How do I know if what the AI told my kid is true?

Reading harder does not work. Reading sideways does

In a study of 45 experienced internet users — ten PhD historians, ten professional fact checkers and twenty-five Stanford undergraduates — Wineburg and McGrew watched people judge live websites out loud. The historians and the students "read vertically, staying within a website to evaluate its reliability", and "often fell victim to easily manipulated features of websites, such as official-looking logos and domain names". The fact checkers did something different: they "read laterally, leaving a site after a quick scan and opening up new browser tabs in order to judge the credibility of the original site". They "arrived at more warranted conclusions in a fraction of the time".

Two things in that are worth sitting with. The fact checkers were faster, not slower — this is not a tax on your evening. And subject expertise did not save the historians. People who had spent careers evaluating evidence were fooled, because they were using the wrong move, not because they lacked knowledge. That should lower your expectations of "my child is smart, they will spot it" quite sharply.

Translated for a chatbot, lateral reading is almost absurdly simple: do not interrogate the answer, open one other place and see whether it says the same thing. A chatbot is a website with no logo and no domain name to be fooled by, so the sideways move is the only honest one left.

This does not arrive on its own

It is tempting to assume children who grew up online absorb this. The largest study of it says otherwise. Breakstone and colleagues gave six tasks to 3,446 US high-school students with a live internet connection, and report flatly: "The students struggled on all of them." Ninety-six per cent never discovered that a site claiming to publish factual climate reports had ties to the fossil fuel industry. Two thirds could not tell news stories from ads. More than half took an anonymously posted video shot in Russia as "strong evidence" of US voter fraud.

What they did instead is the useful part. Students "were often duped by weak signs of credibility: a website’s ‘look’, its top-level domain, the content on its About page, and the sheer quantity of information it provided." A fluent, well-organised, confident, long answer is exactly that surface — and an AI produces it whether or not there is anything behind it.

Hold the limits straight, though: these were high-school students, assessed on websites rather than chatbots. It is not a measurement of what a nine-year-old can do. It is strong evidence that this skill does not arrive with age or screen time, which is the only thing this page needs it for.

Tools that make a child think rather than hand over an answer · Explaining to a child what the thing actually is · The same check, when the thing to verify is a picture or a video · And the half that comes first: asking it well

Four ways it goes wrong, and the check that catches each

Useful because the checks are different. A parent who only watches for invented facts will miss the two failures that matter most at homework age.

The habit, in three questions

1. 1. Before: "what do you think the answer is?" Have your child commit to a guess before the AI answers. This is the one with experimental support behind it: Buçinca and colleagues found that cognitive forcing — making a person commit before seeing the machine’s suggestion — significantly reduced overreliance, where simply adding explanations did not. There is a catch, below.
2. 2. During: "how would we know if that is wrong?" Not "is it right", which invites a yes. This asks what evidence would settle it, which is the question a fact checker asks and the one that produces the sideways move rather than a harder stare at the same page.
3. 3. After: open one other tab. One. Not a research project — the fact checkers in the study were faster than the people who stayed put. The habit you are building is that an answer is a starting point, and that leaving it is normal rather than an accusation against the tool.
4. 4. Once a week, catch it out on purpose. Ask it something you already know, or something with a false premise baked in, and see what happens. This is the step children actually enjoy, and it does more for a child’s scepticism than any number of warnings — because they found it themselves.

The uncomfortable finding, stated plainly

The study behind step one found something that most advice on this subject quietly omits. Buçinca and colleagues tested interventions designed to make people engage with an AI’s suggestion rather than accept it, with 199 participants. Cognitive forcing worked: it "significantly reduced overreliance compared to the simple explainable AI approaches". But "people assigned the least favorable subjective ratings to the designs that reduced the overreliance the most."

In other words: the thing that works is the thing people like least. Which means if you install the "guess first" habit at your kitchen table and your child finds it annoying, that is not a sign you are doing it wrong. It is the expected result, and the measured one. Worth knowing before you abandon it on day three because of the sighing.

The limits, since this page is about not taking confident claims at face value: those were adults in a controlled study on a decision task, not children on homework. The mechanism is what transfers. Nobody has measured this on a nine-year-old at a kitchen table, and this page is not going to pretend otherwise.

The research under this page

Three studies, none of which studied children using a chatbot — the population this page is actually about. What each contributes, and where it stops, is set out below.

