Online assessment is an essential part of online education, and if conducted properly, has been found to effectively gauge student learning. Generally, textbased questions have been the cornerstone of online assessment. Recently, however, the emergence of generative artificial intelligence has added a significant challenge to the integrity of online assessments. In particular, it has been reported that large language models, like ChatGPT-4o, show high performance on textbased questions. In comparison, ChatGPT-4o exhibited significantly reduced performance on figure-based questions in our study. In an effort to counter the recent encroachment of ChatGPT-4o into online assessment, we propose a step-by-step instructional guide for a method in creating figure-based multiple-choice questions that are resistant to ChatGPT-4o. This involves generation of a ChatGPT-4o-resistant figure, writing the question text based on it, and evaluating the final question on ChatGPT-4o. If successfully created, ChatGPT-4o response could be subject to random guessing. Our results showcase four representative examples for introductory biology courses and illustrate a systematic approach to compose questions based on qualitative analysis of ChatGPT-4o responses. In combination with other assessment methods, our method aims to serve as a tool to alleviate the current challenge that educators face for online assessments.