Unit 3 — Can I Trust What AI Tells Me?

Unit 3 · Think · KC-3 Critical Evaluation

Can I Trust What AI Tells Me?

"AI answers with confidence — even when it is wrong. Critical thinking is not the enemy of AI; it is the condition for using AI correctly."

2 Sessions · 100–120 min

9 Activities

Grades 7–12

KC-3

Learning Outcomes

LO-1 · Identify and explain what AI hallucinations are and why they occur

LO-2 · Write verification prompts that ask AI to assess its own reliability

LO-3 · Apply the Cross-Verification Framework (CVF) to any AI response

LO-4 · Distinguish between Fact, Opinion, and Invention in AI output

LO-5 · Recognize how AI bias shapes responses to the same question differently

LO-6 · Produce a structured Fact-Check Report before sharing AI-sourced information

Session Flow

🕵️

S1 The Invented Expert → Hallucination Hunt

🧠

S1 Why AI Hallucinates → Verification Prompt

🔍

S2 Recall Warm-Up → Cross-Verification Lab

⚖️

S2 Fact/Opinion/Invention → Spotting Bias → Fact-Check Report

✍️

D3–D5 Quiz + Rubric + Reflection

Bridge from Unit 2In Unit 2, you learned HOW to ask AI well. In Unit 3, you learn WHEN to question the answer — a well-formed answer does not mean a correct answer.

Why this unit?

This unit works best when the teacher models uncertainty. Normalize saying “I'm not sure — let's verify.” Students internalize healthy skepticism when they see adults practice it. Read the full Teacher Guide and run Activity 1 yourself with an AI tool before delivery — note any hallucinations you encounter.

Unit 3 of 5

Keep this question in mind throughout both sessions

If AI always sounds confident in its answers… how do you know when to trust it and when to question it?

My answer NOW

Share with a neighbor

Talk to the person next to you. What is their answer? Do you agree?

My answer AFTER

Return here at the closing of Session 2.

Watch For…

Post the Essential Question visibly for both sessions. Display for a full silent minute before asking for answers. Record 3–4 answers without judging.

The Invented Expert

Hook & Discuss

Look at a real graded report — silently — then discuss what you notice.

10 min · Session 1 · Engage

Why This Activity?

Students enter the unit through a realistic, emotionally engaging scenario. The hook surfaces existing assumptions about AI reliability before any instruction begins. Show the image silently — give students 60 seconds to read it, then ask: “What do you notice?”

The Scenario — A Real Student Report

A student wrote a report with AI assistance. It received an A+. The teacher later discovered that the book listed — "The Future of Quantum Learning" (1927) — does not exist. Einstein never wrote it. AI invented it — with complete confidence — and neither the student nor the teacher caught it.

Discuss (one question at a time)

1. How would you tell the difference between a real piece of information and an invented one, if both look identical?

2. Is AI's confident tone enough proof that what it says is correct?

3. What is your responsibility as a user when you share information from AI with someone else?

Watch For…

Students who say “I always check AI” — probe: what does checking actually look like for them? Students who feel AI is always reliable — validate, then introduce the tension gently. Do not reveal answers yet — the goal is productive discomfort.

Hallucination Hunt

Explore

Ask AI about a niche topic, then verify every claim against a second, independent source.

15 min · Session 1 · Explore

Why This Activity?

Students directly experience AI hallucination rather than hearing about it abstractly. First-hand encounter creates lasting conceptual understanding and genuine motivation to verify. Groups that query very well-known topics see fewer hallucinations — redirect to niche or specific fact-based queries.

Step 1 — Ask AI about a niche topic (a historical figure, a local landmark, a scientific discovery)

Step 2 — Record the AI response

Step 3 — List the claims, verify each with a second source, then mark: ✅ Verified · ❓ Unverified · ❌ Incorrect/Invented

ReflectDid you find anything AI invented that sounded completely real? What does this tell you about using AI for research?

Watch For…

Groups that accept all AI output as correct — pause and ask them to run a single claim through a search engine. Clarify the difference: a hallucination is fabricated, not just mistaken. Introduce the term “Hallucination” during class debrief.

