Citation networks can reveal many important information regarding the development of science and the relationship between different areas of knowledge. Thus, many studies have analyzed the topological properties of such networks. Frequently, citation networks are created using articles acquired from a set of relevant keywords or queries. Here, we study the robustness of citation networks with regards to the keywords that were used for collecting the respective articles. A perturbation approach is proposed, in which the influence of missing keywords on the topology and community structure of citation networks is quantified. In addition, the relationship between keywords and the community structure of citation networks is studied using networks generated from a simple model. We find that, owing to its highly modular structure, the community structure of citation networks tends to be preserved even when many relevant keywords are left out. Furthermore, the proposed model can reflect the impact of missing keywords on different situations. Data includes the list of words in pickle format (readable in Python) and the citation network with metadata (as a xnet file). The xnet file can be opened by installing the xnetwork package: pip install xnetwork then use the following code to load it as a igraph network: import xnetwork as xn g = xn.xnet2igraph("<path to the xnet file>")