Higher-Order Thinking and English Public Speaking in an AI-Empowered Production-Oriented Approach: A Quasi-Experimental Study of Chinese EFL University Students

English public speaking in EFL education requires both linguistic competence and higher-order thinking, yet existing instruction rarely addresses these demands together. This study examined an AI-empowered Production-Oriented Approach (POA) targeting higher-order thinking (HOTS) and English public speaking performance (EPS) among Chinese EFL university students. In an explanatory sequential mixed-methods design, 160 sophomore non-English majors were assigned to an AI-empowered POA condition ( n = 80) or a multimedia-supported POA condition ( n = 80) over 16 weeks. Because each condition was a single intact class, group and class effects are confounded, so the findings are read as associational rather than causal. The experimental group used generative AI chatbots and speech evaluation tools across the three POA stages; the control group received the same instruction without AI. Mixed-design ANOVAs showed significant Group × Time interactions for HOTS and EPS, with large gain-score effect sizes that describe these two classes rather than a population (HOTS: d = 0.92; EPS: d = 1.10). Analysis and evaluation showed the strongest HOTS gains; content construction and language quality the strongest EPS gains. The interactions for creation ( p = .127) and audience engagement ( p = .084) were not significant. Qualitative analysis identified four themes: AI as a scaffold for idea generation and evidence evaluation, a rehearsal and revision tool, teacher mediation against over-reliance, and feedback overload. Findings suggest AI-empowered POA may support simultaneous development of higher-order thinking and public speaking, with teacher mediation as the central regulating mechanism.