Every previous generation learned "don't believe everything you read." Your child's generation needs the harder version: don't believe everything you see and hear. AI can generate a photo of an event that never happened, a voice saying words never spoken, and an essay full of confident wrong answers — and it keeps getting better at all three.
The temptation is to protect kids with fear. The better move is to give them a game they can win.
First, the framing (one conversation)
Tell them plainly: "Computers can now make pictures and videos of things that never happened. Not because computers are evil — because people tell them to. So we're going to become detectives." Detectives, not victims. Kids who feel skilled engage; kids who feel scared just get anxious — or worse, stop telling you what they saw.
If you want the actual words, here's a version that lands with most kids:
You: "Can I show you something cool and slightly weird? Computers can now make photos of things that never happened. See this picture of a cat in a spacesuit? No cat, no suit, no camera. A computer made it up because somebody typed a sentence."
Kid: "So it's fake?"
You: "Completely fake. And here's the thing — some fakes are just for fun, like this one. But some are made to trick people. So this family is going to train like detectives. Detectives don't get scared of tricks. They get good at catching them."
Tune it by age. With a 5–7-year-old, stop right there and keep the word "pretend" — "some pictures are real and some are pretend, and we're learning to tell which is which." With an 8–10-year-old, add the why: "people make trick pictures to get clicks, to sell things, or to make someone look bad." With an 11–14-year-old, drop the cat and lead with respect: "You're going to see fakes this week — you've probably already seen some. I don't want you to be the kid who gets fooled, and definitely not the kid who forwards the fake. Want to see if you can beat me at spotting them?" Competition beats lecture at this age, every single time.
If they've done the broken-telephone hallucination game, connect it: a deepfake is a hallucination on purpose.
The weekly ritual: Spot the Fake (10 minutes)
Once a week, look at a handful of images together — some real photos, some AI-generated (there are quiz sites for exactly this, or generate your own examples in a supervised session). Everyone votes: real or fake? Keep score across weeks.
Two things happen. Your child gets measurably better — and, more importantly, they watch themselves get fooled sometimes. A child who has been fooled in a game, safely, with a laugh, is far better calibrated than one who's been told "you can always tell." Because increasingly, you can't. Media-literacy educators consistently find that this kind of inoculation — getting tricked in a low-stakes setting and then seeing exactly how — builds more durable skepticism than any amount of warning. The lecture wears off; the memory of "wow, that one got me" doesn't.
Rotate the medium: images one week, a too-good-to-be-true "news" headline the next, an AI-written vs. human-written paragraph after that.
How to run week one (a concrete walkthrough)
The ritual takes ten minutes; the first setup takes about twenty. Here's the whole thing.
1. Gather your material (15 minutes, before the game). You need about six items: three real, three generated. Two easy sources, no special tools required. First, real-or-AI quiz sites — search "real or AI image quiz" and you'll find several free ones built exactly for this, with difficulty levels and answer reveals. Second, make your own deck: pull three genuinely real photos (your own camera roll works — a holiday shot, a pet, a meal) and generate three AI images yourself in a few minutes. Match the subjects roughly — a real dog against a generated dog is a fair fight; a real dog against a generated dragon is not.
2. Set the table. One screen, whole family, everyone votes on every item — including you. Votes get written down or said on the count of three, before any discussion, so nobody follows the leader.
3. Reveal and debrief. After each answer, one question: "What tipped you off?" — and for wrong guesses, the better question: "What made it so convincing?" Celebrate the misses. "Ooh, it got you! It got me too. Look at it again — now can you see it?" Being fooled is data, not failure, and how you react to their first miss decides whether they keep playing honestly.
4. Start the scoreboard. Fridge, whiteboard, back of a notebook — anywhere visible.
The scoring system kids actually care about
- 1 point for a correct call.
- +1 bonus for giving a reason before the reveal ("fake, because the fence melts behind his elbow") — right or wrong, the reason earns the point, because reasons are the skill.
