LLMs can persuade only psychologically susceptible humans on societal issues, via trust in AI and emotional appeals, amid logical fallacies

People are resistant to changing their minds through AI conversations: The study found "longitudinal inertia in convictions," meaning participants largely stuck to their initial opinions even after repeated exposure to AI arguments on polarizing topics like climate change and social media misinformation, suggesting LLMs have limited persuasive power on deeply held beliefs.

Both humans and AI rely heavily on flawed reasoning: NLP analysis revealed that "both humans and LLMs relied on fallacious reasoning in 1 conversational quip every 6," challenging the assumption that AI systems are inherently more logical than people and indicating that conversations with LLMs don't necessarily elevate the quality of debate.

Only psychologically vulnerable people are susceptible to AI persuasion: The research identified "individuals more susceptible to LLM-based opinion changes," suggesting that persuasion by AI is not universal but rather targets people with specific psychological characteristics, raising concerns about differential vulnerability to AI influence.

Trust and emotional appeals drive AI persuasiveness more than logic: The study found that LLMs persuade through "trust in AI and emotional appeals," not superior reasoning, meaning the mechanisms of AI persuasion may be similar to human manipulation tactics rather than evidence-based argumentation.

AI's perceived humanness significantly influences how persuasive it is: The research showed that how human-like people perceived the AI to be was a strong predictor of opinion change and conviction shifts, suggesting that anthropomorphizing AI systems may increase their influence over people's beliefs.