Hide or Highlight: Understanding the Impact of Factuality Expression on User Trust

Hiding inaccurate information can paradoxically increase user trust: The study found that "opaque" (removing less factual content) and "ambiguity" (making uncertain content vague) strategies led to higher trust levels compared to transparently highlighting problematic information, suggesting users may prefer not seeing potentially false content over being warned about it.

Transparency about AI limitations may backfire: Highlighting factual errors through "transparent" labeling did not effectively build trust as researchers expected, indicating that directly pointing out where an AI might be wrong could undermine user confidence rather than improve it.

Users value perceived answer quality alongside trust: The opaque and ambiguity approaches maintained users' perception of answer quality while building trust, suggesting that the most effective approach balances both factors rather than prioritizing one over the other.

Current factuality disclosure methods have trade-offs: The research tested five different approaches to communicating AI accuracy, and no single method was universally optimal, indicating that organizations need to carefully consider their specific context when deciding how (or whether) to reveal AI uncertainty to users.

There's a tension between honesty and user confidence: The findings raise important questions about responsible AI design—whether hiding uncertain information serves users better than transparent disclosure, even though transparency is often considered an ethical principle in AI development.