- Streaks: five correct in a row earns something small but real — detective badge, dessert pick, or the right to choose next week's category.
- The Fooled-Us Hall of Fame: any fake that fools the entire family gets a name and goes on the list. Kids love this — it turns the scariest fakes into trophies.
- Parents play for real. Your kid will beat you some weeks. Let it stand. A child who has out-detected an adult stops thinking of this as a kid lesson.
Age-band variations
- Ages 5–7: three or four images, big contrasts, framed as "real or pretend?" Stick to animals, food, and places — nothing scary, no people in distress. The win condition at this age isn't accuracy; it's the vocabulary. A five-year-old who says "that's a pretend picture" has learned the lesson.
- Ages 8–10: the full game as described above. Add fake headlines around month two: read one real and one absurd-but-plausible headline, vote, discuss.
- Ages 11–14: harder material. Screenshots of things actually going around that week (you curate them first), voice clips if a quiz site offers them, AI-written vs. human-written paragraphs. Better yet, hand over the deck some weeks: they source the images and try to fool you. The tester always learns more than the tested.
The three questions (the checklist that sticks)
When something surprising shows up on a screen, detectives ask:
- "Who's showing me this?" A news site with a name and an address, or a random account?
- "Does anyone else say it too?" Real events leave many footprints. One-source wonders are suspects.
- "What does the picture want me to feel?" Fakes are usually engineered for a spike — outrage, fear, "no way!" The stronger the feeling, the slower the belief.
Three questions, no tech knowledge required, and they work on AI fakes, old-fashioned Photoshop, and plain human lies alike. Put them on the fridge next to the story machine book.
Run the drill out loud once so it becomes a habit rather than a poster. Next time anything surprising crosses a screen you're both looking at, narrate it: "Okay, detective, question one — who's showing us this? Hmm, an account with a string of numbers in the name. Question two — is anyone else reporting it? Let's check. Question three — what did it want us to feel when we saw it?" Two or three walkthroughs and kids start doing it unprompted, usually with great ceremony.
What about the artifacts — six fingers and garbled text?
Teach them as bonus clues, not the method: weird hands, melted text in backgrounds, earrings that don't match, physics that's slightly off. But be honest that these clues expire — every generation of models fixes the giveaways. The three questions don't expire.
The house rule that matters most
"Shocked? Show someone before you share." The single most damaging thing kids do with fakes isn't believing them — it's forwarding them in the first thirty seconds. Making "show a parent first" the reflex buys the pause where the three questions can happen.
The voice on the phone: clone scams and the family codeword
One more detective skill, and this one protects the whole family — including the grandparents.
AI can now clone a voice from a short sample, and there's a well-known scam pattern built on it, sometimes called the "grandparent call": the phone rings, a voice that sounds exactly like a family member says something urgent — an accident, an arrest, trouble while traveling — and asks for money, gift cards, or account details, fast, and usually with "don't tell anyone else." The target is whoever picks up: a grandparent, a parent, or a kid home alone. The tell is never the voice anymore. The tell is the shape of the call: urgency + money + secrecy. Real emergencies almost never come packaged with "don't tell Mom."
The defense is old-fashioned and works: a family codeword. Pick one together — a word or short phrase that would never come up naturally, isn't guessable from your social media, and everyone can remember. Then set the rule: any call that's urgent and asks for money or secrecy gets one response — "What's the codeword?" A real family member answers instantly. Anyone else stalls, deflects, or hangs up. Backup rule if the codeword is forgotten in the moment: hang up and call the person back on their real number. Real emergencies survive a call-back. Scams never do.
Here's the script for introducing it, calm and matter-of-fact:
"One more detective skill. Computers can copy voices now — well enough that a call could sound like me, or Grandma, and not be us. So we have a family codeword: ___. If anyone ever calls sounding like family and asks for money, or says 'don't tell anyone,' you ask for the codeword. No codeword? Hang up and call the real number. Even if it really sounds like us — especially if it really sounds like us."
