Framing the Narrative: How Question Framing Shapes LLMs’ Political Views
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Large Language Models (LLMs) are quickly becoming a dominant source of information, raising critical questions about their influence on public opinion, particularly in politically polarized contexts. This paper investigates how the framing of user prompts can shape the political leaning of responses generated by LLMs, building on prior research that highlights LLMs’ sensitivity to prompt wording and their tendency toward certain political biases. We theorise that three interacting mechanisms are at play: (1) the training text corpora, (2) Reinforcement Learning with Human Feedback (RLHF), and (3) framing effects similar to human cognitive biases. Through a series of experiments on contentious political issues, we investigate whether variations in prompt formulation can significantly influence the ideological tone of LLM outputs. If this is indeed the case, these findings would highlight the potential impact that LLMs may have on the opinions of their users and would inform ongoing debates about the societal risks posed by LLMs, especially in exacerbating affective polarization.
Fields of Study & Subjects
Ethics and Social Impacts of AI
Misinformation and Its Impacts
Social Media and Politics
Publication details
- Publication date
- 30 March 2026
- Language
- EN
- Document type
- Preprint