Tinnitus heterogeneity is well-documented across phenotypes, etiologies, risk factors, comorbidities and associated burden. However, variability in treatment response remains insufficiently explored and often masked by the group-level comparisons of clinical studies. Moreover, little is known about the temporal trajectories of symptoms during treatment. Longitudinal monitoring via smartphones using Ecological Momentary Assessment provides rich inter- and intraindividual data on fluctuations and trajectories of symptoms. In this study, we investigated whether individual 12-week trajectories of daily self-reported tinnitus symptoms during treatment could be meaningfully sub-grouped. 147 patients provided 9634 observations while undergoing single or combined applications of hearing aids, cognitive-behavioural therapy, structured counseling, and sound therapy. A four-class growth mixture model best fit the data. One class was characterized by an increase in tinnitus symptoms over time (18%), another showed stable symptom trajectories (40%), while the remaining two classes described symptom reductions with different onsets during treatment (early improvement: 20%; late improvement: 21%). The identified classes did not differ in baseline characteristics, indicating that this information could not be used to predict symptom trajectories. Additionally, all four classes were represented in nearly each treatment arm. Notably, retrospective patient-reported outcome measures (PROMs) did not consistently align with latent symptom trajectories. These results underscore the heterogeneity and non-linearity of symptomatic change both within and across treatment modalities. We propose that app-based trajectories reveal details about symptom improvement that cannot be seen in standard PROMs.