Teach AI Literacy With a Simple Claim-Checking Exercise

Students do not need an advanced technical lesson to begin questioning AI output. A short AI literacy activity can show why a fluent answer still needs checking. Give learners a generated paragraph and a reliable source, then ask them to identify which statements the source supports, contradicts, or leaves unresolved.

The task is deliberately small. Students are not trying to prove whether all AI is reliable. They are examining one answer against one defined source. That boundary makes the exercise manageable and turns a vague warning about mistakes into a concrete reading skill.

Choose a low-stakes source

Use a short text about a familiar topic: a fictional museum schedule, a classroom recycling procedure, or a description of a plant’s growth conditions. The source should be clear enough that students can locate evidence without needing specialist knowledge.

A fictional museum notice could say that the building opens Tuesday through Sunday, closes at five, and offers a workshop on Saturday morning. You control the details, so you can create a fair exercise without relying on changing public information. Label the notice as fictional.

If you use a real source, record its title, publisher, date, and link. Check it before class. Avoid medical, legal, or emotionally sensitive claims in an introductory activity; the learning objective is evidence matching, and a simple context serves that purpose well.

Prepare a paragraph with mixed claim quality

Generate or write a sample answer that includes accurate statements, one clear error, and one unsupported addition. If you edit the paragraph yourself, tell students it is a prepared teaching sample. Do not present a deliberately altered text as untouched output from a named system.

For the museum example, the paragraph might correctly state the closing time, incorrectly say the museum opens Monday, and add that workshops include free materials when the notice says nothing about materials. These differences create three separate judgments rather than one blanket label of “wrong.”

You can also include a sentence that is partly supported. “The museum opens every weekend and runs workshops on both days” contains one correct component and one incorrect one. This encourages students to divide compound statements before evaluating them.

Explain the evidence categories

Use four labels: supported, contradicted, not established, and unclear. Supported means the source contains adequate evidence for the claim. Contradicted means the source provides conflicting information. Not established means the source does not settle the point. Unclear means the wording or source needs clarification before a fair judgment.

Emphasize that “not established” does not automatically mean false. The museum may supply materials, but the provided notice does not tell us. A careful answer would omit that promise or say that visitors should check the workshop details.

This distinction matters beyond classroom work. People often mistake a missing fact for a false fact, or treat a plausible addition as confirmed. The activity gives students language for uncertainty without requiring them to guess.

Run the AI literacy activity in pairs

Give students the source first and ask them to read it independently. Then distribute the generated paragraph. Have each pair underline factual claims and assign a number to each one. They should split sentences when different parts require different evidence.

For every claim, students record a category and the supporting source phrase. A response such as “sounds wrong” is incomplete. Ask them to point to the line that establishes the opening days or to explain that no line addresses the materials.

One student can propose a classification while the other checks the evidence. After several claims, switch roles. This creates a reason to discuss disagreements rather than simply dividing the worksheet and working separately.

Discuss the claims that caused disagreement

Ask groups which statement was hardest to classify. Often, the interesting cases involve words such as “always,” “all,” or “free.” These small additions can change the meaning of an otherwise accurate sentence.

A student might argue that workshops probably provide materials. Acknowledge that this could be reasonable speculation, then ask whether the source confirms it. The exercise is about the evidence available for a published claim, not whether an assumption sounds sensible.

You can connect this habit to everyday AI reading. A learner encountering an article through Aiera.blog can ask which statements are supported by a named source and which are interpretation, prediction, or practical advice.

Ask students to repair the paragraph

Have each pair produce a corrected version using only the supplied evidence. They should fix the Monday opening claim and remove or qualify the unsupported materials statement. Encourage plain wording rather than elaborate disclaimers.

Then ask students to explain one change. “We removed ‘free materials’ because the notice does not mention materials” is a complete justification. “We made it more accurate” is too vague to reveal the reasoning.

Compare different valid repairs. One pair may omit the uncertain detail; another may write that material arrangements are not stated. Both can be appropriate depending on the purpose of the answer. Discuss which version better serves a visitor who needs to plan a workshop visit.

Keep the lesson balanced

Do not frame the exercise as a contest to catch AI being foolish. The paragraph should contain useful, supported information as well as problems. Students need to learn selective trust, including the ability to recognize when a claim does match its evidence.

UNESCO’s student competency framework includes critical engagement with AI within a broader educational approach. This classroom exercise is one practical way to make that engagement visible without claiming to measure a student’s complete AI literacy.

Also note the limit of the exercise: checking against one source establishes agreement with that source, not universal truth. In a more advanced lesson, students can investigate whether the source itself is current, credible, and relevant.

Assess the process, then increase difficulty

Use three simple criteria: did students identify separate claims, match judgments to evidence, and make justified corrections? Grammar can matter for the final paragraph, but it should not hide the main learning objective.

For a later activity, introduce two sources with different dates or scopes. Ask students to explain why they disagree before choosing which one applies. Keep the first lesson simpler so learners understand the categories before handling source conflict.

End by asking each student to write one question they will use when reading an AI answer. “Where does this claim come from?” is a strong starting point. The lasting skill is not suspicion of every sentence. It is the ability to connect a statement with evidence and express uncertainty when that connection is missing.