Question 1

Should we display confidence indicators or uncertainty warnings in our AI search results?

Accepted Answer

Not straightforwardly. Li and Aral's experiment found that explicit uncertainty signals actually *suppress* user trust and decision quality, while hallucinated citations (false confidence cues) increase it. This inversion of the transparency assumption means adding disclaimers or confidence percentages may backfire—the interface choice itself becomes a policy decision about epistemic risk, one you need to test with your actual user population rather than assume will improve outcomes.

Question 2

How does this research apply across different user segments?

Accepted Answer

The study's differential vulnerability finding is critical: education level and demographics significantly mediate how users respond to AI trust cues. A 12,000-query experiment across seven countries and a U.S.-representative sample means interface choices don't affect all users equally. Before rolling out changes, you need to understand which populations are most susceptible to misleading signals and whether your design amplifies or mitigates that risk.

Question 3

Why should we treat search interface design as a governance question, not just a UX one?

Accepted Answer

Because every choice about reference links, confidence markers, and social proof distributes epistemic risk unevenly across your user base. Li and Aral's framing shows that generative interfaces actively shape how users form judgment and trust, not passively deliver information. That means design decisions encode assumptions about *who bears the cost* when AI fails—making it inseparable from questions of fairness and accountability.