Children’s early vocabulary predicts later reading, academic attainment, and adult earnings, yet low-income parents talk with their preschool-aged children substantially less than higher-income parents do. We evaluate Chat2Learn, a six-month text-message intervention designed to help parents discuss specific words with their children. We randomized 730 low-income parents in the United States into a control group and two treatment arms. The definition arm provides a word, its meaning, and a brief script. The conversation arm provides the same word but asks parents to use it in open-ended talk and pretend play. The definition and conversation arms increased total vocabulary scores by 0.37 and 0.23 standard deviations, respectively, and the two estimates are not statistically distinguishable. Average gains were concentrated on the words featured in the intervention. Conditional quantile estimates for treated words are positive and comparatively similar across the distribution. For untreated words, the conversation arm shows no consistent pattern, whereas the definition arm produces positive estimates at the two highest deciles, consistent with a modest upper-tail shift. In the main specification, the definition arm is associated with an approximately five-percentage-point reduction in parents’ fixed-mindset beliefs from a control-group mean of 18 percent, although this estimate becomes marginally significant in some specifications. We also developed exploratory measures of children’s curiosity; neither arm produced a robust effect. We detect no statistically significant effects on parental stress, fatigue, or enjoyment of learning activities.