Role-playing AI chatbots have emerged as effective tools for foreign language learning, providing engaging conversational practice in judgment-free environments. However, current iterations face a critical gap wherein they cannot provide corrective feedback without first disrupting the learner’s immersion in role-play. This study introduces Narrative Suggestion, a new implicit corrective feedback approach that embeds grammatical corrections in collaborative storytelling suggestions, keeping learners engaged in role-playing rather than shifting their focus to grammatical instruction. Through comparative evaluation against traditional implicit corrective feedback using a web application, we assessed 17 language learners across engagement, immersion, and correction visibility metrics through our three-component Activity-Integrated Feedback framework based on persuasive design principles. Results demonstrate that Narrative Suggestion significantly outperforms Conversational Recasts in engagement (24.1% more chat turns, Cohen’s \(d = 1.302\) ) and learner preferences for Activity Enhancement ( \(d = -0.61\) ) and Immersion Preservation ( \(d = -0.79\) ). Qualitative analysis also revealed proficiency level as a critical factor for Correction Visibility wherein beginner learners still preferred more explicit corrective feedback for clarity. These findings validate our framework and method, providing theoretical and practical foundations for designing future persuasive corrective feedback systems in AI-mediated language learning environments.