Background: In the field of Learning Design, it is common that researchers analyse manually design artefacts created by practitioners, using pedagogically-grounded approaches (e.g., Bloom's Taxonomy), both to understand and later to support practitioners' design practices. Automatizing these high-level pedagogically-grounded analyses would enable large-scale studies on practitioners' design practices. Such an approach would be especially useful in the context of mobile learning, where practitioners' design practices are under-explored and complex (e.g., involving both formal and informal learning activities, happening between physical and digital spaces). Objectives: We inquire about the kind of designs that practitioners create in mobile learning by analysing the entire databases of two m-learning tools, Avastusrada and Smartzoos, which promote inquiry learning outdoors. Methods: We use supervised machine learning to classify the textual content of the designs based on the cognitive level required from learners, the inquiry-based learning phases they cover, and their connection with the learning context (e.g., the role played by the situated environment). Results and Conclusions: Results from the in-the-wild studies emphasize practitioners' tendency to design contextualized activities, but that include few higher-order thinking tasks and elements of inquiry learning. This raises questions about the real-life pedagogical value of similar mobile learning tools and highlights the need for providing pedagogical guidelines and technical solutions that would promote the adoption of good learning design practices. Major takeaways from the study: While we show that machine learning techniques (informed by learning design elements) can enable large-scale studies and provide useful insights, to best understand and support practitioners' design practices it would be necessary to combine them with other quantitative and quantitative analyses (e.g., a qualitative understanding on why practitioners take specific design decisions). Future research could use similar machine learning approaches to explore other design settings, as well as explore scenarios where similar algorithms can be embedded in design tools, to guide practitioners' design practices.