Constructionist pedagogies, emphasizing the hands-on creation of physical or virtual artifacts, have gained prominence in K-12 AI literacy education. Despite its growing importance, current empirical efforts remain fragmented across diverse settings, hindering a cohesive understanding of effective practices and theoretical foundations. This systematic review addresses this gap by synthesizing patterns, trends, and theoretical insights from 34 empirical studies conducted between 2014 and 2024. The review identifies a range of material tools used in pedagogical designs, along with constructionist practices such as experiencing, experimenting, making, and optimizing. These designs enhance students' technical knowledge, ethical awareness, and attitudes toward AI. However, challenges persist, including conceptual and ethical misunderstandings among students, as well as technical barriers like resource limitations and usability issues. Furthermore, this review identifies five core learning mechanisms in constructionist environments: (1) employing constructionist tools as cognitive tools, (2) internalizing AI knowledge through embodied cognition, (3) fostering epistemic practices, (4) exploiting generative AI as a learning partner, and (5) leveraging students' funds of knowledge. These findings provide actionable recommendations for system development, pedagogical design, assessment frameworks, and future empirical research in AI literacy education.