Chatbots are conversational agents designed to engage in text-based conversations with end-users. They have been the subject of several scientific and experimental researches in universities as well as in the industry since the emergence of artificial intelligence techniques and natural language processing. Obtaining the right example of training data is crucial for the creation of chatbots. Most chatbots require the collection of intentions as well as expressions in human language. This is generally a manual and tedious process and hence the need for automatic learning data volume generation tools to build robust chatbots. One of the solutions that exists is the increase of data in natural language. This article aims to study the impact of increased data on chatbot performance. To do this, we carried out experiments on real data. This is the human resources data used to build a generic chatbot within our company Alcatel-Lucent Enterprise (ALE).