The Decision to Verify: How Warmth and User Characteristics Shape Reliance on Conversational Agents for Information Search

A research study examines how users interact with conversational AI systems when fact-checking is accessible through hybrid search interfaces. The findings reveal that users continue to over-rely on AI answers despite having web search available, with verification behavior driven primarily by user characteristics like prior trust rather than answer quality, while conversational warmth indirectly increases reliance by boosting agreement with incorrect responses.

This academic research addresses a critical challenge in AI deployment: understanding why users fail to verify AI-generated information even when fact-checking mechanisms are readily available. The study challenges the assumption that access to traditional web search automatically corrects AI overreliance, revealing instead that behavioral patterns are deeply embedded in user psychology rather than system design alone.

The research emerges from growing concerns about conversational AI reliability in high-stakes applications. As language models become primary information sources, their tendency to generate plausible-sounding but incorrect answers—hallucinations—creates real risks. Previous work focused on technical solutions or alert systems, but this study shifts focus to user behavior, recognizing that trust dynamics fundamentally shape information verification.

For developers and platforms integrating conversational search, these findings carry significant implications. Building trustworthy systems requires more than enabling fact-checking; it demands understanding user demographics, digital literacy levels, and the subtle psychological effects of conversational design. The discovery that conversational warmth paradoxically increases agreement with incorrect answers suggests that maximizing user satisfaction through friendly AI personas may inadvertently undermine accuracy outcomes.

The research indicates that different user segments require different intervention strategies. Some users verify information regardless of context, while others default to trust, making one-size-fits-all design solutions ineffective. Future systems may need adaptive verification prompts, dynamic trust calibration based on user profiles, or explicit transparency about AI limitations. The finding that consulting multiple AI sources improves accuracy while traditional web search does not suggests that information architecture decisions—which sources to prioritize—carry underestimated importance in shaping user outcomes.

• →User reliance on conversational AI persists despite access to web search fact-checking tools
• →Verification behavior is primarily driven by pre-existing user characteristics rather than answer quality or context
• →Conversational warmth increases agreement with incorrect AI responses, creating a trust-accuracy tradeoff
• →Consulting multiple AI sources improves accuracy more than traditional web search does
• →Trustworthy conversational AI design requires user-adaptive strategies rather than one-size-fits-all solutions