How do I use this pack?
The teacher pack complements the main page in teacher mode with everything you need to print. It is built so that you can print it once and reuse it for years. The interactive elements (“Spot the AI mistakes”, “Real or fake?”) run directly on the website and work beautifully on the projector.
Recommended sequence
- Preparation: First read through the main page in teacher mode. For each chapter it gives you learning goals, timing, discussion prompts, quiz answers and anticipated student questions.
- Print the material: Print the worksheets that follow here (ideally a class set minus one, with one held in reserve). Print the parent letter as a class set.
- In the lesson: Follow the timing in teacher mode. The worksheets serve as an activity, not a test. The interactive spot-the-mistake game on the projector opens the lesson.
- Assessment: Optional class test at the end of the unit. Grade using the included rubric.
What's included?
- 4 worksheets — 15–20 min of individual or pair work each, with an answer key directly underneath.
- Class test — 30 points across all topics, with a grading rubric and weighting recommendation.
- 12 homework tasks — 3 tasks per topic across 3 difficulty tiers.
- Parent letter — a template to adapt, with placeholders.
- Curriculum mapping — mapped to media-literacy / computing standards (with DE/AT/CH notes), Bloom matrix, time variants.
A calm, steady stance
The aim of the unit is not to frighten, but to foster informed composure: students should understand how fakes are made, how to recognise them and how to check calmly rather than sharing in a panic. Avoid alarmist examples; emphasise that an alert, questioning eye is the best defence.
Printing tips
- Press Ctrl+P (Windows) or ⌘+P (Mac). In the print dialog you can also choose “Save as PDF”.
- When printing, the navigation and background colours are hidden — the result works in black and white.
- Each worksheet starts on a new page. You can select individual pages in the print dialog.
Lesson overview & learning goals
Overarching competence
Students can recognise, situate and calmly check AI fakes (deepfakes) and disinformation. They understand the technical basics in outline, know the typical tell-tale signs, are aware of their limits, and act prudently and responsibly in everyday life.
Learning goals by Bloom's taxonomy
- L1 — Knowledge: state the word origin of “deepfake” and three kinds of manipulation (image, video, voice).
- L2 — Comprehension: explain how an AI learns from training data and why the intent to deceive is the decisive criterion.
- L3 — Application: apply a checklist to a suspicious piece of media and carry out a reverse image search.
- L4 — Analysis: recognise why visual tells are a hint but not proof, and why checking the origin is more reliable.
- L5 — Evaluation: develop their own reasoned position on “Is it OK to make deepfakes of classmates ‘for fun’?”
- L6 — Creation: design their own awareness campaign (poster, short-video script) for handling fakes calmly.
Double lesson at a glance (90 min)
| Time | Phase | Content & method |
|---|---|---|
| 10 min | Warm-up | “Have you ever seen something online that turned out to be fake?” — buzz-group, collect on the board. |
| 15 min | Development 1 | What deepfakes are and how they are made (ch. 2–3, website in Simple mode). Worksheet 1 or 2. |
| 20 min | Development 2 | “Spot the AI mistakes” on the projector + tell-tale checklist (ch. 4). Worksheet 3. |
| 20 min | Development 3 | “Real or fake?” + checking tools, lateral reading (ch. 5). Worksheet 4. |
| 15 min | Transfer | Protective rules, family code word, the legal position (ch. 6). Discussion. |
| 10 min | Consolidation | Quiz review, guiding principle “informed, not afraid”. |
Differentiation
- Easier level: the class stays in the website's Simple mode; focus on the checklist and the spot-the-mistake game.
- More demanding level: switch to In Detail mode — there, GANs, diffusion models, voice cloning, forensic signals, C2PA provenance and the legal position are placed in technical context.
Materials & equipment
- Projector/smartboard with internet access for the interactive elements.
- Printed worksheets (see below).
- Optionally one device per small group for the reverse image search.
- Important: Do not use real private photos of students; all examples on the website are original illustrations, not real people.