Why AI Hallucinates

Explain

Build a mental model: AI predicts language — not truth.

10 min · Session 1 · Explain

Why This Activity?

Students need a mental model of why hallucinations occur before they can develop systematic responses. Understanding the mechanism reduces fear and builds informed agency. Use the analogy of an “extremely well-read guesser.”

🔮

Predicts, Doesn't Look Up

AI generates text by predicting the most statistically likely next word — it does not "look things up" like a search engine does.

🧩

Fills Gaps Confidently

Trained to sound fluent and helpful, it may fill knowledge gaps with plausible-sounding content it has no actual evidence for.

🚫

No Uncertainty Alarm

AI does not know what it does not know — it has no built-in signal that says "I'm not sure about this."

The most important point in the unitAI predicts language — not truth. It does not verify facts before answering. Fluency is not accuracy.

Quick Poll — is AI hallucinating the same as AI lying?

Answer: No — AI is not intentionally deceptive; it lacks the awareness to know it is wrong.

Watch For…

Students who anthropomorphize AI (“It lied to me!”) — gently reframe: no intent, just pattern prediction. Students who conclude AI is useless — validate the frustration, pivot: “So what do skilled users do?”

Write a Verification Prompt

Build & Test

Return to your AI from Activity 2 and ask it to reflect on its own certainty.

15 min · Session 1 · Elaborate

Why This Activity?

Students practice an immediately applicable skill: turning AI into a self-checking partner. This activity closes Session 1 with agency — students leave with a tool they can use tomorrow.

Prompt Starters

"Look at the response you gave me above. For each claim, tell me: how confident are you, and how could I verify it?"

"Which parts of your previous answer might be incorrect or uncertain? Please flag them."

"Give me 3 reliable sources I can use to check what you just told me."

Which starter did you use, and how did AI respond?

Now write YOUR OWN original verification prompt (go beyond the starters)

ReflectDid AI's self-assessment match what you found in Activity 2? What surprised you?

Important LimitVerification prompts help, but AI can still be wrong about its own uncertainty. Always check with an independent source.

Watch For…

Students who copy the exact prompt starters — push for personalization: “What's a topic YOU care about verifying?” AI that resists the meta-prompt or gives vague self-assessments is a teachable moment: even the verification prompt has limits.

Exit Ticket — Session 1 (D1)

5–10 minutes — Individual and honestThis is not graded — it helps your teacher understand what you learned today.

In your own words, what is a Hallucination in AI?

Write ONE verification prompt you could send to AI after receiving a factual response. Make it your own — don't copy from the slide.

Complete this sentence: "A confident AI tone means …"

Quick CheckI can define Hallucination and give an example · I understand why AI produces hallucinations · I can write a verification prompt · I know why confident tone ≠ accurate content.

Watch For…

Sort Exit Tickets into three groups before Session 2: Clear understanding / Partial / Needs re-teaching. Open Session 2 by addressing the most common gap revealed.

Warm-Up: Recall + Real Examples

Engage

Recall Session 1, then see real published examples of AI hallucination.

8 min · Session 2 · Engage

Why This Activity?

Bridging from Session 1 reactivates prior learning and surfaces any remaining misconceptions before new content is introduced. Real-world examples ground the session in relevance.

Recall Prompts

1. What is a Hallucination? Give an example we found last session.
2. What is one verification prompt you wrote?
3. True or False: If AI sounds certain, the information is correct.

🌍 Real-World Bridge

Real published examples of AI hallucination exist — fabricated legal citations, invented historical events. This happened to real people, in real professional contexts. Today we go further — not just checking if something is true, but distinguishing fact from opinion and from invention.

Watch For…

Students who conflate “wrong” with “hallucinated” — re-clarify the distinction quickly. Students who didn't retain Session 1 vocabulary — offer a quick visual glossary.

Cross-Verification Lab

Explore & Apply

Apply the 4-step Cross-Verification Framework to a real AI Response Card, using the SOURCE test.

