Changes in attitudes toward meat consumption after chatting with a large language model

Researchers have started to explore the persuasive power of large language models (LLMs) and initial results suggest that LLMs can be as persuasive as humans, even for controversial and moralized topics. We conducted a preregistered proof-of-concept study to test how a brief conversation with ChatGPT-4o influences attitudes and beliefs about meat consumption. We found that participants (n = 101) reported weaker commitment to eating meat and weaker beliefs that eating meat is necessary and natural, but not weaker beliefs that eating meat is normal or nice, after their conversation (vs. before). Our study provides preliminary evidence and can act as a blueprint for future explorations of how conversations with LLMs can change attitudes on meat consumption and related topics.