Caregivers vary in how often and in what ways they engage verbally with their child, and this variation is associated with positive child outcomes (Hart & Risley, 1995). Does this variation reflect the activities (e.g., play, book sharing) in which caregivers and children engage? Or, is variation in verbal engagement a stable feature of children’s learning environments regardless of what activities occur? We sampled daylong LENA audio-recordings in English- (n = 45) and Spanish-speaking (n = 45) families with 24-month-old children in the U.S., transcribing the six densest 10-min segments of speech directed to target children (tCDS) per family. On average, caregivers spent ~50% of time engaged in child-centered activities (e.g., book sharing, play), ~20% in adult-centered activities (e.g., cooking), and ~30% not engaging verbally with the target child (non-tCDS). Caregivers were more likely to imitate/expand on their children’s utterances during all child- than adult-centered activities. Book sharing, occurring in only ~50% of families, was associated with longer utterances (MLUw) and more responses than adult-centered activities. Children who experienced more minutes in child-centered activities, especially book sharing, heard significantly more total words than children who experienced fewer minutes in child-centered activities. In families without book sharing, time in other child- and adult-centered activities was associated with more total words. Adult-centered activities tended to be more lexically diverse than other child-centered activities. Evidence for moderate stability within families was found in word tokens and mean utterance length. These results highlight that variation in features of child-directed speech derives from multiple sources.