Accelerators and startups acceleration programs are increasing rapidly in the world. An important task in the acceleration program manager’s routine is select promising startup to compose each acceleration batch. Meanwhile, Artificial intelligence and large language models (LLMs) are changing process and creating exhibit remarkable capabilities across a variety of domains and tasks. This study investigates the effectiveness of GPT-4 in selecting startups for a Brazilian vertical health accelerator, aiming to compare the selection outcomes of humans and the AI model to determine GPT-4’s utility in the acceleration program selection process employing a non-randomized controlled trial method. The main results showed high performance levels of the GPT-4 model in accurately predicting suitable startups for acceleration, with 86.31% of accuracy and others notable precision, recall, and specificity rates. This suggests that GPT-4 can significantly aid startup accelerator managers in the selection process, enhancing the efficiency and effectiveness of identifying startups with high potential for the batches.