Fuzzy Adaptive Resonance Theory (ART) is an unsupervised clustering algorithm that is not as vulnerable to the same weaknesses as other approaches such as backpropagation neural networks. The Fuzzy ART algorithm uses a choice function as the main mechanism for clustering. In order for the Fuzzy ART algorithm to be effective, the choice function needs to be adjusted per dataset. Previous work has show the effectiveness of using genetic programming and grammatical evolution to evolve the choice function rather than deriving manually. However, these studies have been done independently and there has been no work comparing the performance of both these approaches. This paper compares GP and GE for evolving choice functions for Fuzzy ART on various text clustering datasets including four benchmark sets (Enron, SMS Spam, IMDB and Amazon) and a real-world data set (ChatGPT tweets). The results show that the GE and GP evolved choice functions perform far better than the standard choice function. The GE evolved choice functions are better than GP evolved choice functions for the ChatGPT, IMDB and Amazon datasets. The GE and GP evolved choice function perform equally well on the Enron and SMS Spam datasets. Additionally the GE takes less time to evolve a choice function than GP. No other unsupervised learning technique has been applied to the ChatGPT dataset in the literature before, making Fuzzy ART using a GE evolved choice function the current state-of-the-art.