The use of written or graphic representations is essential in mathematics. Graphic representations are mainly used and researched as instruments for problem solving. There is a gap in research for interventions that use learner-generated graphic representations as documents for reflection processes for promoting the development of children's graphic representation competences. This is the focus of the study presented here. The study examines to what extent such an intervention has an effect on how the children take into account a mathematical structure in their self-generated graphic representations, how they ensure a mathematical matching with the word problem, and what degree of abstraction they choose. Additionally, the effect on the solution rates is investigated. The results show that children in the intervention group more frequently pay attention to a mathematically appropriate structure, compared with children in the control groups. This result is statistically significant. At the same time, children keep the degree of abstraction relatively constant. Solution rates improve continuously, but the difference is not significant.