The articles refer to large-scale experiments published in the journals Nature and Science and show that dialogue-based interactions with large language models can cause measurable shifts in political attitudes – in some cases more strongly than traditional election advertising.
CAIS professor Hendrik Heuer classifies the results with particular reference to the mechanisms of action identified in the Science study. In his view, the central role of information density is particularly noteworthy: it is not personalization or narrative design, but the quantity and structure of the information conveyed that proved to be decisive for the persuasiveness of AI systems. Heuer considers this result to be “very reassuring” as it suggests that changes in political opinion are more likely to come about through substantive debate than through emotional manipulation.
At the same time, the articles make it clear that these effects are associated with risks. Models that were particularly optimized for persuasion tended to produce false information in the course of intensive dialogues. Hendrik Heuer’s assessment thus positions itself between recognition of the potential of AI-supported communication for informed political discourse and the need for critical reflection on its use in the democratic process. At the same time, the article underscores the role of CAIS as a place for scientifically sound assessment of socially relevant AI developments.