Worksheets
Each worksheet suits about 15–20 minutes of individual or pair work. The answer key is in a fold-out box directly underneath — cut it off before printing the class set, or print double-sided (student side at the front, answer key at the back — do not hand to students).
Worksheet 1 — What are deepfakes?
1. Definition (2 points)
In your own words, explain in no more than 2 sentences what a deepfake is:
2. Word origin (2 points)
Which two English words make up “deepfake” — and what do they mean?
3. Drawing the line (3 points)
Is every edited image a deepfake? Explain the difference between a harmless photo filter and a deepfake:
4. Three kinds (3 points)
Name three kinds of deepfakes and one short example of each:
5. Reflection (2 points)
Who could be harmed most by a faked video of a person? Give your reasoning in 1–2 sentences:
Total points: ___ / 12
🔑 Answer key for teachers — Worksheet 1
1. A deepfake is a piece of media (image, video, audio recording) created or altered with Artificial Intelligence so that it looks convincingly real, even though it never actually happened that way.
2. “Deep learning” (the AI technique behind it) + “fake” (a forgery). Accepted: a sensible explanation of both parts.
3. No. A fun filter (e.g. dog ears) deceives no one about reality. A deepfake aims to deliberately deceive that something is real. Strong answer: names the intent to deceive as the decisive criterion.
4. Accepted: image (a faked photo / “this person does not exist”), video (face swap, a politician saying something invented), voice (a cloned voice on the phone, the “grandparent scam 2.0”).
5. Any well-reasoned answer accepted: private individuals (bullying, reputational damage), politicians (disinformation), families (shock calls), companies (fraud). Marking: a reason given = full marks.
Worksheet 2 — How are deepfakes made?
1. Training data (3 points)
What does an AI need in order to create a convincing deepfake of a person? Explain in 2–3 sentences:
2. Analogy (2 points)
Compare how a deepfake AI learns with something from your everyday life (e.g. how you recognise a person from far away):
3. Match the terms (4 points)
Connect each term with its explanation (draw lines):
| a) Training data | → Reproduce a voice from a few speech samples |
| b) Face swap | → The example recordings the AI learns from |
| c) Voice cloning | → Two networks competing: one fakes, one checks |
| d) GAN | → Put one person's face into someone else's video |
4. Why easier today? (2 points)
It used to take expensive equipment and a lot of skill. Name two reasons why deepfakes are easier to make today:
5. The data question (2 points)
Why are publicly posted photos and videos especially useful for deepfake makers? What does that mean for you?
Total points: ___ / 13
🔑 Answer key — Worksheet 2
1. Many example recordings (training data) of the person — photos, video clips or speech samples from different angles and situations. The more material, the more convincing the fake.
2. Accepted: any fitting analogy. Example: “Like recognising a friend by their walk from far away because I've seen them so often — the AI ‘sees’ a person very often and can then put them back together anew.”
3. a → the example recordings; b → a face into someone else's video; c → reproduce a voice; d → two networks competing.
4. Accepted (two of): apps instead of expensive software · pre-trained models freely available · falling computing costs · a phone is often enough · lots of public photos available.
5. Public content supplies training material for free. Conclusion: post sparingly, don't make profiles fully public, less material = less scope for misuse. Marking: insight + a personal consequence = full marks.
Worksheet 3 — How do I spot a fake?
1. Tell-tale checklist (5 points)
Name five places in an image where AI fakes often go wrong:
2. Why hands? (2 points)
Why does an AI go wrong especially often with hands of all things?
3. Video tell (2 points)
You are watching a suspicious video. What do you watch for in how image and sound work together?
4. More important than looking (3 points)
Why isn't simply looking enough? What is more reliable — and how do you do it in practice?
5. Mini-practice: reverse image search (2 points)
Describe in your own words what a reverse image search is and what it tells you:
Total points: ___ / 14
🔑 Answer key — Worksheet 3
1. Accepted (five of): hands/fingers · teeth · ears & jewellery (asymmetry) · the hair-to-background transition · light & shadow / reflections · garbled text in the background · lip-sync with the audio (video).
