Demographic Prompting at Scale: When More Attributes Hurt LLM--Human Agreement

2026-07-12

AI Summary

This study investigates how incorporating annotator demographic attributes into prompts affects alignment between large language models and human judgments. Through systematic evaluation of five open-source models across five tasks—using prompts that include combinations ranging from one to all available attributes—the authors find that model–human agreement peaks when prompts contain one to three high-signal attributes, while including all attributes degrades performance. The results reveal that alignment efficacy depends not on the overall influence of attributes on human annotations, but on their learnable signal strength and directional consistency. Neuron probing further uncovers a correlation between specific activation patterns and alignment gains. These findings indicate the existence of an “over-specification threshold” and demonstrate that prompt effectiveness is jointly governed by signal quality, task characteristics, and model architecture, offering fine-grained guidance for controllable alignment.

Abstract

We investigate how annotator demographic attributes, supplied as prompt cues, shape the alignment between large language model (LLM) predictions and human annotations across five tasks. Using five open-source LLMs, we systematically vary the number and composition of demographic components in the prompt, spanning every combination from single-attribute through full-attribute configurations. Our experiments reveal three principal findings. First, alignment consistently peaks with one to three high-signal attributes and degrades under the full attribute set, establishing a clear over-specification threshold. Second, the overall magnitude of demographic influence on human annotations does not predict which attributes improve LLM alignment; instead, both the learnability and the directional coherence of each attribute's annotation signal need to be considered jointly. Third, neuron probing reveals that specialized activation correlates with alignment gains only under coherent annotation signals, and that activation volume alone does not imply steerability. Together, these results demonstrate that demographic prompting is not a monolithic intervention: its utility is highly context-dependent, shaped by attribute signal quality, task characteristics, and model architecture.

Problem

• demographic prompting
• LLM-human agreement
• attribute over-specification
• annotation alignment
• large language models

Innovation

• demographic prompting
• LLM-human alignment
• over-specification threshold
• annotation signal coherence
• neuron probing