In this study, we examined observable strategic regulation and perceived learner ownership across two ChatGPT interaction protocols during repeated English as a foreign language (EFL) reading and writing tasks. Twenty low-intermediate university students in two pre-existing General English I sections participated in a six-session, qualitatively driven mixed-methods study with a nonrandomized comparison. The metacognitive-partner protocol prompted planning, monitoring, and evaluating, while the answer-oriented protocol allowed standard ChatGPT assistance. Learner–AI transcripts and learner reflections provided the primary evidence, while quantitative measures were summarized descriptively. In these classes, metacognitive-partner transcripts contained more learner-articulated planning at task initiation and more visible monitoring and evaluation than answer-oriented transcripts. Learners in the metacognitive-partner condition also more often reported ownership of AI-supported decisions and products. Because the prompts invited several of the coded moves and the intact classes were not randomly assigned, these patterns describe interaction within the two protocols and learners’ reported perceptions. They do not establish a causal effect, independent strategy acquisition, verified control over submitted work, or durable transfer.