2. Hands are complex and appear in countless positions; the AI easily mistakes the number of fingers or bends them. Accepted: words to the effect of “a difficult, highly variable structure”.
3. Whether the lip movement and the voice are exactly in sync — a slight offset is a warning sign. Also: flickering at the edges of the face.
4. Visual tells are only a hint, not proof, and become rarer with every AI generation. More reliable is checking the origin: a reverse image search + comparing several independent sources. Full marks: names “not proof” + origin + a concrete method.
5. You upload an image (or its address) to a search engine instead of searching for words. You see where else the image appears, when it first showed up and in what context — that's how you expose old or out-of-context photos.
Worksheet 4 — Why fakes work & what I can do
1. Headline check (3 points)
Read the headline and decide: real or fake? Name three warning signs you can spot it by.
“SHOCK: Famous star is giving away their entire fortune today — only those who share THIS link get a cut!”
Real or fake?
Three warning signs:
2. Why do we share so fast? (2 points)
Why do fakes often spread faster than sober news?
3. Three checking tools (3 points)
Name the three tools that almost always help before you believe or share something:
4. The shock call (3 points)
Someone calls, sounds like a family member and urgently needs money. What do you do? Describe your steps:
5. Responsibility & the law (2 points)
Is it OK to make and share a deepfake of a classmate “for fun”? Give your reasoning:
Total points: ___ / 13
🔑 Answer key — Worksheet 4
1. Fake. Warning signs (three of): capitalised “SHOCK” & strong emotion · pressure to share a link immediately · the promise of money · no verifiable source · found nowhere else on reputable outlets.
2. Fakes work through emotion and speed: whatever makes us angry, anxious or gleeful gets shared faster — without thinking. In echo chambers, like-minded people reinforce each other.
3. Pause (don't share straight away) · reverse image search (where does the image come from?) · compare several sources (do reputable outlets report the same thing?).
4. Stay calm, do not send money straight away. Hang up and call back on the known number. Ask a question only the real person can answer (a code word). If in doubt, involve other family members.
5. No. Even “for fun” it can be serious and a criminal offence: the right to one's own image, personality rights, possibly insult/defamation. It can seriously harm the person affected. Full marks: a clear position + at least one reason (legal or human).
Class test 30 points
Final test across all topics. Suggested time: 30–40 minutes. Answer key and grading rubric follow directly underneath.
Part A: Multiple choice 1 pt each · 6 pts
Tick the one correct answer in each case.
1. What makes a piece of media a deepfake?
- A fun filter on a selfie
- An AI creates or alters it so that it looks convincingly real
- It was taken with an expensive camera
- A hand-painted portrait
2. What does an AI need above all to create a convincing deepfake?
- The person's phone number
- Nothing — the AI just guesses everything
- Many example recordings (training data) of the person
- A very fast internet connection
3. Which spot is especially often wrong in AI images?
- A blue sky
- A white wall
- The hands
- A plain single-colour background
4. What is the most reliable tool against a fake?
- Only looking at the hands
- Checking the origin (reverse image search, several sources)
- Sharing the content immediately and asking afterwards
- Trusting your gut feeling
5. How do you best protect yourself from a call with a cloned voice?
- Trust the voice — it does sound real
- Transfer money immediately to help
- Hang up and call back on the known number
- Just ignore the call and never answer the phone again
6. What does “lateral reading” mean?
- Reading a page especially slowly and thoroughly from top to bottom
- Checking in several tabs who is behind a source
- Reading texts from right to left
- Reading only the headline and sharing
Part B: Short answer 2 pts each · 8 pts
7. Explain in 1–2 sentences what a deepfake is:
8. Name four typical tell-tale signs of an AI image fake:
9. Why is simply looking not solid proof of a fake?
10. Name the three checking tools to use before you believe or share something:
Part C: Application 10 pts
11. You receive a video from a class chat that supposedly has a teacher saying something embarrassing. Describe step by step how you would check whether it is real (5 pts):
12. An older relative gets a call in the voice of her grandson, who urgently needs money. Explain to her in 3–4 sentences what she should do and why (5 pts):
Part D: Reflection 6 pts
13. Discuss in 5–8 sentences: “Should it be allowed to create and share deepfakes of real people ‘for fun’?” Argue both for AND against, then take a reasoned position of your own.
