Early post-publication behavioural signals of misinformation on social media

Social-media users often encounter contested claims before professional fact checks become available. This study examines whether the expression of seed posts and the early audience responses they receive differ across nine Snopes claim-level verdict categories, and whether these behavioural signals support prospective classification. The corpus links 588 public X seed posts to 28,600 quote-post observations during the first 24 hours. Analyses preserved the associated Snopes article as the grouping unit and assessed the temporal onset of distributional differences, the contribution of textual, affective, and engagement features, and predictive performance at prespecified early cutoffs. Among posts with observed quote activity, six of nine verdict categories met the one-vs-rest persistence criterion from hour 1, and 33 of 36 class pairs met the criterion at some point during the first day. Category-associated differences in post expression and audience response therefore became observable early at the group level. Individual classification, however, remained modest and non-monotonic. The strongest hour-1 macro- was .237, compared with .113 for the strongest dummy baseline, and performance did not improve systematically through hour 24. Seed-post features carried substantial early information, while quote-content and engagement features added more limited information as responses accumulated. Early post-publication behaviour provides a temporally resolved account of how posts associated with different claim-level verdicts are expressed and received, but it does not permit reliable stand-alone prediction of the associated claim-level verdict. The study provides a behavioural benchmark for future research on human–platform interaction and carefully evaluated decision support.