We investigate the usefulness of text generation AI, in constructing categorical databases using ontology logs. Database design often struggles with complex relationships between tables, especially in large-scale systems. Categorical databases apply principles of category theory to ensure data consistency by defining objects (tables) and morphisms (foreign keys) under strict rules. An ontology log is a method to represent concepts and their relationships using both diagrams and natural language, helping clarify database schemas. We used ChatGPT to generate natural language sentences that express composite relationships between tables. For example, a sequence of foreign keys like “Grades → Student → Department” can be composed into a sentence such as “Grades in an educational outcome of Department.” ChatGPT produced multiple candidate sentences for each relationship, which were then evaluated for consistency and accuracy by database designers. Our results indicate that text generation AI can quickly produce diverse and coherent descriptions, thereby supporting the efficient design of categorical databases. We also expect that further development of this approach may allow automatic extraction of business processes from requirement documents and the proposal of suitable database structures for effective data management and analysis.