This systematic review investigates the measurable effects of machine translation (MT) on second language (L2) writing performance. It asks how MT is used in EFL/ESL writing instruction and revision, what effects have been reported, and how learner proficiency and instructional guidance shape these effects. Scopus was searched for peer-reviewed studies published in 2022–2026, and a procedure adapted from PRISMA was followed. Fourteen studies met the criteria. Studies were coded, appraised for quality, and counted by outcome. MT served as free access, a revision tool, feedback, a base for post-editing instruction, and a comparison with other tools. Nine of the 10 studies that measured accuracy reported better accuracy. Four of the six studies that measured fluency reported gains. Only four of nine studies reported gains in syntactic complexity, and each gain had limits. Lexical results were mixed to positive (five of nine studies reported clear gains), but only one study tested vocabulary directly. In the studies that tested this, gains made with MT fell when MT was removed, but stayed above pretest levels in the clearest case. Studies that added explicit instruction, such as post-editing training, reported gains mainly in accuracy. Guidance seems to shape the effect of MT, but no study clearly isolated guided from unguided MT use. Only three studies compared proficiency levels, and effects were not simple. Lower-proficiency learners gained most in syntactic complexity in one study, while higher-proficiency learners kept vocabulary gains longer in another. Study quality was high in two, moderate in five, and low in seven. The evidence shows short-term writing performance more than long-term language development. Therefore, MT should be taught with guidance and assessed with and without MT. Future studies need longitudinal designs with delayed tests without MT, guided versus unguided comparisons, and comparisons across proficiency levels and contexts.