20 min · Session 2 · Explore

Why This Activity?

Students apply a structured method — not intuition — for evaluating AI output. The lab format makes verification feel like a skill, not a chore. Working in groups simulates real collaborative fact-checking.

The Cross-Verification Framework (CVF)

Identify Claims

Underline every factual claim in the AI response. What is the AI asserting as fact?

Source Hunt

Find at least 2 independent sources per claim. Who else says this, and are they credible?

Compare & Flag

Mark: ✅ Confirmed · ❓ Unverifiable · ❌ Contradicted. Does the evidence match the claim?

Summarize

Write a 2-sentence verdict on the AI response's reliability. Would you trust this response? Why / why not?

The SOURCE Credibility Test

Letter	Question to Ask the Source
S · Sector	Who published this? (.gov, .edu, established news org, or personal blog?)
O · Objective	Is the goal to inform, to sell, or to persuade?
U · Updated	Is it recent enough to be relevant to this claim?
R · Referenced	Does it cite its own sources or evidence?
C · Consistent	Do other credible sources say the same thing?
E · Expert	Is the author qualified to make this claim?

Rate each sourceHigh (H) · Medium (M) · Low (L) credibility. This rating becomes part of your Fact-Check Report.

Topic of your AI Response Card & claims identified

Verification Log — claim, source used, credibility (H/M/L), verdict (✅❓❌)

Watch For…

Groups that rely on a single Google search as “verification” — apply the SOURCE test together. Groups that verify only easy claims — guide them to flag harder-to-check assertions too. Groups that rate all sources “High” reflexively — push them to apply each SOURCE letter explicitly.

Fact vs. Opinion vs. Invention

Explain & Sort

Learn to categorize AI claims, then sort 8 AI-generated sentences.

10 min · Session 2 · Explain

Why This Activity?

Many students conflate these three categories. Making the distinction explicit equips them to evaluate not just accuracy but also the nature of AI claims — a meta-cognitive leap.

FACT

A verifiable claim that can be confirmed or refuted. Clue words: "was born," "discovered," "costs," "studies show."

OPINION

A judgment or interpretation — reasonable people may disagree. Clue words: "is considered," "many believe," "arguably."

INVENTION

A fabricated claim presented as fact — cannot be verified because it does not exist. Details sound specific but trace to nothing.

Sorting Exercise — Label each sentence: F / O / I

#	AI-Generated Sentence	Your Label
1	"Marie Curie was born in Warsaw in 1867."
2	"Curie is widely considered the greatest scientist of the 20th century."
3	"In 1903 Curie published The Science of Radiance, her most influential book."
4	"Radioactivity was first observed by Henri Becquerel in 1896."
5	"Einstein's 1919 paper on quantum education changed school systems worldwide."
6	"Many experts argue that Curie's work is underrepresented in school curricula."
7	"The Curies received the Nobel Prize together in 1903."
8	"Science education has benefited enormously from Curie's legacy."

NoteSentences 3 and 5 contain invented information — the books/papers named do not exist. Compare your answers with your group.

Watch For…

Students who label all AI output as “fact” — remind: AI does not label its own claims; the reader must categorize. Disagreement on Opinion vs. Fact is productive; use it to deepen discussion about subjectivity.

When AI Disagrees With Itself — Spotting Bias

Explain

Compare two AI tools' answers to the same debatable question — and ask why they differ.

7 min · Session 2 · Explain

Why This Activity?

Bias appears in the unit concepts but had no learning activity. This short activity makes it visible, experiential, and directly tied to the unit's critical thinking goals. Students discover that the same question can produce meaningfully different AI answers — and ask why.

Same Question, Two AI Tools

"Is social media beneficial for teenagers?"

Response A

"Research consistently shows that social media strengthens social bonds, supports mental health through peer connection, and gives young people a voice in public discourse. Used responsibly, it is a net positive for adolescent development."

Response B

"Studies indicate that heavy social media use is linked to increased anxiety, sleep disruption, and social comparison in teenagers. While some benefits exist, the risks to adolescent wellbeing are significant and well-documented."