🔑 Answer key for teachers — Class test
Part A: 1b · 2c · 3c · 4b · 5c · 6b
Part B:
7. A piece of media (image/video/audio) created or altered with AI that looks convincingly real but never actually happened that way.
8. Four of: hands/fingers · teeth · ears & jewellery · the hair-to-background transition · light/shadow · garbled background text · lip-sync offset.
9. Because AI fakes get better with every generation and tells become rarer; they are only a hint. Certainty comes only from checking the origin.
10. Pause · reverse image search · compare several independent sources.
Part C:
11. Ideal answer (≈ 1 pt each): don't share/believe straight away · reverse-search single frames · watch lip-sync with audio and the edges · check whether reputable sources report it · involve a teacher/trusted adult rather than spread it. Bonus: recognises the potential harm to the person affected.
12. Ideal answer: stay calm, do not pay straight away (1) · hang up and call back on the known number (2) · a question / code word only the real person knows (1) · explain that voices can be cloned with AI (1).
Part D:
13. Full marks: a nuanced discussion with clear pro/con arguments and a reasoned position of one's own. Pro (rarely, with reservations): artistic/satirical use with labelling and consent. Con: personality rights, the right to one's own image, possible serious harm, criminal liability, broken trust. The depth of the argument is graded, not the position.
Grading rubric
| Points | Grade (DE/AT) | Grade (CH) | Assessment |
|---|---|---|---|
| 27 – 30 | 1 / Very good | 5.5 – 6.0 | Complete understanding, independent reflection, confident use of the checking methods. |
| 23 – 26 | 2 / Good | 4.5 – 5.0 | Confident knowledge, minor gaps, reflection present. |
| 18 – 22 | 3 / Satisfactory | 3.5 – 4.0 | Basics understood, application/reflection superficial. |
| 14 – 17 | 4 / Sufficient | 3.0 – 3.4 | Key terms present, many gaps in application. |
| 0 – 13 | 5 / Insufficient | < 3.0 | Basic terms not understood — extra support recommended. |
Grades follow the DE/AT 1–5 and CH 1–6 systems; map the point bands to your own grading scale as needed.
Weighting recommendation
- Knowledge (Parts A + B): 14 pts — testable with a clear right/wrong line.
- Application (Part C): 10 pts — room for partial marks per task (grade the steps individually).
- Reflection (Part D): 6 pts — grade the depth of the argument here, not the position taken.
Homework collection — 3 difficulty tiers
3 tasks per topic: Easy Medium Challenging. Answer hints are in expandable details directly underneath (collapsible on screen, always open when printed).
Topic 1 — What are deepfakes?
Explain to an adult (parents, grandparents) in your own words what a deepfake is. Write down in 3 sentences what you said.
🔑 Answer hint
Accepted: a correct short definition (an AI fake of an image/video/voice that looks real) and at least one example. Bonus: no unexplained jargon.
Make a table with 5 media examples × the columns “real / edited / deepfake?” and a short reason. Find your own examples (no real private people).
🔑 Answer hint
Full marks: all columns filled in, with the reason naming the intent to deceive as the distinguishing factor in each case. Example entry: “advertising photo, retouched | edited | no, it isn't trying to deceive about reality”.
Research a real reported deepfake case (e.g. from the news). Summarise on 1 A4 page: what happened, who was affected, how was it exposed?
