Retelling the News: Exploring AI-Driven Conversational News Interactions

Paper Title

Retelling the News: Exploring AI-Driven Conversational News Interactions

Publication Info

• Topic area: AI-driven conversational interfaces for news consumption
• Keywords: conversational AI, news interaction, large language models, user experience, personalization, trust, transparency, news consumption, design space, conversational user interfaces

Background and Problem

• Problem / challenge: Traditional news formats often fail to engage users effectively, with barriers such as information overload, lack of personalization, and declining trust in media. AI-driven conversational formats offer potential but face challenges in trust, accuracy, and user interaction.
• Significance: Enhancing news consumption through conversational AI could improve accessibility, engagement, and understanding, addressing critical issues like misinformation and declining civic engagement.
• Motivation and related work: Previous research has explored AI in news delivery, engagement, and personalization, but gaps remain in understanding how conversational formats reshape user experiences and trust. This paper builds on prior work by examining the experiential qualities of conversational news and mapping its design space.

Solution

• Proposed approach: A two-part study combining a co-speculation workshop and an in-the-wild field trial of a conversational news agent to explore design possibilities and user experiences.
• Novelty:
  1. Identification of five design dimensions shaping conversational news agents: relational stance, personalization strategies, conversational practices, positioning toward journalism, and source transparency.
  2. Empirical insights into user experiences with conversational news, highlighting benefits like clarity and engagement alongside challenges in trust and accuracy.
  3. Development of a conversational news probe using a news anchor role, offering a focused, text-based interaction format.
  4. Four design considerations for balancing personalization, engagement, contextual cues, and credibility in conversational news systems.
• Procedure and key techniques:
  • Co-speculation workshop: Eight participants explored three roles for conversational news agents (personal assistant, news anchor, companion) and produced artifacts like scripts and sketches to map the design space.
  • Field trial: Six participants used a conversational news agent for seven days, with data collected via questionnaires, diaries, chat logs, and interviews. The agent was designed to provide concise news summaries and support user-driven exploration.

Results

• Concrete findings:
  • Participants rated the agent highly for clarity (mean score: 6.17/7) and engagement (5.17/7) but showed lower intent to continue use (3.33/7).
  • Trust in the agent was generally high, though inaccuracies and inconsistencies (e.g., conflicting responses about Sweden’s NATO membership) were noted.
  • Users appreciated the lack of distractions (e.g., clickbait and visuals) but found the conversational format demanding in terms of active participation and phrasing queries.
• Advantage over baselines: The conversational format reduced distractions, supported curiosity-driven exploration, and offered a more focused news experience compared to traditional news interfaces.
• Experiments / evaluation:
  • Workshop: Participants explored speculative roles for conversational agents, identifying themes that informed the design space.
  • Field trial: Participants interacted with a news agent using a Retrieval-Augmented Generation (RAG) approach, with data analyzed qualitatively to identify interaction patterns and experiential themes.
• Limitations and future work:
  • Small sample size limits generalizability.
  • Single news source (Verdens Gang) constrained perspectives.
  • Lack of journalist/editor input and long-term engagement data.
  • Future work should explore broader configurations within the design space and refine technical implementations to address accuracy and transparency issues.

Summary

This paper investigates how conversational AI can transform news consumption by enabling interactive, user-driven exploration. Through a co-speculation workshop and a field trial, the study identifies five design dimensions for conversational news agents and highlights user experiences with a news anchor-style agent. While the format supports clarity and engagement, challenges in trust, accuracy, and personalization remain. The findings inform four design considerations to balance personalization, engagement, contextual cues, and credibility, offering a foundation for future development of conversational news systems.