Check an AI Lesson Against Its Sources

Check an AI-generated lesson by tracing each important claim to the source that supports it. Separate direct observations, reasonable interpretations and unsupported additions. A citation at the end of a paragraph does not establish that every sentence in the paragraph follows from the cited material.

Use this original source passage

“During a two-week fictional library trial, 40 students used a new booking form. Thirty-two completed it without staff assistance. Eight requested help, most often at the date-selection step. The trial did not compare completion time with the previous form. Staff planned to simplify the date instructions before another trial.”

This passage is invented for practice. It contains enough facts to check a short lesson, but it is not a real study or evidence about library systems. Keep the passage visible while reviewing the claims below. In a real lesson, record the source title, version, section and access date.

Audit an overconfident summary

Suppose an AI draft says: “The new form made all students faster and eliminated confusion. The study proves that digital booking improves learning outcomes. Staff permanently adopted the form.” Each sentence sounds like a plausible success story, but the source does not support those conclusions.

A corrected summary is: “In a two-week trial, 32 of 40 students completed the form without staff assistance. Eight requested help, especially with date selection. Staff planned to revise those instructions and run another trial. The passage does not establish a time saving or a learning benefit.”

Check the calculation and its label

The independent-completion proportion is 32 / 40 = 80%. Label it as completion without staff assistance among the trial participants. It is not an 80% improvement, since no baseline was measured. It is also not evidence that 80% of all students would complete the form independently in another setting.

Ask a learner to explain which additional measurement would be needed for a claim about speed. A suitable answer names comparable completion times under the old and new forms and considers differences in participants or tasks. This is a reasoning exercise, not permission to invent a control group that the passage never mentioned.

Build a traceability table for your lesson

Use columns for claim, source location, support status, correction and reviewer note. Review numbers, definitions, causal statements and conclusions first. Then inspect transitions that introduce stronger language such as proves, always or guarantees. Small connective phrases can turn a limited observation into a broad assertion.

Cornell's ethical-AI guidance recommends external accuracy checks. Another AI model can help identify claims to examine, but agreement between two generated responses does not replace opening the source. If a source is unavailable, mark the claim unresolved and remove or qualify it until you can verify it.

Preserve uncertainty in the final activity

Include questions that ask what the passage does not establish. This teaches learners to recognize boundaries instead of memorizing a simplified success narrative. Keep answer explanations tied to the passage and accept equivalent wording when the reasoning is correct.

Use the AI quiz review to check the assessment built from the lesson. The guided-practice sequence shows how to move from a worked example to independent reasoning. Review again when sources or lesson claims change; a previous check does not cover a new revision.

Sources and further practice

Cornell: Ethical AI for Teaching and Learning supports the reference principle used here. The exercise, example data and review routine on this page are original Paradigm Study teaching examples.

For a broader workflow, see educators and creators. Bring your attempt and the step that confused you into Paradigm Study for a lesson or focused practice. Start a learning notebook.