Anchors promote sales by communicating product information through multimedia channels in live streaming e-commerce (LSE). This study focuses on how anchor nonverbal behaviors influence the sales during the LSE. Specifically, the curvilinear effect of anchor body movements on customer purchase, and the moderating effects are hypothesized from the perspective of information overload. 9,641 product pitches of live videos from Douyin are sampled for regression. A deep learning model is leveraged to measure the extent of the anchor body movements. The empirical evidence on Tobit model with anchor and day fixed effects indicates that the extent of anchor body movements, as an indicator of information density, has a significant inverted U-shaped effect on sales. And this curvilinear relationship remains consistent across different time periods. Moreover, a comprehensive set of conditions adjusting the information overload in LSE context are revealed. Specifically, the initial audience size, introduction duration, and product price, in relation to information source, information quantity, and task complexity respectively, moderate the curvilinear effect of anchor body movements on sales. A subsequent analysis with action clustering further explains this curvilinear effect in terms of specific action segments. Representative actions that enhance or impair sales are found. This study provides insights regarding nonverbal behaviors of anchors in LSE, uncovering the impact of information overload in this multimedia context. Situational practical implications are offered to enhance the performance of the anchors and promote the sales.