Large Language Models (LLMs), as the one used in ChatGPT, can assist in retrieving information to answer questions, generating images and audio, and supporting complex activities such as data visualization design. Some studies explore the use of these generative tools to perform certain design activities, but we have not found studies on the entire process. In general, these studies focus on generating and ideating the final product rather than on the design process itself. This work investigates the ability of these models to assist beginners in using methods and tools throughout a data visualization design process. Thus, we sought to answer the question: How can LLMs support the visualization design process? In a preliminary study with ChatGPT, we explored different strategies for building prompts to generate and justify data visualization recommendations. This study also helped us define criteria to evaluate the quality of responses from this type of model. In this dissertation, we investigated techniques and tools used in design activities, created prompts to execute these techniques with designers who are not specialists in visualization, and evaluated, with specialists, the process of building design guides for activities to understand the differences and similarities between the two. To do so, we defined a usage context and a set of criteria that allowed us to assess the interaction as well as participants perception when communicating with the model. At the end of the methodological process, it became clear that ChatGPT can support both abstract and more tangible activities at different levels. However, it is not advisable to recommend this type of model to users who do not have sufficient knowledge of the field to assess the quality of the model s responses.