Chat but verify: Combating misinformation in conversational Generative AI with verification affordance
Pre-registered online experiment (N=477) testing whether the conversational style of generative AI assistants leads people to trust misinformation, and whether a verification mechanism mitigates it. Participants interacted with a purported generative AI assistant that returned inaccurate information under a 2 (conversationality cues: low or high) by 4 (verification affordance: absent, cue, forced action, voluntary action) design. High conversationality cues raised trust by inhibiting the negative machine heuristic; acting on the verification affordance lowered the perceived credibility of the false output.
Publisher
Journal of Computer-Mediated Communication (Oxford University Press); University of Georgia; Pennsylvania State University
Published
29 Jul 2026
Added
today
Key Findings
- High conversationality cues produced higher trust in the misinformation, mediated by reduced activation of the negative machine heuristic (the belief that machines are mechanistic, cannot make subjective judgments and cannot detect emotions)
- Acting on the verification affordance reduced the perceived credibility of the generated information
- The authors frame conversationality as a heuristic cue that hallucinating systems can exploit and argue that verification affordances should be designed into conversational interfaces to reduce overreliance on cognitive heuristics
Methodology Notes
Pre-registered between-subjects online experiment guided by the HAII-TIME model, N=477; single interaction with a mock assistant seeded with misinformation; self-report trust and credibility measures; recruitment platform not stated in the abstract. Open access (CC BY 4.0), volume 31 issue 3, published online 29 July 2026. The OUP article page blocks automated access; verified from the Crossref record (title, authors, affiliations, dates, licence, abstract) on 2026-09-17.
Sources
Authors
Liao, Mengqi, Sundar, S. Shyam
Tags
Cite This
APA
Liao, Mengqi, Sundar, S. Shyam. (2026). Chat but verify: Combating misinformation in conversational Generative AI with verification affordance. Journal of Computer-Mediated Communication (Oxford University Press); University of Georgia; Pennsylvania State University. https://doi.org/10.1093/jcmc/zmag012