Correcting misinformation is crucial for addressing societal challenges like climate change, as its propagation erodes trust in science. While the continued influence effect (CIE) is well-studied in text, effective correction strategies for video-based social media content remain largely unexplored. This is an urgent gap given that AI-driven tools now facilitate the rapid dissemination of misleading synthetic content. This study examines factors that optimize correction efficacy in this digital context.

We conducted a pre-registered online experiment (N = 879) exposing participants to a short-form AI-generated video containing prevalent climate misinformation (i.e., that volcanoes emit more CO₂ than humans). We tested interventions using a 2x2 design, crossing correction content (causal alternative vs. simple retraction) with multimedia signaling principles (high vs. minimal), alongside two control groups.

Preliminary results indicate that all corrections significantly reduced misinformation reliance compared to the misinformation control (d = -0.83, p < .001). Causal alternatives were significantly more effective than simple retractions (β = -0.13, p < .001). Crucially, a marginally significant interaction emerged (p = .093), showing that high signaling (reducing extraneous cognitive load) disproportionately benefited the more complex causal alternative. Mediation analysis confirmed this mechanism: high signaling increased perceived processing fluency, which in turn significantly reduced misinformation reliance (indirect effect p = .025). This research advances mechanistic understanding of the CIE in video formats and yields theory-driven guidelines for effective science communication. By combining explanatory depth with cognitive scaffolding, practitioners can better counter the rising threat of synthetic climate misinformation.