Costello, Pennycook and Rand: Just the Facts: How Dialogues with AI Reduce Conspiracy Beliefs

Costello, Pennycook and Rand: Just the Facts: How Dialogues with AI Reduce Conspiracy Beliefs

Thomas H. Costello, Gordon Pennycook and David G. Rand argue in their paper that personalized AI dialogues reduce conspiracy beliefs mainly through relevant, well-reasoned counterevidence—not rapport, storytelling or appeals to empathy. They report two experiments with 2,193 U.S. adults recruited online, testing possible explanations for the effect. The reported reduction remained when the researchers changed how the exchange was framed, shortened the AI’s responses, changed how beliefs were measured, and replaced dialogue with one tailored message. It was substantially reduced only when the AI was instructed to persuade without factual evidence and arguments. The authors’ account makes the proposed mechanism important: the findings point to the substance of the evidence, rather than a conversational style alone, as central to these participants’ changed beliefs. They suggest that people might be similarly effective if they could supply comparable evidence. The abstract reports these experiments with online-recruited U.S. adults; it does not establish whether the same results hold in other populations or settings.

Publication: 2fea85f3-388b-40d7-8837-ac1ff97e2e2f

attribution — pass: The reading correctly attributes the paper and its claims to Thomas H. Costello, Gordon Pennycook and David G. Rand. It keeps the mechanism claim and the suggestion about humans framed as the authors' argument rather than established fact.

Authors:Thomas H. Costello, Gordon Pennycook, David G. Rand

summary — pass: The summary accurately conveys the study design, the robustness tests, the one condition that substantially reduced the effect, and the proposed mechanism. 'Changed how the exchange was framed' compresses the disclosure, debate-framing and inform-not-persuade conditions, but it does not distort them. The language is accessible, and the question follows from the authors' own suggestion. The abstract's hedge ('appear') is softened slightly to 'argue', but attribution is preserved. Minor optional repair: list the disclosure and inform-not-persuade conditions explicitly.

AI dialogues appear to reduce conspiracy beliefs primarily by supplying relevant, well-reasoned counterevidence.

date — pass: The originalPublished value of 2026-11-03 matches the arXiv v1 submission date. The feed date of 2026-11-05 is correctly not used.

[Submitted on 3 Nov 2026]

support — pass: Each claim is source-stated and supported by the abstract. The limitations correctly note the sample scope and that only the abstract was available. The question restates the authors' own conditional suggestion as an open issue without adding unsupported benefits or costs.

Only instructing the AI to persuade without factual evidence and arguments substantially reduced the debunking effect.

media — pass: The selected media is the verified arXiv abstract page for this paper. It is the direct original source and the most relevant link.