The Dialogic Paradox Across Platforms: Media Richness, Sentiment, and Phase-Dependent Engagement on Twitter During the COVID-19 Delta Rebound

Background. Crisis communication on social media during pandemic rebound periods remains under-studied, and two competing frameworks — Media Richness Theory (MRT) and Dialogic Communication Theory (DCT) — offer divergent predictions about which content features drive audience engagement under acute uncertainty. Recent evidence from Chinese government microblogs (Zhang et al., 2022) suggests dialogic markers may reduce rather than enhance engagement during rebound periods, contrary to DCT's direct prediction, but no comparable evidence has been reported for Western platforms.Objective. To identify which content features of COVID-19 vaccine-related tweets predicted audience engagement during the Delta variant rebound of July–September 2021, test whether these effects hold across sentiment operationalizations and pandemic phases, and compare the resulting patterns with prior Weibo-based findings.Methods. We analyzed 71,845 unique vaccine-related tweets from the Delta rebound window, with an extended analysis across three pandemic windows totaling 182,053 tweets. Sentiment was operationalized using two complementary models: VADER (lexicon-based) and Twitter-RoBERTa. Latent topics were extracted using BERTopic. Engagement was modeled with a Zero-Inflated Negative Binomial (ZINB) specification with eleven predictors.Results. Media richness (URL presence) was the strongest content-level predictor of engagement (IRR = 4.56, p < 0.001). All three dialogic markers were negatively associated with engagement: question presence (IRR = 0.76), hashtag count (IRR = 0.87 per tag), mention count (IRR = 0.89 per mention). This dialogic paradox held across all three pandemic windows and mirrors Zhang et al.'s Weibo findings, indicating cross-platform convergence. Positive sentiment had a modest effect that approximately doubled between tools (VADER IRR = 1.37 vs RoBERTa IRR = 1.91). The extended analysis revealed a phase-dependent reversal in the media-richness effect: beta = +1.26 at Delta peak but beta = -0.37 in the post-Delta inter-peak window (Wald z = 61.9, p < 0.001).Conclusions. During peak-uncertainty crisis phases, informational richness outperforms dialogic engagement — a pattern that contradicts DCT's direct predictions but replicates across platforms. Media Richness Theory requires reformulation with a crisis-phase parameter. For public health communicators, prioritizing URL-bearing content over conversational engagement markers is the strongest single lever identified in this study.