Introduction: Text-message interventions with universal decision rules (treatment algorithms) demonstrate modest and inconsistent effects on physical activity. Efforts to increase effect sizes by targeting, tailoring or generically customizing content have largely been unsuccessful. In this study, system identification methods from control systems engineering were applied to develop person-specific dynamical models of individual responses to motivational text messages. Methods: Emerging adults not meeting aerobic physical activity guidelines (verified by accelerometer during a 1-week run-in period) received 0-6 messages/day while wearing a Fitbit Versa watch for 3 months. Experimental messages targeted cognitive or affective determinants of physical activity (move more) or sedentary behavior (sit less); comparator messages were inspirational quotes. For each participant, two dynamical models were estimated using difference equations: one to model the effects of move more and sit less messages and the other to model the effects of affectively- and cognitively-targeted messages. Both models included comparator messages. Effects represent the expected change in behavior immediately after message receipt (vs expected behavior without a message). Model order was determined individually to balance uncertainty and overfitting (range = 2-9). Results: The sample (n = 20) was mostly female (60%), White (65%), and not Hispanic or Latino (95%) with a mean age of 24.4 years (SD = 3.3; range = 19-29). Idiographic patterns of behavior change were readily observed from the personalized dynamical models. Most participants’ step counts increased after messages to move more (60% of participants’ responses exceeded the 95% error interval [responders], M = 94.3 steps; 83% of responders increased steps) and sit less (50% responders, M = 114.6 steps; 90% of responders increased steps), as well as control messages (75% responders, M = 59.5 steps; 60% of responders increased steps). Step counts also increased after messages targeting cognitive determinants (80% responders, M = 97.1 steps; 75% of responders increased steps) and affective determinants (75% responders, M = 52.9 steps, 67% of responders increased steps), as well as control messages (75% responders, M = 57.8 steps; 60% of responders increased steps). Conclusions: Motivational text messages can alter physical activity dynamics, and responses to different message types are highly personalized. Among responders, most message types increased expected step counts but some messages were iatrogenic and decreased expected step counts. Computational models of those dynamics provide a foundation for personalizing decision rules to select the type and time the delivery of messages to promote physical activity and improve cardiovascular health.