This paper provides a comprehensive survey of text-to-3D shape generation methods, categorizing them into three main families based on the use of 3D and text data. The first family uses paired text and 3D data (3DPT), the second family uses unpaired 3D data (3DUT), and the third family does not rely on 3D data (No3D). The survey highlights recent advancements in 3D representations, large-scale pretraining, and differentiable rendering, which have driven progress in this field. However, challenges remain, including the scarcity of paired 3D and text data, the inability to edit generated outputs intuitively, and the complexity of training high-quality 3D models without explicit 3D data. The paper discusses the limitations of existing methods and outlines promising future directions, focusing on the third family of methods that do not require 3D data. It also explores hybrid approaches that combine text-to-image models with 3D representations to enhance the quality of generated 3D shapes.This paper provides a comprehensive survey of text-to-3D shape generation methods, categorizing them into three main families based on the use of 3D and text data. The first family uses paired text and 3D data (3DPT), the second family uses unpaired 3D data (3DUT), and the third family does not rely on 3D data (No3D). The survey highlights recent advancements in 3D representations, large-scale pretraining, and differentiable rendering, which have driven progress in this field. However, challenges remain, including the scarcity of paired 3D and text data, the inability to edit generated outputs intuitively, and the complexity of training high-quality 3D models without explicit 3D data. The paper discusses the limitations of existing methods and outlines promising future directions, focusing on the third family of methods that do not require 3D data. It also explores hybrid approaches that combine text-to-image models with 3D representations to enhance the quality of generated 3D shapes.