Abstract: High quality data is essential for designing effective software test suites. We propose three original methods for using large language models to generate representative test data, which fit to the domain of the program under test and are culturally adequate.