Learning design guide · PrepTalk team
Scaffold-first AI learning: a practical guide
In PrepTalk, scaffold-first AI learning means designing the learner’s task and contribution before deciding how AI should help. The activity gives learners a structure for understanding, practicing, writing and revising. AI feedback supports their next step.
See how PrepTalk applies itWhat does “scaffold-first” mean?
Scaffolding is support for doing something a learner cannot yet manage independently. It can include a prompt, a worked example, language to practice or a smaller step within a larger task. In their foundational tutoring study, Wood, Bruner and Ross examined support that let children tackle a construction task beyond their unaided efforts.
“Scaffold-first” describes the design approach used in PrepTalk’s teaching model: define what the learner needs to practice, then design assistance around that contribution.
How can you recognize it in an AI learning app?
Inspect a complete activity, including what happens when an answer is incomplete. A useful product description should let you answer these questions:
What is the learner practicing?
Look for a specific task and a clear contribution from the learner, such as choosing a position and supporting it with a reason.
What does AI contribute?
Check whether it identifies a problem, asks a useful question, supplies an example or writes the response. These forms of help create different opportunities for practice.
What happens after feedback?
Look for a chance to revise and try again, and a clear explanation of when the activity moves forward.
Whose words form the final text?
Follow the writing from first attempt to finished product. Check which wording came from the learner and which was supplied by the tool.
What happens without the support?
Ask what evidence exists about later independent performance. Completion during an assisted activity does not establish that learning has transferred.
A concrete example: replying to a forum post
Suppose a learner is replying to a post about the flexibility of remote work. The activity first establishes what the writer means. The learner practices expressions for responding, chooses how to respond, and writes their own reply in parts.
If the learner writes “I agree that remote work is good,” a hint can ask which benefit the original writer described. The learner then revises the sentence. Accepted writing collects alongside useful phrases in the notepad, ready for the final edit.
The support is attached to the task: understand a position, respond to it and explain the response. The full walkthrough shows how the parts fit together.
Which app uses this approach?
PrepTalk is being developed specifically around a scaffold-first model for foreign-language writing. It combines teacher-designed activities, useful-phrase practice, targeted AI hints, a notepad of learner writing and a final editor. Its pilot scope is English and French at CEFR B1 and B2.
PrepTalk is currently in development for a controlled pilot, with public illustrations and a written walkthrough. It does not yet offer open self-service learner access. If you need an app you can use today, check availability as well as the teaching approach.
Other tutoring tools may also provide scaffolding without using this particular term. Use the questions above to examine how each activity distributes the work between learner and AI.
How is it different from asking a chatbot for help?
An open-ended chatbot can provide explanations, examples or feedback depending on the request and its configuration. A scaffold-first activity makes the intended task, required learner contribution and progression part of the product design. Evaluate the actual behavior rather than relying on either label.
| Question to ask | What PrepTalk is designed to do |
|---|---|
| Who defines the activity? | A teacher-authored task specifies the purpose, content and ordered stages. |
| What does the learner supply? | Their own choices, ideas and written components. |
| How is difficulty handled? | Targeted hints and opportunities to revise the response. |
| What is saved as the text develops? | Accepted learner writing and useful phrases in the notepad. |
| How does the activity finish? | The learner reviews and edits the complete text. |
Does scaffold-first AI improve learning?
That requires evidence. A plausible teaching model, successful task completion and a polished final text are different from improved independent writing. Studies need to examine what learners can do later, including when the assistance is removed.
PrepTalk’s research pages explain the design rationale and proposed evaluation questions. They do not report established learning gains from PrepTalk. Ana Rosso’s forthcoming Routledge chapter presents the model; its forthcoming status should not be read as a completed evaluation.