AI summaries on social media are reshaping how users form opinions about political topics, yet their influence remains largely unexamined despite their widespread deployment. This paper investigates how two types of AI summaries affect user opinions and engagement: textual summaries of discussion narratives and percentage breakdowns of agreement/disagreement. Through a 144-participant experiment on simulated online discussion threads, we found that displaying commenter agreement percentages amplified social conformity towards the majority views beyond reading comments alone. Conversely, AI narrative summaries created misperceptions of balance in polarised threads, reducing opinion change. While these summaries did not influence participants’ willingness to engage, toxic discussions deterred participation even when participants held majority views. Based on our findings, we provide critical design interventions for industry and researchers to mitigate these tools’ polarising effects, paving the way for responsible AI deployment on social media platforms.

As multi-agent Large Language Models (LLMs) gain traction, designers must consider how to surface their internal reasoning in ways that foster appropriate trust. We present a design-led, qualitative, comparative structured observation study, exploring how users interpret and evaluate transparency in multi-agent LLMs. Participants interacted with five interface variants, each instantiating different combinations of transparency-related design dimensions, across two task types: information-seeking and logical reasoning. We surface participants’ mental models, the cues they interpret as signals of transparency and trustworthiness, and how they weigh the costs and benefits of increasing process visibility. Transparency needs were dynamic and context-sensitive, with the ideal “Goldilocks” (i.e., “just right” transparency) level shaped jointly by task demands, interface affordances, and user characteristics such as task expertise and dispositional AI trust. We highlight tensions between process visibility, information sufficiency, and cognitive effort, and synthesise these insights into design considerations for aligning transparency with user needs in future multi-agent LLM interfaces.

Virtual AI-generated Influencers (VAIIs) are increasingly being used by corporations and public agencies, raising questions about how their visual design and communication strategies impact end-users’ propensity to accept the messages they deliver. We examined the impact of human-likeness (how closely a VAII resembles a human) and self-disclosure (whether the message contains personal information) on message acceptance, alongside dispositional factors like empathy and anthropomorphising tendencies. In a mixed-methods experiment, participants (N=120) watched short-form videos featuring VAIIs of varying human-likeness (High/Moderate-High/Moderate-Low/Low) and self-disclosure (present/absent). We observed the strongest message acceptance from the VAIIs with the lowest human-likeness, and message rejection for VAIIs with moderate to low human-likeness. Additionally, participants’ message acceptance was influenced by their empathy tendencies. Our qualitative analysis revealed further insights into participants’ perceptions of the human-likeness of VAIIs, their discomfort with self-disclosure, and their tendency to anthropomorphise VAIIs. These findings provide important implications for the design of VAIIs.

The start of the 2020s ushered in a new era of AI through the rise of Generative AI Large Language Models (LLMs) such as ChatGPT. These AI chatbots offer a form of interactive agency by enabling users to ask questions and query for more information. However, prior research only considers if LLMs have a political bias or agenda, and not how a biased LLM can impact a user’s opinion and trust. Our study bridges this gap by investigating a scenario where users read online news articles and then engage with an interactive AI chatbot, where both the news and the AI are biased to hold a particular stance on a news topic. Interestingly, participants were far more likely to adopt the narrative of a biased chatbot over news articles with an opposing stance. Participants were also substantially more inclined to adopt the chatbot’s narrative if its stance aligned with the news—all compared to a control news-article only group. Our findings suggest that the very interactive agency offered by an AI chatbot significantly enhances its perceived trust and persuasive ability compared to the ‘ static ’ articles from established news outlets, raising concerns about the potential for AI-driven indoctrination. We outline the reasons behind this phenomenon and conclude with the implications of biased LLMs for HCI research, as well as the risks of Generative AI undermining democratic integrity through AI-driven Information Warfare.

Online polarisation can tear the fabric of civility through reinforcing social media’s perceptions of division and discord. Social media platforms often rely on content-moderation to combat polarisation, contingent on the reactive removal or flagging of content. However, this approach often remains agnostic of the underlying debate’s ideas and stifles open discourse. In this study, we use prompt-tuned language models to mediate social media debates, applying the strategies of the Thomas-Kilmann Conflict Mode Instrument (TKI). We evaluate multiple mediation strategies in providing targeted responses to the debates, as shown to a debate audience. Our findings show that high-cooperativeness TKI strategies offered more persuasive arguments, while an accommodating argument strategy was the most successful at depolarising the audience’s opinion. Furthermore, high-cooperativeness strategies also increased the perception that the debaters will reach a consensus. Our work paves the way for scalable and personalised tools that mediate social media debates to encourage depolarisation.