A completed answer can be a useful classroom product, but it is not a complete picture of what a student understands. Generative AI makes that distinction especially visible: a tool can help produce text while the learner’s reasoning remains unclear. The educational question is therefore how assistance fits the learning objective and what evidence of understanding an activity should reveal.
Start with the skill being taught
UNESCO’s AI competency framework for students places critical judgement alongside technical understanding and ethical awareness. Its structure moves through understanding, applying and creating, treating students as people who can examine and shape AI use rather than simply receive outputs.
That provides a useful starting point for lesson design. If the objective is to practise building an argument, the teacher needs to see how the learner selects evidence and connects it to a conclusion. If the objective is to critique an explanation, a generated draft might instead become material to analyse.
These are different activities even when the final page looks similar. Making the objective explicit helps determine which assistance is appropriate, what students should disclose and which parts of the work must demonstrate their own understanding.
Make the reasoning visible
Consider a hypothetical lesson built around a short, teacher-selected source. Students first identify its main claim, then compare it with a generated summary. They mark omissions, unsupported additions and wording that changes the meaning. Finally, they write a revised summary and explain two of their editorial choices.
This is an illustrative teaching design, not a report of a measured classroom outcome. Its purpose is to show how the same technology can be used to create something for examination rather than supply an answer that ends the activity.
Another exercise might ask learners to propose questions before consulting a tool, then record which questions the output actually answers. The comparison can reveal a mismatch between fluent writing and relevance. The important evidence is the student’s explanation of that mismatch.
Support does not remove the need for evaluation
UNESCO’s guidance on generative AI in education and research advocates a human-centred approach and calls attention to privacy, equity and pedagogical validation. It does not establish that every tool improves learning or that one classroom policy suits every age and context.
Access also matters in an individual activity. A task should not quietly assume that all learners have the same device, subscription, language support or permitted account. Teachers need a way for students to meet the learning objective when a particular commercial tool is unavailable.
Similarly, introducing a tool is different from assessing its educational value. A useful evaluation asks what students can explain or do after the activity, what support they needed and which difficulties remain. A polished final document alone cannot answer all of those questions.
Set clear rules for the work
The OECD AI Principles identify human agency and transparency as important values. In a classroom, a practical application is a clear instruction that explains whether AI assistance is allowed, what it may be used for and how its contribution should be described.
A learner might note that a tool suggested an outline while the student selected and checked the evidence. Another task may require an unaided first attempt. The rules should match the skill being assessed and be explained before the work begins.
The strongest educational use of AI is a question to investigate, not a conclusion established by the technology’s presence. An activity becomes easier to judge when the learner’s decisions, evidence and revisions are visible. That keeps attention on understanding rather than the mere production of a finished answer.