Both responses sound confident and factual. Which one is correct?

What might explain why two AI tools gave such different answers on the same topic?

What does this tell you about trusting a single AI response on a debatable topic?

Key ConceptAI bias can come from training data, prompt wording, or system design. It often looks like a reasonable, confident answer, not an obvious error — that is what makes it worth checking. For contested topics especially, always seek more than one source and more than one AI tool.

Watch For…

Students who conclude “AI is always biased and useless” — reframe: bias exists on a spectrum; the skill is recognizing and accounting for it. Students who pick one response as simply “more true” — push: on genuinely debatable topics, the answer is not always binary.

My Fact-Check Report

Elaborate · Summative

Select a topic, generate an AI response, apply the CVF, and produce a written Fact-Check Report.

15 min · Session 2 · Elaborate

Why This Activity?

This is the summative performance task of Unit 3. Students synthesize all unit skills into a concrete, transferable product. The report format mirrors real-world fact-checking practice. Share the Rubric (D4) before students begin.

Independently select a topic, generate an AI response, and apply everything you learned today. This is a real product you can stand behind.

Topic & AI Query — state the topic and the exact prompt used

AI Response Summary — paraphrase the AI's response (do not copy)

Verification Log — at least 4 claims: source, credibility (H/M/L), verdict (✅❓❌)

Reliability Verdict — rate the AI response (High/Medium/Low) and justify

Verification Decision — would you share this response with others as-is? What would you change first?

Reflection — what will you do differently next time you use AI for research or publishing?

Watch For…

Circulate — check students flag at least one ❓ or ❌ claim. Students who only verify “safe” obvious facts — redirect: the value is in checking what seems credible but might not be. Students who use AI to verify AI must be flagged directly: “Can a source verify itself?”

Exit Ticket — Session 2 (D2)

7 minutes — IndividualUse the vocabulary you learned in both sessions.

What are the 4 steps of the Cross-Verification Framework (CVF)? List them in order.

Give one example from real life (or today's class) of each: Fact, Opinion, Invention/Hallucination.

In Activity 7B, two AI tools gave different answers to the same question about social media. What does this tell you about relying on a single AI response for a contested topic?

Complete this sentence: "Before I share AI-generated information with anyone, I will …"

🎉 Unit 3 is nearly complete. In Unit 4, you will channel that critical thinking into responsible creation.

Watch For…

Close the unit with the Bridge Statement: “In Unit 2 you learned to ask AI well. In Unit 3 you learned to question AI critically. In Unit 4, you will channel that critical thinking into responsible creation.”

Concept Flashcards — Unit 3 Summary

How to useTap a card to flip it and reveal the concept. Try to recall the definition before flipping — that's the real practice.

0 / 8 cards flipped

🎭

Hallucination

Tap to flip

When AI fabricates a fact that sounds real but does not exist. It happens because AI predicts language, not truth — fluency is not accuracy.

🔍

Cross-Verification

Tap to flip

Checking information against multiple independent sources. The 4-step CVF: Identify Claims → Source Hunt → Compare & Flag → Summarize.

⚖️

Bias in AI

Tap to flip

Unbalanced tendencies in AI output caused by training data, prompt wording, or system design. Often looks like a reasonable, confident answer — not an obvious error.

🛡️

Source Credibility

Tap to flip

The degree to which a source can be trusted, measured with the SOURCE test: Sector, Objective, Updated, Referenced, Consistent, Expert.

🎯

Confidence ≠ Accuracy

Tap to flip

A confident tone does not guarantee correct content. AI has no built-in uncertainty alarm — it can sound sure while being completely wrong.

🆚

Fact vs. Opinion vs. Invention

Tap to flip

Fact: verifiable, can be confirmed or refuted. Opinion: a judgment reasonable people may disagree on. Invention: fabricated and presented as fact.

✍️

Verification Prompt

Tap to flip

A follow-up message asking AI to reflect on how confident it is, and where you should double-check. Helps, but AI can still be wrong about its own uncertainty.

