Introduction: Despite the assumed importance of school-focused possible identities for academic motivation and outcomes, interventions rarely assess the effect of the intervention on possible identities. This may be due to difficulty coding open-ended text at scale but leaves open a number of questions: 1) how do school-focused possible identities change over the course of the school year, 2) whether these changes are associated with changes in school outcomes, and 3) whether a machine coding approach is viable. Methods: In Study 1 (n = 247 Chicago 8th-graders) we assess fall-to-spring change in school-focused possible identities. We test whether a change in school-focused possible identities predicts 8th-grade academic outcomes. We include robustness checks. Then we examine school context effects. In Study 2 (n = 1006 Chicago 8th-graders) we address the problem of coding at scale, using a separate data set to train a machine-learning algorithm. Results: On average, school-focused possible identities decline over the school year. But nearly a third of students have increasing school-focused possible identity scores. This increase is associated with improved grades. School context influences whether linked strategies matter. Our machine-learning algorithm accurately classifies school-focused possible identities in our original sample and this school-focused classification reliably predicts academic trajectories. Conclusions: Change in school-focused possible identities is normative over the course of the school year, interventions should take this into account. On average, students have fewer school-focused possible identities by spring. This decline is associated with declining academic trajectories. However, when school-focused possible identities increase, so do grades. Whether strategies matter is context-dependent.