Humans are constantly in relation with other humans, frequently using heuristics to efficiently navigate an increasingly complex social environment. One type of heuristic used are stereotypes, usually applied to other social groups, but also to one's own group, and thereby to the self. Stereotypes can, but don't have to be negative, and can be learned from the social environment based on statistical contingencies between social group concepts and a stereotype-compatible characterization. However, in the case of a negative stereotype, this is potentially in conflict with the general tendency to see the self in a positive light, the so-called self-positivity bias. Here, we use a modified minimal-group paradigm to induce participants to feel linked to an avatar, and thus member of one of two social groups present in the experiment. Then, we expose participants to stereotype information which is positive for the out-group, but negative for the self-relevant in-group. Based on data collected in a small pilot study, we simulated the experimental design both for feasibility overall and to estimate the number of items and participants necessary in particular. Then, we conducted a lab-based study with human participants (N=80). Participants showed allegiance to their self-relevant group and a self-positivity bias before learning that it is associated with a negative description. After having acquired this information, participants switched their group alliance to the positively described social group. In addition, they judged their experimentally intended in-group as more negatively than the (intended) out-group. This result supports that preserving a positive self-image has significant influence on task-performance in an experimental setting. This stresses the importance of context in experimental, but also observational data acquired in a more ecological setting, where conclusions drawn from the data need to be interpreted in relation to the situation of data collection.