🔑 Answer hint
Accepted: a reputable source named, a factual summary, clearly separated: what happened / harm / how it was exposed. Bonus: own judgement on why the case is instructive. Note: watch for sensitive content; provide preset topics if needed.
Topic 2 — How are deepfakes made?
Describe in 4–5 sentences how an AI “learns” a face. Use an analogy of your own.
🔑 Answer hint
Accepted: the AI sees many recordings (training data), spots patterns, can then reassemble the face anew. A fitting analogy (e.g. recognising a friend by their walk).
Try Google's Teachable Machine: train a mini-AI for 2 categories (e.g. “thumbs up” / “thumbs down”) with 30 images each. Describe in 4–6 sentences what you observed.
🔑 Answer hint
Accepted: a documented experiment. Findings: more examples = better; worse recognition against a different background; surprisingly good even with 30 images. Bonus: reflection on bias from too little variety — a bridge to deepfake training data.
Explain the difference between a GAN and a diffusion model in your own words (max. 1 A4 page). Use the main page in “In Detail” mode as your source.
🔑 Answer hint
GAN: generator vs. discriminator competing, fast but unstable. Diffusion model: works an image out of noise step by step, more stable, steerable via text. Full marks: both correct and in the student's own words, with one pro/con per method.
Topic 3 — Spotting fakes
Design a small “tell-tale checklist” as a memo card (postcard size) with at least 6 places where AI images often go wrong.
🔑 Answer hint
Accepted: 6+ tells from the checklist, clearly laid out. Bonus: poster-/memo-worthy, own symbols.
Carry out a reverse image search on a (harmless) image of your choice. Document: where did the image originally come from, when did it first appear?
🔑 Answer hint
Accepted: a documented search with a result (first source, date, context). Bonus: reflection on how an out-of-context image would be exposed this way. Note: only harmless images, no private people.
Research what C2PA / Content Credentials are, and explain on 1 A4 page why “proving the origin” could matter more in the long run than “spotting the fake”.
🔑 Answer hint
C2PA = cryptographically signed provenance metadata (a “digital nutrition label” for a piece of media). Argument: detectors lose the cat-and-mouse race, while provenance verifiably proves the source. Full marks: concept correct + a coherent argument + a limit named (a screenshot strips the signature).
Topic 4 — Disinformation & staying safe
Write your personal “4 rules for handling suspicious content” on a sheet (big enough to put up).
🔑 Answer hint
Accepted: rules that show composure, e.g. “Check first, then share” · “Check the source” · “Compare several sources” · “If in doubt, ask”. Bonus: poster-worthy.
Agree an emergency code word against shock calls with your family and write a short message (for a family group) explaining why it helps.
🔑 Answer hint
Accepted: a sensible code-word concept + an explanation that cloned voices can be deceptively real and a code word tells a genuine emergency call from a faked one. Bonus: a note to “call back on the known number”.
Create an awareness campaign for your year group: 1 poster or a 60-second short-video script on “Real or fake? — check calmly instead of sharing fast”.
🔑 Answer hint
Accepted: audience-appropriate (no jargon), a clear message, at least one concrete checking method. Full marks: attractively designed, a calm tone rather than scaremongering, includes a call to action.
Parent-letter template
You can adapt this template to your school and class. Replace the [placeholders in purple] with your own details and print the template for your class.
[Your school]
[Address]
[Date]
To the parents of class [Year X]
Subject: Teaching unit “Deepfakes & disinformation — media literacy”
Dear parents,
over the coming [weeks / double lesson] your child's class will be looking at deepfakes and disinformation. With Artificial Intelligence it is now possible to fake images, videos and even voices in a deceptively real way. We want to enable your children to recognise such content calmly, check it critically and handle it prudently — without fear, but with an alert eye.
What your child will learn:
- What a deepfake is — and what sets it apart from harmless photo editing.
- How such fakes are made technically (training data, AI models, cloned voices).
- How to spot fakes — and why checking the origin matters more than simply looking.
- Why disinformation spreads so quickly and how to expose it with simple steps.
