Since its release, ChatGPT has raised concerns in many teaching contexts given its threat to reliably evaluating learners' knowledge and skills. Within task-based pedagogy, however, this technology opens new avenues for second language (L2) teaching when adopting the technology as a writing buddy. Our study explores how ChatGPT as a model impacts the revision process of advanced L2 writers of German. Twenty-two university students participated in a two-week classroom-based intervention, producing two summaries of popular-scientific texts in L2 German. After writing a first draft, they compared their summaries with texts produced by ChatGPT (3.5) and revised, where necessary, their own text. In this paper, we analyze all students' rubrics-based ratings of the ChatGPT models and present data of six focus students' screen-recorded revision processes that we coded for revision focus, source, and success. Results reveal students' growing awareness of characteristics of ChatGPT-output, such as linguistic accuracy and fluency, as well as its flaws in content provision. Revision data demonstrate that students skillfully made use of the models to improve their own texts. Our study provides evidence that using ChatGPT as models in writing and revision processes can stimulate higher-order thinking in the revision process of advanced L2 students.