Despite the increasing adoption of generative artificial intelligence (GenAI) to facilitate second language (L2) writing, current reviews are insufficient in their depth and scope to effectively highlight the forefront of research trends. To bridge the gap, this paper synthesizes findings from 73 empirical studies on GenAI-assisted L2 writing based on the six dimensions of the revised technology-based learning model, including theories, GenAI, participants, objectives, methods, and outcomes. Results reveal that (1) Sociocultural, cognitive, and sociocognitive theories are most frequently used. (2) ChatGPT is the most widely used GenAI model, and "defining objectives + refining outputs" is one of the most common prompt engineering strategies. (3) GenAI primarily assists learners of English as a second language (ESL) across various proficiency levels. (4) Major objectives include learner perceptions (attitudes, perceived usefulness) and academic writing practices. (5) Researchers prefer mixed-methods and qualitative studies over quantitative ones, focusing on content, organization, grammar, and mechanics as primary writing metrics. (6) Affordances (feedback provision, content development) and challenges (learner dependency, inconsistent feedback) coexist, with studies generally reporting positive and mixed effects with moderate and large effect sizes. It highlights prospects for writer-friendly GenAI platforms and enhanced GenAI competence and pedagogy while identifying crucial research directions.