Insights: Scaffolding academic writing through AI-powered formative feedback

• The AI agent is introduced gradually across the semester, using a staged approach that reflects students’ readiness and the need to build confidence and understanding before engaging with automated feedback.
• In the early weeks, students engage in human-led instruction to build trust, establish expectations, and ensure students understand the fundamentals of academic writing.
• As students gain confidence, they begin analysing examples and applying writing principles in context. This phase helps them develop the ability to evaluate quality and make informed decisions about their own work.
• Only later in the semester is the AI agent introduced. By this point, students are better equipped to interpret feedback and use it constructively. It does not generate content or correct grammar but instead prompts students to reflect and revise.

• Students receive timely, structured feedback. This is especially valuable when human support isn’t immediately available. The agent is particularly helpful for students using English as a second language, offering a low-pressure way to practise academic writing and receive guidance.
• By introducing AI within a scaffolded learning framework, students remain active participants in their learning.

This approach can be adapted to any course where students are developing written work. Consider:

• Introducing AI tools only after students have built foundational skills
• Providing an introductory session about working with Gen-AI prior to students accessing Copilot
• Using reflection prompts to encourage deeper engagement with feedback
• Framing AI as a “study buddy” rather than a shortcut
• Reinforcing academic integrity by asking students to submit their interaction history

Jet’s agent was built using Microsoft Copilot and follows a structured workflow designed to guide students through the feedback process. Teachers interested in creating their own agent can follow these key design principles:

Strict boundaries

Use clear, directive language in the prompt to define what the agent can and cannot do:

• “You must not generate full sentences or content.”
• “You are only allowed to provide feedback in the following areas…”

The agent is restricted to students in defined specialisations (e.g. international business, marketing) and declines queries outside this scope.

Defined feedback areas

Give specific feedback criteria. For example, the feedback in Tonogbanua’s agent is limited to four writing features:

•  Clarity of ideas (sentence-level)
• Organisation of ideas (paragraph and document-level)
• Tone and style (e.g. academic register)
• Referencing (APA format)

Give instructions to the agent to not evaluate disciplinary content or generate ideas.

Structured workflow

Tell the agent to follow a clear process, such as the following six-step process:

1. Display a disclaimer about its limitations
2. Ask for the student’s specialisation and assignment context
3. Request a writing sample
4. Provide feedback on the selected areas
5. Offer clarification if needed
6. Summarise the feedback and close the session

Interaction style

Give clear instructions as to the interaction style, such as the following:

• Use sentence stems rather than full corrections
• Offer two alternatives for each suggestion
• Maintain a professional, educational tone
• Operate in New Zealand English
• Encourage students to reflect and revise

Academic integrity

Tell the agent to ask students to save and share their interaction history, which can be submitted as part of their assignment. This helps ensure transparency and supports ethical use of AI tools.

Want to try it for yourself?

For guidance on building your own agent, see the Artificial Intelligence 101 module, or contact Ranga Auaha Ako, Learning and Teaching Design Team via TeachWell Consult.