<p>This exploratory classroom study examined the feasibility and uptake of a structured Predict–Observe–Explain (POE)–Socratic ChatGPT routine in Jordanian Tawjihi physics (Grades 11–12). ChatGPT was deliberately restricted to a questioning role: groups used it to receive prompts for justification, evidence checking, and explanation refinement rather than final answers. Across four regular lessons on mechanics and electric circuits, 100 students from four intact sections completed post-session reaction surveys and formative concept quizzes, while group-process indicators were recorded through a structured classroom rubric. The results indicate generally positive student reactions, increasing learning and engagement ratings across sessions, higher formative quiz scores across sessions, and stronger observed evidence use and conceptual coherence over repeated enactments. The study contributes an implementation-focused model for responsible generative artificial intelligence (AI) use in secondary physics by integrating conceptual-change sequencing, Socratic dialogue, and teacher oversight in a high-stakes examination context. Because the study used intact classes without a comparison group, the findings should be interpreted as evidence of feasibility, student uptake, and within-classroom associations rather than causal effects.</p>