Rethinking news framing with large language models

Area of Science:

• Media Studies
• Computational Social Science
• Psychology

Background:

• Mainstream media significantly influences public opinion.
• Existing research on media framing is limited in scalability and generalizability.
• Subtle biases like selective fact presentation and tone can impact attitudes.

Purpose of the Study:

• To introduce a novel framework using large language models (LLMs) to generate synthetic news articles.
• To systematically vary content selection and tone while maintaining factual accuracy.
• To investigate the impact of these controlled framings on public attitudes.

Main Methods:

• Development of a framework leveraging LLMs for synthetic news generation.
• Systematic variation of news content selection and tone.
• Conducting a large, pre-registered randomized experiment (N = 2,141) to assess framing effects.

Main Results:

• Selective presentation of accurate information significantly shifts policy views and emotional responses.
• Negative framings demonstrate stronger effects than positive framings.
• Individuals reporting lower levels of topic information show more pronounced effects.

Conclusions:

• Subtle media bias possesses significant persuasive power.
• LLMs are valuable tools for scalable and controlled research on media effects.
• Understanding framing effects is crucial for media literacy and informed public discourse.