An AI-polished assignment can hide the most interesting part of learning: the decisions that produced it. If a teacher sees only the final version, it can be difficult to tell whether the student understands the argument, recognised an error or simply accepted a smoother sentence.
A useful classroom exercise makes the edit itself visible. Ask students to preserve an original draft, make one controlled revision and explain what changed. The aim is to reveal judgement while giving students a concrete way to examine AI assistance.
Choose the learning objective first
Decide what the exercise is meant to teach. You might focus on paragraph structure, identifying unsupported claims, explaining a mathematical step or adjusting language for a specified audience. Choose one objective so that the activity can be assessed clearly.
The method below is a suggested teaching design, not a validated intervention. Adapt it to the subject, level and institution. Use approved tools, appropriate privacy arrangements and applicable age requirements. A teacher-supplied sample response can let students participate without creating an AI account.
Stage 1: produce and preserve the first attempt
Give students a short task they can attempt themselves. For example, ask them to explain a course concept in 150 words using a supplied source. Keep the original draft, even when it is incomplete. It is evidence of the starting point, not an object of embarrassment.
Tell students which materials are allowed and how the work will be assessed. Do not ask them to upload classmates’ work, personal information or restricted course material into a tool without authorisation.
Stage 2: request one bounded kind of assistance
A broad “improve this” request can change too many things at once. Ask for assistance that matches the objective. If the lesson is about evidence, the tool might identify sentences that need support. If it is about structure, it might propose a clearer order without adding facts.
Read this draft alongside the supplied source. Identify up to three places where the explanation is unclear or unsupported. Do not rewrite the entire answer or introduce new information. Explain the issue and ask a question that would help the writer revise it.
This original prompt keeps the student in the revision role. The tool’s observations are suggestions to inspect. A comment that sounds plausible may still misunderstand the source or the assignment.
Stage 3: make a decision about each suggestion
Ask students to mark each suggestion as accepted, modified or rejected. Require a reason tied to the learning objective. “It sounds better” may be a useful starting impression, but it should not be the complete explanation.
| Suggestion | Decision | Reason |
|---|---|---|
| Move the definition to the first paragraph | Accept | The reader needs it before the example. |
| Add a broad claim about all students | Reject | The supplied source does not support that population. |
| Replace a technical term with a simpler phrase | Modify | The term matters, so define it rather than remove it. |
These are illustrative decisions, not answers to a particular assignment. In a real lesson, the student should point to the source or the course criterion that supports the explanation.
Stage 4: explain the final edit
Collect the original, the relevant AI suggestion, the revised passage and a short explanation. Invite the student to describe one improvement, one rejected suggestion and one uncertainty that remains.
Offer equivalent ways to explain: a written annotation, a short recorded explanation, a conversation or another suitable accessible format. If the objective is subject understanding, do not accidentally turn the assessment into a test of fast speaking or confidence under pressure.
Assess understanding separately from polish
Use a rubric that distinguishes factual accuracy, reasoning, source use and communication. A fluent final answer may deserve credit for presentation while still needing work on the underlying argument. An awkward sentence may contain a sound idea that deserves recognition.
The revision record is one source of evidence. It is not a perfect detector of authorship or understanding. A prepared explanation can also be copied. Combine it with discussion, in-class work or another assessment method appropriate to the subject.
What this design makes possible
The teacher can see how a student encountered and evaluated assistance. The student has to make an explicit choice rather than treating the AI version as automatically authoritative. Both can discuss a mistake without reducing the entire lesson to whether a tool was used.
UNESCO’s guidance argues for human-centred and age-appropriate use of generative AI in education, including attention to privacy and pedagogical validation. The exercise proposed here is consistent with that direction, but its effectiveness in a specific classroom should be observed and reviewed.
A small pilot before a course-wide change
Try the exercise on one short task. Ask students what the revision record helped them notice and which part felt confusing. Review whether their explanations revealed the intended learning objective. Adjust the task before extending it.
For a related source-checking habit, read our practical route through a research paper. The shared principle is useful: keep the source, the interpretation and the judgement visible.
Reference: UNESCO: Guidance for generative AI in education and research.
Featured image: conceptual illustration created with AI for MaryChuks.com.
Discover more from Marychuks.com AI, Psychology, Business & CreativeVerse
Subscribe to get the latest posts sent to your email.
I like that the first draft stays in the picture… It gives a student room to say why they kept one suggestion and left another out.