(1) Background: In the era of generative AI (GenAI), assessing AI literacy is essential for understanding how effectively non-expert users can interact with AI. However, existing assessment tools primarily focus on users’ understanding of AI principles or rely on self-reported scales, neglecting critical thinking and actual interaction capabilities. To address this gap, this study aims to design and validate evaluation indicators targeting the behavioral process of human–GenAI interactions and analyze the impact of critical thinking. (2) Methods: Grounded in information literacy and critical thinking frameworks, this study operationalized human–AI interaction capabilities into behavioral indicators and rubrics through observation, surveys, and pilot studies. Data were collected from 121 undergraduates completing two real-world tasks with GenAI, and their interaction processes were documented and evaluated. (3) Results: The indicators showed acceptable inter-rater and internal consistency reliability. Exploratory and Confirmatory Factor Analysis confirmed a three-dimensional structure. Further analysis showed that interaction capabilities varied across gender, academic background, AIGC use frequency, critical thinking disposition levels, and question chain logic. (4) Conclusions: The developed evaluation indicators are reliable and valid. Further analysis reveals that a high critical thinking disposition can offset the disadvantage of lower usage frequency. This highlights the significance of critical thinking in enhancing human–GenAI interaction capabilities.