<p>The emergence of generative artificial intelligence (GenAI) tools in second and foreign language (L2) contexts has introduced new possibilities for second language acquisition (SLA). While preliminary studies have highlighted the potential of GenAI technologies, there remains a notable gap in empirical evidence concerning how such technologies influence the cognitive and affective aspects of L2 learning. Moreover, most existing studies have relied on cross-sectional designs, overlooking the evolving nature of L2 learners’ cognitive and affective development. To bridge these gaps, this longitudinal study harnessed the latent growth curve modeling (LGCM) approach to track the sustained effects of Gemini on L2 learners’ writing proficiency and anxiety. The study involved 274 undergraduate English learners enrolled in four academic writing classes at a public university in Iran. Participants were randomly assigned to either a treatment group (<i>n</i> = 151), which received GenAI-assisted instruction using Gemini, or a control group (<i>n</i> = 123), which received traditional instruction, over a 16-week course. The writing proficiency of the participants was measured using a standardized rubric-based assessment, and their writing anxiety was assessed through an L2 writing anxiety scale. The LGCM outcomes revealed significant growth in the writing proficiency of the treatment group (<i>MD</i> = 4.49, <i>SE</i> = 0.181, <i>CR</i> = 24.84, <i>p</i> &lt; .001), alongside a marked decrease in their writing anxiety (<i>MD</i> = –5.42, <i>SE</i> = 0.114, <i>CR</i> = 47.48, <i>p</i> &lt; .001). Together, the findings underscore the pedagogical value of integrating GenAI tools like Gemini into L2 writing instruction, as they offer tailored, immediate feedback that enhances learners’ proficiency while alleviating their writing-related anxiety.</p>