This master’s thesis examines how a structured use of artificial intelligence (AI) can support English as a Second Language (ESL) writing in secondary education without encouraging metacognitive laziness. The project grew out of classroom practice and explores how guided interaction with ChatGPT can help students think about their own thinking while writing. Grounded in theories of metacognition, cognitive offloading, dialogic feedback, and desirable difficulties, the study implements a three-step writing process: students first handwrite their drafts, then type and expand them, and finally use ChatGPT via Fobizz to receive feedback that they must accept, adapt, or reject with justification. Successive drafts, short reflection notes, and open-ended questionnaires provide insight into how students plan, monitor, and evaluate their writing. The results show that when AI feedback is embedded within a teacher-structured framework that demands explanation and accountability, learners engage more deeply with their own revision process. They become more reflective, clearer in expression, and more confident in making choices about their texts. The thesis concludes with practical implications for classroom use: prompts should invite reasoning, not simple correction; AI feedback should remain dialogic and reflective; and writing tasks should align with Lehrplan 21 (curriculum) competencies on strategy use and self-regulation. Overall, the study offers a concrete example of how AI can be integrated responsibly into school-based ESL writing to strengthen, not replace, students’ metacognitive awareness.