Answers are grounded in this paper and cite their sources.
Generative-AI systems increasingly provide answers to everyday questions, but users need to decide not only whether to accept an answer, but whether acceptance should depend on answer correctness. This randomized online experiment tested whether visible source information and an accuracy warning supported more selective reliance on AI-style answers. A final analytic sample of 800 adult participants was assigned to one of four between-subjects cue conditions in a 2 × 2 × 2 mixed design crossing source transparency (absent vs. present), accuracy warning (absent vs. present), and answer correctness (incorrect vs. correct, within participant). Each participant evaluated 24 AI-style answers, yielding 19,200 trial-level observations. The primary analysis used all trials and modelled acceptance as a function of source transparency, accuracy warning, answer correctness, and their interactions. Participants accepted correct answers more often than incorrect answers, and the correct-incorrect acceptance gap was largest in the source-plus-warning condition. In the applied-risk model restricted to incorrect-answer trials, acceptance of incorrect answers was highest in the control condition (1080/2400, 45%) and lower in the source-only (864/2400, 36%), warning-only (672/2400, 28%), and source-plus-warning conditions (504/2400, 21%). Verification behaviour was highest in the combined condition (2496/4800, 52%). These results indicate that brief on-screen cues lowered erroneous-answer acceptance and encouraged verification in low-risk information-evaluation tasks, while the distinction between response bias and sensitivity remains an important boundary condition.
No takes yet. Share an insight, caveat, or question.
Xing et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: