Self-set educational goals are central to self-regulated learning and an observable manifestation of students' motivation. In this paper, we develop and trial a tagset for the characteristics of university students' self-set goals. The novelty of this approach consists in using data-driven, non-exclusive tags, rather than theory-derived, exclusive categories, on the basis of freely formulated goals in natural language. A 400-goal sample out of the 2,262 educational goals collected from 732 students at three universities was used to develop a tagset of six metatags and 28 tags. Six coders independently assigned tags to the collected goals. Based on these tag assignments (n = 376,458), Krippendorff's [alpha] was used to approximate intercoder reliability. Surprising intercoder agreement scores are discussed. Further, relative frequencies for each of the tags were calculated. These point to relevant aspects of students' motivations. The tagged dataset may serve as input for AI-powered study assistant systems, whilst the tagset itself may be used in future studies to gain further insights into university students' study motivations.