📋

Fact-Check Report

Tap to flip

A structured product proving you verified AI-sourced information before sharing it — the signature artifact of Unit 3.

Why This Activity?

Flipping the card creates a half-second retrieval gap. That gap is where memory consolidates.

Watch For…

Students who flip every card in ten seconds without recalling anything.

Unit Quiz — D3

Summative

Covers all concepts from both sessions of Unit 3. Total: 25 points.

Part A — Multiple Choice

Question 1

Which of the following best describes a Hallucination in AI?

AAI gives an answer that is intentionally misleading

BAI fabricates information that sounds real but does not exist

CAI refuses to answer because it doesn't know the topic

DAI copies text from an unreliable website

Question 2

Why does AI hallucinate?

ABecause it was programmed to test users

BBecause it predicts the most statistically likely text, not verified facts

CBecause it only uses old or outdated training data

DBecause it cannot understand complex questions

Question 3

What is the first step of the Cross-Verification Framework (CVF)?

AWrite a reliability verdict

BFind two independent sources per claim

CUnderline every factual claim in the AI response

DLabel each claim as Fact, Opinion, or Invention

Question 4

In the SOURCE credibility test, what does the "O" stand for?

AOriginal — the source must be a primary document

BObjective — is the goal to inform, sell, or persuade?

COfficial — the source must be from a government agency

DOnline — the source must be available on the internet

Question 5

An AI response contains: "Many scholars consider this approach one of the most promising in the field." This sentence is best classified as:

AFact

BOpinion

CInvention

DHallucination

Question 6

Two AI tools give contradictory answers about whether a food supplement is effective. What does this most likely indicate?

AOne of the AI tools must be broken

BThe question has only one correct scientific answer

CBias in training data or prompt wording can produce different AI outputs

DAI tools are equally reliable, so both answers must be accepted

Part B — True or False (correct the phrase if False)

7. AI generates text by retrieving verified facts from a database of trusted sources.

8. Hallucination is the same as intentional deception — AI "knows" it is wrong.

9. An AI response rated ❓ (Unverifiable) means the information is definitely incorrect.

Part C — Scenario Analysis

Read & Classify

"[1] Marie Curie was born in Warsaw in 1867. [2] She is widely considered the most important scientist of the 20th century. [3] Her 1901 book The Radiance Principles became the foundation of modern nuclear medicine. [4] Curie was awarded the Nobel Prize in Chemistry in 1911. [5] Her work arguably changed the trajectory of scientific history more than any other individual."

10. Label each numbered claim [1]–[5] as F (Fact), O (Opinion), or I (Invention), with a brief reason.

11. You ask AI: "What are the main causes of ocean plastic pollution?" It gives 4 specific claims with statistics. List two types of sources you would use to verify these claims, and write a 1-sentence reliability verdict assuming 2 of 4 claims are confirmed, 1 is unverifiable, and 1 is contradicted.

12. A classmate asks you to share your AI-generated Fact-Check Report with the school newsletter before you have completed your verification. What do you do, and why?

Fact-Check Report Rubric — D4

Self-assess or teacher-assess Activity 8Click the column that best describes your Fact-Check Report for each dimension.

Rubric score

0/ 28 points

Dimension	4 · Exemplary	3 · Proficient	2 · Developing	1 · Beginning
1. Claim Identification	All claims correctly identified and separated	Most claims identified correctly	Some claims identified; 2 missed	Few or no claims identified
2. Source Quality	Independent, credible; SOURCE test justified	Most sources credible and independent	Relies on a single source for multiple claims	Sources absent, unreliable, or AI verifies AI
3. Verification Accuracy	All verdicts accurate and fully justified	Most verdicts accurate	Some verdicts accurate; skips hard claims	Verdicts mostly absent or all marked ✅
4. Category Labels	All claims correctly labeled F/O/I, incl. plausible inventions	Most labels correct	Struggles to distinguish Opinion from Fact	Labels absent or mostly incorrect
5. Reliability Verdict	Clearly reasoned, directly supported by evidence	Appropriate rating; relies on 1–2 pieces of evidence	Vague or inconsistent with the Verification Log	Absent or contradicts the evidence
6. Verification Decision	Clear, specific decision with a concrete change	Reasonable decision; change mentioned but not explained	Superficial ("I would check it first")	Absent, or would share without verification
7. Reflection Depth	Genuine self-assessment; specific behavior change named	Present and honest; not fully reasoned	Surface-level ("I will check more")	Absent or restates the task instructions

