AI-Human Empathy Comparison

Best for: Research Scientist, AI Product Manager, AI Scientist, NLP Engineer, AI Ethicist

Related on AIssential

• Friendlier LLMs tell users what they want to hear — even when it is wrong — AI models are increasingly used for emotional support, with users perceiving AI responses as more empathic than human ones.
• Beyond Sentiment Classification: A Generative Framework for Emotion Intensity Evaluation in Text — Generative LLMs fine-tuned on continuous emotion intensity scores outperform classification and generalize to unseen affective dimensions.
• Can LLMs Think Like Consumers? Benchmarking Crowd-Level Reaction Reconstruction with ConsumerSimBench — LLMs struggle to reconstruct specific, socially grounded consumer reactions in public discourse, despite strong technical performance.
• When large language models are reliable for judging empathic communication — LLMs reliably judge empathic communication, matching experts and outperforming crowdworkers when properly benchmarked.
• Teaching Values to Machines: Simulating Human-Like Behavior in LLMs — LLMs can be induced to manifest human-like value structures, aligning strongly with human behavior patterns in simulations.
• Exploring conversational AI and poetry but not as we know it — LLMs challenge traditional notions of language, creativity, and authenticity, particularly in fields like poetry.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.