A polished lesson can still contain an incorrect explanation, an ambiguous diagram, or feedback that gives away the answer. Fluency and visual quality are reasons to inspect the work carefully, not evidence that it is ready for teaching.
Treat generated material as a draft. The following review is a practical editorial workflow for educators, not a certification scheme. Use the checks that fit your subject and school policy, involve a specialist when needed, and keep a record of the version you actually reviewed.
1. Read the objective before the opening scene
Write down what the learner should understand or be able to do. Then inspect the activity without its decorative story. Does the learner need that understanding to complete the meaningful part of the task? A game that rewards speed or trial and error may produce many correct choices without revealing the intended reasoning.
Check prerequisites and the final evidence task together. If a task expects vocabulary that has not been introduced, decide whether to teach it, supply access support, or change the task. Avoid quietly lowering the objective merely to make the experience easier to finish.
2. Verify the claims and the representations
List the central factual claims and check them against suitable sources. Open the references themselves. A plausible title or a citation-shaped string is not enough. For a mathematics activity, work the examples independently. For a science diagram, inspect the labels, units, relationships, and any simplifications that affect the explanation.
Consider a food-web activity. If arrows indicate the direction in which energy passes, the legend and every arrow must use that convention consistently. A visually attractive arrow pointing the other way changes the meaning. If the model omits relevant relationships, the teacher notes should state those limits rather than presenting the simulation as a complete prediction of a real ecosystem.
3. Try the wrong answers as well as the right ones
Walk through the experience as a learner. Try a likely misconception, an empty response, a plausible alternative, and a repeated attempt. Look at what the feedback actually asks the learner to do. Does it explain the relevant relationship, reveal the answer immediately, or send the learner into an unhelpful loop?
Inspect any hint sequence. A supported success is useful information, but the evidence record should retain the assistance that led to it. If the next question is intended as an independent check, make sure the earlier answer is not still visible in a way that defeats that purpose.
- Check that instructions and success conditions agree.
- Test an expected misconception and a justified alternative.
- Confirm there is a sensible stopping point and a recovery route.
4. Look for barriers unrelated to the learning goal
Read the instructions at the size a learner will see. Try the meaningful controls with a keyboard. Check whether the lesson depends on hearing audio, distinguishing colors, dragging precisely, or responding quickly. Provide a transcript, readable labels, a static view, or an alternative response where it preserves the objective.
A change in response mode should not automatically be treated as help with the answer. An oral explanation can show the same reasoning as a typed one when writing is not the assessed skill. If writing itself is the objective, discuss how the adaptation affects the evidence rather than assuming the two tasks are interchangeable.
CAST organizes UDL around engagement, representation, and action and expression. The review questions here use that lens to identify barriers while keeping the specific learning goal visible.
CAST Universal Design for Learning Guidelines5. Review the boundaries around the content
Check age suitability, the source and rights record, and the information the activity asks learners to provide. An example can usually be invented without uploading a child’s private work or family details. Keep generated interactions within the school’s approved boundaries, with a clear path back to the teacher.
AI suggestions should remain inspectable. A recommendation to change a task is different from permission to publish it, and a content review is different from a legal compliance determination. Use the school’s decision process for questions outside your authority and keep sensitive context out of public examples.
UNESCO’s 2023 guidance emphasizes human agency and validation of educational uses of generative AI. This article offers a classroom review routine consistent with that direction, without claiming formal endorsement.
UNESCO: Guidance for generative AI in education and research6. Record a decision someone else can understand
Finish with a clear outcome: approve this version for a stated use, request specific revisions, or hold it until a question is resolved. Record the scope, sources checked, important limitations, reviewer, and date. ‘Looks good’ is difficult to act on when another teacher later finds an error.
If a revision changes a scientific claim, the assessment demand, or a learner interaction, review that change before relying on the earlier approval. Keep a practical fallback for the lesson itself. The aim is to make review an ordinary part of teaching preparation, with enough detail to protect useful work and enough clarity to change it.
Take this into your practice
Review the experience a learner will actually receive—including wrong answers, access barriers, and later changes—and approve a specific version for a clear purpose.