Continuation writing was recently introduced as a new element of China’s English as a Foreign Language (EFL) assessment. The assessment task requires students to complete the story within a set word limit. Preparing students for this task is challenging for Chinese EFL teachers, as little pedagogical guidance is supplied. Given the recent trend of incorporating Digital Multimodal Composing (DMC) in Chinese EFL classrooms, using generative AI to provide sample writings presents a potential option.. This process could be viewed as students creating mentor texts to inform their writing development. We argue that AI-generated continuation writing samples are predominantly built with complex but culturally hollow sentences. Particularly, AI-generated samples may prioritize “native-speakerism” in which local cultures are under-represented. Integrating AI-generated continuation writing samples without cultural awareness risks unintended recolonization of EFL classrooms. We illustrate our argument with data generated via AI chatbots. We mobilize Hallidayan systemic functional linguistics (SFL) to identify patterns in complex AI-generated sample texts. This study highlights potential risks when the dominant power reproduced by the language patterns in AI-generated texts is not critically considered.