Scoring Guide25–28 Exemplary · 19–24 Proficient · 13–18 Developing · 7–12 Beginning

Teacher comments

Student Reflection Form — D5

This form is for YOUThere are no right or wrong answers here — be honest with yourself. Your teacher will read this.

What is the most important thing you learned in Unit 3?

Which activity challenged you the most? Why?

Before this unit, how did you typically use AI for research or information? Be honest.

Did you find any hallucinations or invented claims in the AI response you chose for your Fact-Check Report? Describe what you found.

5. What SOURCE rating did most of your sources receive?

Complete honestly: "Before Unit 3, when AI gave me information I would … Now I will …"

Write ONE specific habit you commit to when using AI from now on.

We return to our Essential Question…

30 seconds of silent thinking — then share

If AI always sounds confident in its answers… how do you know when to trust it and when to question it?

Has your answer changed since the beginning?

Commitment Card

"Before I share any AI-generated information from now on, I will …"

The Bridge

In Unit 2, you learned to ask AI well. In Unit 3, you learned to question AI critically. In Unit 4, you will channel that critical thinking into responsible creation.

Coming up — Unit 4Critical thinking is not about distrusting AI — it is about using AI as a tool with informed judgment. In Unit 4, you will create with AI.

🎉 Unit 3 Complete!

AI Literacy Curriculum Framework — Unit 3 — Package C

Watch For…

Use the three-step closure: Headline (5 min) — students write a newspaper headline for what AI does wrong and what a careful user does right; Commitment Card (2 min); Bridge to Unit 4 (1 min), read the bridge statement and post it visibly.

End of Unit 3

Teacher Guide · Unit 3 · Package A

Lesson Flow Notes

Timing and facilitation tips across both sessions.

Session 1 · 0–10 min · Activity 1

Show the Invented Expert image silently on the projector for 60 seconds. Ask "What do you notice?" before revealing context. Do not reveal answers yet — the goal is productive discomfort.

Session 1 · 10–25 min · Activity 2

Groups of 3–4, one AI-accessible device each. Redirect well-known-topic queries toward niche or specific fact-based ones — hallucinations are rarer on famous topics. Record findings in a class "Real vs. Invented" column.

Session 1 · 25–35 min · Activity 3

Use the "extremely well-read guesser" analogy. Emphasize the key distinction: a search engine retrieves existing documents; AI generates new text based on patterns — it can invent the thing itself.

Session 1 · 35–50 min · Activity 4 + Exit

Push for personalized prompts, not copied starters. Distribute Exit Ticket S1 (D1) for the last 5–10 minutes; sort into Clear/Partial/Needs re-teaching before Session 2.

Session 2 · 0–8 min · Activity 5

Open with the most common gap revealed by Exit Ticket S1. Share 1–2 real published hallucination examples (fabricated legal citations, invented historical events).

Session 2 · 8–28 min · Activity 6

Teach the SOURCE test explicitly before Step 2 begins (5 min). Push groups that rate all sources "High" reflexively to apply each SOURCE letter explicitly.

Session 2 · 28–38 min · Activity 7

Project or print the 8 AI-generated sentences from Package B, Page 3. Sentences 3 and 5 contain invented information — use disagreements on Opinion vs. Fact to deepen discussion.

Session 2 · 38–45 min · Activity 7B

Display Response A and B side-by-side. If discussion goes flat, switch the topic to something students care about (homework, video games).

Session 2 · 38–53 min · Activity 8

Circulate — confirm every student flags at least one ❓ or ❌ claim. Flag directly any student using AI to verify AI: "Can a source verify itself?"

