This study investigates information spillover mechanisms in China’s stock market under market shocks through constructing limit order book (LOB) networks. A methodological framework is developed that extracts informative variables from dynamically evolving LOB data, employs a high-dimensional vector autoregressive (VAR) model with post-double-selection LASSO regularization for Granger causality testing, and implements bootstrapping techniques to compute generalized forecast error variance decompositions. Four principal findings emerge from our empirical analysis. First, LOB networks demonstrate superior descriptive capabilities over conventional price networks by incorporating critical market microstructure variables, thereby enabling precise identification of information spillover mechanisms during market shocks. Second, the spillover dynamics exhibit complex network structures comprising both own-impact paths and significant cross impact paths. The coexistence of these mechanisms reveals that while information predominantly circulates within individual stocks, substantial cross market spillovers occur through key bridging nodes, collectively reflecting the market’s intrinsic correlation structure. Third, regulatory constraints on short selling induce notable bid-ask asymmetries in information diffusion, with predominant spillover directions originating from informationally advantaged traders to retail participants via bid-side order flows. Fourth, intensified herding behavior and overreaction patterns are observed in spillover dynamics during market downturns.