Reclaiming the Writing Process: A CEFR-Aligned, Scaffold-First Model for AI in Foreign Language Writing

The problem: support that takes over

The chapter examines the risks of using open-ended generative tools to produce complete texts without a clear instructional structure. A polished result can obscure whose decisions shaped the writing and what the learner actually practised. The proposed alternative places the pedagogical task before the AI interaction.

A scaffold-first architecture

PrepTalk organises writing into teacher-authored stages grounded in Common European Framework of Reference for Languages (CEFR) descriptors. Each stage has a defined purpose and elicits specific learner input. AI operates within those boundaries, providing targeted clarification, hints and validation support rather than generating a complete draft or rewriting the learner’s text.

The distinction is architectural: learner authorship is supported through explicit task goals, progression rules and limits on what AI is permitted to do. The abstract frames those constraints as part of responsible design, not an optional instruction added to an otherwise open-ended chatbot.

From task design to a visible writing process

The chapter connects the pedagogical rationale to PrepTalk’s interface, including stage-based guidance and the notepad where learner work accumulates. It explains how CEFR alignment informs the writing task and how bounded assistance preserves the purpose of each stage.

A foundation for empirical research

The chapter also outlines structured logging of stage progression and learner interactions. This creates a basis for investigating how scaffolds are used and for comparing different approaches to AI-supported writing.

It presents a design framework, not evidence of improved learning outcomes. Empirical questions about effectiveness, learner behaviour and the influence of different forms of support remain distinct from the framework itself.