Background Patients with Alzheimer’s disease (AD) often show problems in spatial navigation. However, spatial navigation is rarely investigated in real world settings and in people at early disease stages. Capitalizing on recent advancements in digital technologies, we measured the performance of healthy younger (YA) and older adults (OA), as well as older adults with subjective cognitive decline (SCD) while they solved a smartphone‐based wayfinding task. Method Using the novel smartphone application “Explore”, we tracked GPS data and other behavioral indicators in 24 YAs, 25 OAs, and 23 SCDs, while they were asked to find points of interest on the medical campus in Magdeburg. In our analyses, we first quantified similarities between the GPS trajectories and identified wayfinding styles using k‐medoids clustering. Next, we extracted navigation profiles with aggregated performance measures as input features (e.g., wayfinding distance and number of help function calls) in a latent profile analysis. Mixed effect models were fitted to evaluate group differences for each performance measure separately. We then fed the performance measure that showed the largest groups differences into a multinomial logistic regression model to predict age and diagnostic group in unknown users via leave‐one‐out cross‐validation. Result The GPS data could be clustered into three wayfinding styles. However, the correspondence to our groups was rather low (Figure 1a). In contrast, navigation profiles obtained from the latent profile analysis showed a higher correspondence (Figure 1b). Particularly the number of brief stops during navigation differed between the groups, all p ≤ .006 (Figure 2), and predicted age group, p = .015, and diagnostic group above chance, p = .004 (Figure 3). The leave‐one‐out cross‐validation confirmed the successful prediction of group membership in unknown participants, accuracy = 0.57 (chance level 0.33). Conclusion Here, we show how information about the cognitive health status of an individual can be inferred from smartphone‐data, obtained during a brief episode of a frequently performed everyday behavior. In particular, the number of short stops during wayfinding could be used to distinguish between YA, OA, and SCDs. Hence, real‐world navigation data, obtained from mobile devices, might help to identify individuals at increased risk for developing AD.