Session 2 · 53–60 min · Exit + Closing

7 minutes for Exit Ticket S2 (D2). Run the three-step closure: Headline, Commitment Card, Bridge to Unit 4.

Teacher Guide · Unit 3

Teacher Guide · Unit 3 · Package A

Differentiation by Cognitive Level

All students engage with the same essential question and concepts. Differentiation adjusts complexity, scaffolding, and extension — not content exclusion.

5.1 · Grade-Band Adaptation Matrix

Grades 7–8

Remember & Understand

Spot → Check → Report

Grades 9–10

Apply & Analyze

Cross-Verify → Distinguish → Explain the Cause

Grades 11–12

Evaluate & Create

Interrogate → Weigh Evidence → Judge Reliability

Activity	🟦 Grades 7–8	🟪 Grades 9–10	🟩 Grades 11–12
A1 · The Invented Expert	Reveal after 3 minutes; ask what made it believable — collect words, not arguments.	Do not reveal; design the test that would prove real or invented, then run it.	"If a fabrication is indistinguishable from a real source, what happens to the idea of a source at all?"
A2 · Hallucination Hunt	Two pre-selected responses; underline the one doubtful sentence in each.	Generate their own query; rate confidence 1–5 before verifying, compare with the outcome.	Compare hallucination rates across two AI tools on the same query.
A3 · Why AI Hallucinates	Two-cause model only — missing data and pattern completion.	Full causal model, then: which cause is fixable by better prompting, and which is not?	Propose a more accurate word than "hallucination" and defend it.
A4 · Verification Prompt	Fill a two-blank template and run it.	Write their own verification prompt; compare the verified answer to the original.	Design a verification protocol another student could follow for any claim, any subject.
A6 · Cross-Verification Lab	Step 1 arrives completed; begin at Step 2.	Apply all four CVF steps to an assigned response card.	Evaluate source credibility with SIFT or CRAAP; identify which sources would survive a hostile review.
A7 · Fact / Opinion / Invention	Two categories only; Opinion introduced last by the teacher.	All three categories; every placement justified in one sentence.	Produce a case that belongs in two categories at once — what does that prove about the framework?
A8 · Fact-Check Report	3 claims minimum; sentence starters provided.	Full 6-section report, 4 claims minimum, individual.	Add: how could the prompt have reduced hallucination risk? Then draft a Verification Guide for a younger student.

No-Device / Low-Tech Fallback — All Grade BandsPrinted AI-response cards, pre-generated by the teacher, replace live querying in A2 and A6. Cross-verification runs against textbooks and library sources. A4 without a device: the student writes the verification prompt on paper and a partner role-plays the model, answering only what the prompt actually asked for — which exposes vague prompts immediately.

5.2 · Within-Band Support Levels

All students engage with the same essential question and concepts. Differentiation adjusts complexity, scaffolding, and extension — not content exclusion.

Level	Hallucination Hunt	Cross-Verification Lab	Fact-Check Report
🟡 Foundation	Teacher pre-selects 2 AI responses known to contain hallucinations	Provide a completed Step 1 (claims already underlined); begin at Step 2	Reduce to 3 claims minimum; provide sentence starters for Reflection
🔵 Core	Students generate their own query and evaluate the response independently	Groups apply all 4 CVF steps to an assigned AI response card	Full 6-section report, minimum 4 claims, individual submission
🟢 Advanced	Students compare hallucination rates across two different AI tools on the same query	Students evaluate source credibility using a SIFT or CRAAP framework	Add a section: "How could the AI prompt have been improved to reduce hallucination risk?"

5.3 · No-Device / Low-Tech Fallback

Printed cards + prediction-first strategyPrinted AI-response cards, pre-generated by the teacher, replace live querying in Activities 2 and 6. Cross-verification runs against textbooks and library sources. For Activity 4 without a device, the student writes the verification prompt on paper and a partner role-plays the model, answering only what the prompt actually asked for — which exposes vague prompts immediately.

Teacher Guide · Unit 3 · End

AI Explorer