Prompt Framing Impacts LLM Cultural Alignment and Value Responses

New research investigates how different prompt framing strategies affect the cultural alignment of large language models (LLMs). The study tested whether asking LLMs to adapt to users, role-play populations, or forecast human answers to value-laden questions would yield interchangeable results. Using the World Values Survey, researchers evaluated several leading LLMs across various language-country contexts. They found that prompt framing is a primary factor determining cultural alignment, with country-specific cues often shifting answers substantially, though not always towards human response distributions. Third-person forecasting prompts consistently produced the strongest directional alignment for most models, outperforming personalization and role-play. Alignment gains were most noticeable in areas like religiosity and gender roles, while institutional trust questions remained challenging. This indicates that prompt design is crucial for eliciting culturally appropriate responses from LLMs.