← Research at PrepTalk

Research explainer

How can AI support writing without writing for the learner?

In PrepTalk’s scaffold-first model, AI supports a defined part of the writing task. The learner still supplies the ideas, makes the revisions and produces the text.

Start with the task, not a blank prompt

A writing activity begins with a communicative purpose and a sequence of teacher-authored stages. Those stages determine what the learner needs to understand, practise or compose. AI assistance is bounded by that structure rather than deciding what the activity should become.

A hint should leave a decision to the learner

Consider a learner who writes that remote work is good because people like it. Targeted support can ask what makes remote work useful and prompt the learner to develop a more specific reason. The learner chooses the reason and writes the revision. Supplying a polished replacement sentence would change the nature of that task.

This is the distinction illustrated in PrepTalk’s guidance demonstration: the hint identifies what still needs attention; it does not supply the learner’s final wording.

Keep authorship visible

Successful learner contributions become part of the notepad, where the text takes shape across stages. The task structure, support and learner response have separate roles. That separation makes it possible to describe what the system helped with without treating the system as the author.

Preserving this division is a design commitment. Whether a particular scaffold improves learning is a further empirical question, not something established by the interface alone.

This explainer describes PrepTalk’s design and research rationale. For the conceptual framework, read the overview of Ana Rosso’s forthcoming chapter.