- How to protect yourself from scams — such as shock calls with a cloned voice (the “grandparent scam 2.0”).
- Responsibility: why you don't create fakes of other people — not even “for fun”.
Material and source: We use the freely available learning platform deepfakes-verstehen.webhoch.com, provided by the Austrian agency Webagentur Hochmeir e.U. under a free licence (CC BY 4.0). All images shown are original illustrations — no real people. The content is prepared in an age-appropriate way for [Years 7–10].
How you can support at home:
- Talk with your child about content that seems “too good/too bad to be true” online.
- Agree on a family code word for genuine emergencies — as protection against calls with a cloned voice.
- Practise together: with suspicious calls, hang up and call back on the known number.
- Encourage your child to pause briefly and check the source before sharing.
Important notes:
- No private photos of the students are used in the lesson; all examples are illustrations.
- We deliberately treat the topic calmly and without scaremongering — the aim is competence, not fear.
- If you have questions or concerns: [Your email address]
We are glad that your child is gaining this important future skill — and we appreciate your support along the way.
Kind regards,
[Your name]
[Role / class teacher]
Curriculum mapping
This unit covers core learning areas of media literacy. In general, it maps to lower-secondary media-literacy / computing curricula (≈ Years 7–10, ages 12–16). The mapping below is adaptable to the Austrian, German and Swiss education standards for lower secondary.
Curriculum references (general lower-secondary media literacy / computing; AT: Digitale Grundbildung; DE: media-literacy framework / KMK “Bildung in der digitalen Welt”; CH: Lehrplan 21, “Medien und Informatik” module)
| Subject / area | Competence / standard | Topics |
|---|---|---|
| Computing / digital literacy | Explain how AI systems work in outline; identify synthetic media | 1–3 |
| Media literacy | Analyse & evaluate media; check sources critically; recognise disinformation | 3–4 |
| Language / English | Argue, discuss, justify positions; manipulation in language & image | 4 |
| Ethics / religion | Responsibility in a digital world; personality rights | 4 |
| Civics / citizenship | Forming opinions, the public sphere, disinformation & democracy | 4 |
Competence matrix (Bloom's taxonomy)
| Level | Name | Example from this unit |
|---|---|---|
| L1 | Knowledge | Name terms (deepfake, training data, voice cloning, reverse image search) |
| L2 | Comprehension | Explain the difference “photo editing vs. deepfake” |
| L3 | Application | Apply the tell-tale checklist, solve “Spot the AI mistakes”, do a reverse image search |
| L4 | Analysis | Spot warning signs in “Real or fake?”; analyse “a tell ≠ proof” |
| L5 | Evaluation | Develop their own position on “deepfakes for fun?” |
| L6 | Creation | Design an awareness campaign / memo card (Homework 3.1 / 4.3) |
Time allocation
| Variant | Breakdown | Recommended for |
|---|---|---|
| Double lesson (90 min) | Intensive workshop, all topics in excerpts, one worksheet | Project day, cover lesson, taster course |
| Weekly module (3 × 50 min) | Day 1: What & How (topics 1–2); Day 2: Spotting fakes (topic 3); Day 3: Disinformation + test (topic 4) | Standard lessons across one week |
| 5-week project (5 × 90 min) | One double lesson per topic, own research projects, final presentations | In-depth study, project work, gifted-and-talented support |
| Project week (5 × 4h) | Days 1–4: deep dive into the topics with own research · Day 5: campaign presentations + test | “Media literacy” themed week, IT summer camp |
Where do I connect this?
- Helpful beforehand: basic knowledge of the internet / smartphone / social media. Familiarity with the term “app”.
- Follow-on links: AI basics (sister site KI verstehen), data protection/GDPR, source criticism in history/language lessons, cyberbullying prevention.
- Beyond the classroom: Teachable Machine (train your own mini-AI), C2PA / Content Credentials (proof of origin), fact-checking portals (e.g. correctiv, Mimikama).