Building Metacognition in the AI Era

Each process breaks down into specific skills students can practice. Select a skill to see the strategies that build it.

Task analysis, prior-knowledge activation and strategy selection are each built by a single strategy. If your course uses only a few strategies, include at least one from Plan so these skills are practiced.

Research describes two connected parts. Knowledge of cognition is what a learner knows about themselves, tasks and strategies. Regulation of cognition is what a learner does with that knowledge: planning, monitoring and evaluating their work (Schraw & Dennison, 1994). Knowledge only improves learning when a student acts on it, and regulation only works when it draws on accurate knowledge.

Conditional knowledge matters more than ever. When AI is one click away, every task now includes a decision about whether and how to use it, and that decision depends on knowing what the task is for.

Plan: the AI's first response defines the task, prior knowledge is never activated, and "ask AI" replaces choosing a strategy.

Monitor: fluent text feels like understanding, trust in AI lowers checking, and students rely on the tool to tell them they're right.

Evaluate: a strong product hides weak learning, credit goes to the wrong source, and the reflection itself can be written by AI.

What a student learns in evaluating feeds the next round of planning. Regulation is a cycle, not a sequence (Zimmerman, 2002; Winne & Hadwin, 1998).

Students adjust what they do based on how well they think they are doing (Nelson & Narens, 1990). A student who believes they understand stops studying; a student who notices confusion slows down, rereads or asks for help. Monitoring links what students know about their learning to what they do next (Greene & Azevedo, 2007).

If a student's judgment of their own understanding is wrong, every decision that follows is wrong too: they stop too early, skip the hard part, or trust an answer they should have checked. Four of the eleven strategies target monitoring directly, and several others build it in as a step.

Assignments have long assumed that producing the work required doing the thinking. AI breaks that link. In a field experiment with nearly 1,000 high school students, unrestricted GPT-4 access raised practice performance, but students scored 17% lower once access was removed than peers who never had it (Bastani et al., 2025). In a randomized study of essay writing, ChatGPT support improved essay scores without greater knowledge gain or transfer (Fan et al., 2025).

Learners routinely mistake ease of reading for understanding (Bjork, Dunlosky & Kornell, 2013), and AI text is fluent by design. In a survey of 319 knowledge workers, higher confidence in AI was associated with less critical thinking, and higher self-confidence with more (Lee et al., 2025). AI weakens monitoring at exactly the point everything else depends on.

Reflection journals, learning logs and take-home self-assessments were built to make students think about their thinking. AI can now write them in seconds. Metacognitive practices have to be redesigned so the metacognitive act itself stays with the student.

AI tends to amplify the learning habits students already have: strong self-regulators use it strategically, while others offload more of their thinking (Dhurumraj & Labuschagne, 2026). Explicitly teaching metacognition benefits disadvantaged learners in particular (Education Endowment Foundation). Building it into course design is an equity decision.