Session 04 · Senior School

Metacognition, feedbackand workload.

When students should stop using AI, the kind of feedback only humans can offer, and a follow-along build of one Gemini Gem from your own materials.

Duration 60 minutes (×2)
Format Presentation · live build
Take-home One Gemini Gem

The metacognition problem

The question is not whether to use AI. It's when to stop.

GTT 4.6 — activating self-regulated learners — asks teachers to help students plan, regulate and monitor their own learning. The expertise reversal effect (Sweller et al., 2019): scaffolding that helps novices becomes counterproductive for experts.

AI scaffolding never comes down. It's always there.

The skill is recognising the moment scaffolding stops being helpful and becomes a substitute for thinking.

The four feedback levels

Recognitive vs. extra-recognitive feedback.

Hattie & Timperley (2007) identified four levels. AI feedback overwhelmingly operates at the bottom two — task and self.

Task — "fix this error." Surface corrections. AI handles this.
Process — "try testing against a counter-example." Requires expert judgement.
Self-regulation — "reread and ask if a reader would follow." Requires knowing the student.
Self — "you're a good student." Least effective.

Corbin, Tai & Flenady (2025): recognitive vs. extra-recognitive feedback. Recognitive feedback is grounded in mutual recognition; both teacher and student are vulnerable to judgement, and the trust that develops is what makes feedback transformative. Extra-recognitive feedback addresses the work, not the person — technically accurate, relationally empty. AI feedback is structurally extra-recognitive.

The move: let AI handle task-level feedback so your time goes to the recognitive conversations only humans can have.

Student feedback-seeking

Prompting for guidance, not for answers.

Zhan, Boud, Dawson & Yan (2025): the gap between low- and high-feedback-literacy students using the same AI tool.

Low feedback literacy

"Explain photosynthesis to me."

Vague prompt → generic response → passive acceptance.

High feedback literacy

"I think the light-dependent reactions produce ATP and NADPH, but I'm not sure how the electron transport chain fits in. Can you ask me questions to test whether I understand the sequence?"

Specific prompt against criteria → AI as Socratic partner → monitored revision.

The second prompt is teachable. It is also assessable — the prompt itself becomes evidence of metacognitive engagement.

Designing the assessment

The AIAS for Senior School — cycle, don't pick.

Cycle through levels rather than pick one. Students can't game a process that moves in and out of AI; each phase generates evidence of authentic engagement.

01
No AI. Gather original ideas, build foundational knowledge.
02
AI-assisted research. Broaden scope with AI; authenticate with discussion.
03
Independent drafting. Write without AI to establish a baseline.
04
AI as revision partner. Refine, extend, challenge. The student directs.
05
Face-to-face checkpoint. Defend, discuss, demonstrate understanding without technology.

For IB contexts: declaration and authentication happen at every stage, not just at submission. Detection is a dead end — design is the answer.

The follow-along workflow

From messy inputs to a system that knows your context.

Process-based thinking, not app-based. The same three-step pattern works in Gemini Gems, Custom GPTs, and Copilot Agents — pick the ecosystem your school is already in.

Step 01