Then actually brief the grandparents — they're the most-targeted people in the family and the least likely to have heard of voice cloning. Practice it once at dinner, laugh about the codeword you picked, and move on. No sensationalism required; this is a seatbelt, not a horror story.
The flip side: their own creations
A child who makes AI images (in supervised sessions) should learn the mirror rule: we don't make fakes of real people, ever. Classmates, teachers, siblings — real faces are off-limits, full stop. Kids who create understand faster than kids who only consume — which is exactly why this pathway had them build a story machine before this conversation.
"What if my kid..." — troubleshooting the real moments
...shares a fake anyway? It will happen; it happens to adults constantly. Skip the lecture and go straight to repair: "Okay — it got you. It's a good one. First job: send the correction to the same people you sent the fake to. That's what honest people do when they get fooled." Then, later, the reflection: "What did it make you feel right before you hit share?" The rule survives because breaking it costs a small repair, not a shame spiral — a kid who fears your reaction more than the fake will just stop showing you things.
...gets anxious about deepfakes? Some kids hear "you can't trust what you see" and quietly spiral. Shrink the claim back to true size: "Almost everything you see today is still real. Fakes are a small slice — and you now have detective skills most adults don't have." Then convert worry into reps: anxious kids often become the best players, because the game gives the fear something to do. If the anxiety is specific — a fear of fakes being made of them — take it seriously, restate the mirror rule, and tell them plainly that making fakes of real people breaks your family's rules, the platforms' rules, and in many places the law, and that they should always come to you first if it ever happens.
...insists "I can always tell"? The best cure is a loss. Pull up the hard set on a quiz site and let the game do the arguing. One honest defeat teaches calibration faster than ten warnings — deliver it with a grin, not an "I told you so."
...shows you something genuinely disturbing? First words: "Thank you for showing me." That's the house rule working exactly as designed, and it must be rewarded even when the content is awful. Deal with the feeling first, run the verdict second, and only then decide together what to do about it.
...is a teen who thinks the game is babyish? Promote them. They curate next week's round, they try to fool you, they run the debrief for a younger sibling. Gamemaster is a status upgrade; player was a chore. Teens rarely refuse a role that makes them the authority in the room.
Frequently asked questions
What age should we start Spot the Fake?
Around five, in the screen-free "real or pretend?" form with gentle images. The full weekly game with scoring works from about eight, and the voice-codeword conversation is worth having as soon as your child answers a phone — which is younger than most of us expect.
Where do I actually find real-vs-fake image pairs?
Two sources cover everything: free "real or AI" quiz sites (search that exact phrase — several exist, with difficulty levels and answer reveals), and your own deck — real photos from your camera roll paired against images you generate yourself in a supervised session. Matching subjects keeps it a fair fight.
Should I show my child an actual deepfake video?
For 11-and-up, one clearly-labeled example with a debrief is genuinely useful — abstract warnings don't stick the way "I watched it and believed it for four seconds" does. Under 11, images and headlines teach the same skill with less to process.
My child made an AI image of a classmate "as a joke." Now what?
Treat the first offense as a rule they hadn't fully absorbed, not a character verdict: delete it together, restate the mirror rule ("real faces, never — even as a joke"), and talk about how the classmate would feel finding it. If it was shared, the repair includes an apology. Escalate only on repetition.
Do AI-detection apps solve this?
No. Detection tools are unreliable and locked in an arms race with the generators — they miss fakes and flag real photos. Teach the three questions instead; they work regardless of which side of the arms race is winning this month.
What makes a good family codeword?
A word or phrase that never comes up in normal conversation, isn't discoverable from your family's social media, and that a seven-year-old and a grandparent can both remember under stress. Silly beats serious — silly gets remembered. Say it out loud once as practice, then keep it off any screen.
Next in the pathway: the question every family hits eventually — where's the line between AI helping with homework and AI doing it?
