The visual system processes vast amounts of information by internally weighting stimulus inputs based on their local features and behavioral relevance [1]. Although evidence shows that behavioral relevance influences stimulus saliency, it is not well understood how integrating stimulus and social information impacts perception and decision-making. Previous studies indicate that predicting others’ actions improves with real-time observation of their movements [3]. Additionally, game-theoretic simulations show that real-time feedback leads to leader-follower behavior, where one agent acts and the other follows [2]. Here, we studied leader-follower dynamics between players in a dyadic perceptual decision-making task. Players use a joystick to collect targets while integrating both stimulus and social information. The stimulus is the direction of a dot pattern indicating the target’s location, while social information includes real-time feedback on the angular direction and radial eccentricity of the other player’s joystick. We investigated how the information is transferred between players’ variables and from the stimulus using a multivariate form of transfer entropy (TE). We then assessed the extent to which both information sources predict the dyadic score (total targets collected by the players) by fitting three Bayesian linear regressions with predictors: (a) TE from social and stimulus information, (b) TE from social information, and (c) TE from stimulus information. We detected dyads (26%) presenting leader-follower dynamics (Fig. a) and mutual-influence (68%) interaction pattern where both players received information from each other (Fig. b). We observed that the best model (ELPD estimated with WAIC [4], Fig. c) was model (c), showing an increase in the dyadic score with TE from the stimulus ($\beta_{stimulus-player1}$ ∈ [0.245, 1.244] and $\beta_{stimulus-player2}$ ∈ [0.022, 1.149], 94%CI). We showed that TE detects two classes of interaction patterns in a dyadic setting with real-time feedback, enabling leader-follower dynamics consistent with previous studies [2]. Moreover, the dyadic score increased with the transfer of information from the stimulus to the players, suggesting that TE effectively captures the alignment of players’ direction with the stimulus for target collection. Our results support using TE to study social interactions where perceptual decisions follow the integration of stimulus and social information.