Large Language Model (LLM)-Based Automation for Software Test Script Generation

Abstract

With the rapid evolution of software systems and the growing complexity of their testing needs, test automation has become more essential than ever within the software development life cycle (SDLC). Yet, despite advances in tooling, the process of creating test scripts still tends to be manual, time-consuming, and highly dependent on both programming skills and domain expertise. In this study, we explore a practical approach that uses Large Language Models (LLMs) including GPT-based models to automate the generation of test scripts directly from natural language software requirements or user stories. The method combines prompt engineering with lightweight fine-tuning and the use of domain-specific data to generate executable test cases. We applied this framework across a range of real-world software projects and used popular testing tools like Selenium and PyTest to assess performance in terms of script accuracy, maintainability, and the overall reduction in developer workload. Our findings show that LLM-generated tests significantly reduce manual effort and produce results that are close to human-written scripts in terms of coverage and correctness. Overall, the approach offers a flexible solution for integrating AI-powered test generation into CI/CD pipelines and could mark a step forward in how teams approach software testing in agile environments.