From cognitive necessity to cognitive choice

§01 The assumption

What assessment used to do for free

Most of the debate about AI in higher education has been about cheating, authorship and detection tools. Montebello argues that this hides a more basic change. The link between what students hand in and what they have actually thought through is coming loose.

Constructive alignment (Biggs, Tang and Kennedy, 2022) rests on that link. When outcomes, learning activities and assessment fit together, we assume that a good performance shows the student went through the thinking the outcome describes. For a long time the assumption held, because the task itself made the student think. Montebello calls that built-in effort productive friction.

Montebello describes AI as an epistemic engine that moves part of the thinking out of the student. Whether a student engages now depends on motivation, a sense of responsibility and the ability to regulate their own learning. That asks more of students, and most of all in mass higher education, where prior knowledge and support vary a great deal.

§02 The gap

Performance without learning

The article’s central term is the cognitive engagement gap, the distance between completing an assessment and the processes we connect with learning. Making meaning, integrating concepts and regulating your own understanding are examples. The gap doesn’t mean learning stops, or that every assessment with AI in it is worthless. It means the result is no longer a reliable sign that the student worked with the material. And the problem goes beyond measuring the wrong thing, because the assessment no longer requires the processes it claims to measure.

Generative AI makes visible what was previously implicit, that assessment supports learning only when it structurally requires learners to think.

§03 CECA

Five principles, in practice

Montebello’s answer is cognitive engagement-centred assessment (CECA). It isn’t a recipe. The five principles are meant to make the thinking explicit, visible and valued in whatever format you use.

Montebello also presents CECA as a matter of fairness. When engagement is built into the design, learning becomes a shared and supported expectation. It no longer hangs on each student’s confidence, academic capital or metacognitive maturity.

§04 Your teaching

What it means for the way you plan

Assess the process, and say so in the criteria

Single submissions of essays and reports are the most exposed. Formats built around the process make the thinking easier to see.

• submissions in stages
• annotated drafts
• reflective notes and learning logs
• spoken explanations of the student’s own work
• design rationales, where students explain choices, doubts and revisions

These traces have to be part of the criteria. Added on as an extra, they tell students that the polished product is still what counts. Step 3 builds windows like these into one assessment.

Keep the productive struggle

Research on productive failure shows that conceptual understanding grows when students struggle first and get guidance afterwards (Kapur, 2008). AI that hands over a finished answer at once can remove exactly that friction. Montebello suggests letting students work on a problem for a set time before they use AI, asking them to compare an AI answer with their own, or giving them an AI answer with errors to find, criticise and correct. AI then becomes something students examine. Step 3’s “Fix the AI answer” is one version of the last idea.

Our addition. The struggle doesn’t have to happen without AI. In Step 2, AI in the role of a coach gives hints where it could give answers, and in the study by Bastani and colleagues (2025) a tutor version of that kind largely removed the harm to learning.

Teach good AI habits

Montebello points to practices that keep students engaged. They can use several AI systems to expose where they disagree, ask AI for ideas instead of answers, and question it in a Socratic dialogue. None of these habits arrive by themselves. Students have to learn them, practise them and be assessed on them.

Vary the forms

Oral assessment, reasoning in class, authentic practical tasks and dialogue all ask students to explain, justify and adjust their thinking as they go. They raise questions of scale and workload, which Step 4 takes up, but well designed they can strengthen validity without surveillance.

Rethink alignment

Alignment has to become explicitly cognitive. Learning outcomes should describe how students are to think as well as what they should know. Learning activities should model and scaffold that thinking, and the criteria should reward evidence of reasoning, reflection and metacognition. Montebello calls this a move from alignment between content and task to alignment between process and evidence. Without it, the article warns, alignment can become a façade that hides the engagement gap.

Assure learning through design

Make the expectations for AI use clear, treat openness about use as normal where it fits, and design assessments that mean something whether AI is used or not. That also takes development for teaching staff, and support from the institution for trying new forms.

In this framing, the pedagogical question shifts from ‘Did the student use AI?’ to ‘What evidence demonstrates that the student learned?’

§05 Read it for what it is

A conceptual argument

The article is a theoretical analysis. Its terms are lenses for thinking with, not empirical findings, and CECA has not yet been tested. Montebello calls for studies of how different designs affect engagement, how students experience the cognitive choice, and how disciplines differ.

Why read it anyway? It moves the conversation about AI and assessment away from cheating and control, towards the question of what counts as good evidence of learning.