1. Wineburg & McGrew, 2019 — Lateral reading and the nature of expertise: reading less and learning more when evaluating digital information. Teachers College Record, 121(11), 1–40. (45 experienced internet users — 10 PhD historians, 10 professional fact checkers, 25 Stanford undergraduates — thinking aloud as they judged live websites. "Historians and students often fell victim to easily manipulated features of websites, such as official-looking logos and domain names. They read vertically, staying within a website to evaluate its reliability. In contrast, fact checkers read laterally, leaving a site after a quick scan and opening up new browser tabs in order to judge the credibility of the original site. Compared to the other groups, fact checkers arrived at more warranted conclusions in a fraction of the time.") The central technique. Fact checkers read laterally and reached better conclusions faster; historians and students read vertically and were fooled by surface features. Forty-five adults evaluating websites, so the strategy transfers and nothing about children’s execution of it is established. What it does not show: A small expert/novice study of 45 adults evaluating websites, not children evaluating a chatbot. What transfers is the STRATEGY, and the finding that subject expertise did not protect the historians; the study says nothing about how well a nine-year-old can execute it.
2. Breakstone, Smith, Wineburg, Rapaport, Carle, Garland & Saavedra, 2021 — Students’ civic online reasoning: a national portrait. Educational Researcher, 50(8), 505–515. (3,446 US high-school students with a live internet connection, on six constructed-response tasks. "The students struggled on all of them." 96% never learned that a site claiming to disseminate factual climate reports had ties to the fossil fuel industry; two thirds could not distinguish news stories from ads; more than half believed an anonymously posted video shot in Russia was "strong evidence" of US voter fraud. Students "were often duped by weak signs of credibility: a website’s ‘look’, its top-level domain, the content on its About page, and the sheer quantity of information it provided.") Used for "this does not arrive on its own", and for the list of weak credibility signals that a chatbot supplies in abundance. High-school students on websites; we quote no figure as if it described a seven-to-twelve-year-old. What it does not show: High-school students, and about websites rather than chatbots. It is evidence that this skill does not arrive on its own with age or screen time — not a measurement of what a seven-to-twelve-year-old can do, and we quote no figure as if it were.
3. Buçinca, Malaya & Gajos, 2021 — To trust or to think: cognitive forcing interventions can reduce overreliance on AI in AI-assisted decision-making. Proceedings of the ACM on Human-Computer Interaction, 5(CSCW1), 1–21. (N = 199. "People supported by AI-powered decision support tools frequently overrely on the AI: they accept an AI’s suggestion even when that suggestion is wrong", and adding explanations "does not appear to reduce the overreliance". Cognitive forcing — making the person commit to their own answer before they see the AI’s — "significantly reduced overreliance compared to the simple explainable AI approaches". The trade-off: "people assigned the least favorable subjective ratings to the designs that reduced the overreliance the most".) The basis for the guess-first step, and for the honest warning that the most effective design was the least liked. Adults, a controlled decision task, N = 199 — cited for the mechanism, never as a measured effect on a child. What it does not show: Adults in a controlled study on a decision task, not children on homework. Cited for the mechanism and for its uncomfortable trade-off — the intervention that worked best was the one people liked least — and never as a measured effect on a child.
4. Google For Families Help, read 31 October 2026 — Guide your child’s Gemini Apps experience. (Google’s own page for parents: it states that a parent may give a child under 13 access via a supervised account, that "Our filters try to limit inappropriate content from appearing, but they’re not perfect", that Gemini "can hallucinate and present inaccurate information as factual", and that a child should understand "it’s an AI tool, not a person to confide in". Its reporting section adds that if a child "encounters inappropriate or incorrect content, you or your child can report it", and that "You can report it by submitting feedback or your child can mark it as a bad response".) Quoted only to show the vendor is not in dispute: Google’s own page for parents says Gemini "can hallucinate and present inaccurate information as factual". Re-read 31 October 2026. What it does not show: A vendor’s own help page, not an independent assessment, and it describes the product on the date it was read. It is strong evidence about what Google itself tells parents and weak evidence about what the product actually does. On the reporting half specifically: the page names a route and makes no claim at all about what happens after a report is filed, so nothing here supports telling a parent that reporting will change the model’s behaviour — only that the vendor asks to be told.

Questions parents ask

How can I tell if an AI answer is wrong?

Leave the answer and look somewhere else — do not analyse it harder in place. In a study of professional fact checkers, historians and undergraduates, only the fact checkers "read laterally, leaving a site after a quick scan and opening up new browser tabs", and they reached better-warranted conclusions in a fraction of the time. The others, including subject experts, were taken in by surface signals of credibility. One other source beats any amount of re-reading.

Do kids know when AI is making things up?

Mostly not, and age alone does not fix it. In the largest assessment of its kind, 3,446 high-school students struggled on every task, with 96% missing a climate site’s fossil-fuel ties and two thirds unable to separate news from ads. They were persuaded by look, domain, About pages and sheer volume of information — all things an AI answer supplies by default. Treat verification as something explicitly taught, not absorbed.

What is the single best habit to teach?

Have your child say what they think the answer is before the AI says anything. Research on AI-assisted decisions found this kind of "cognitive forcing" significantly reduced overreliance, while simply showing explanations alongside the answer did not. Expect mild resistance: in the same study, the designs that cut overreliance the most were the ones people rated least favourably.

Is it bad for my child to trust AI?

Trusting it is not the problem; trusting it by default is. The useful framing is not "the AI lies" — which is both wrong and unhelpfully scary — but "an answer is a first draft that someone has to check". That version is true, it is how adults should treat it too, and it leaves your child with a skill that applies to every source they will meet rather than a fear of one product.

An AI that asks before it answers

The habit on this page is hard to drill at home because a general assistant answers instantly, every time. Luchi is built to ask the child what they think first and to make them explain what they made — and it writes you a short note in their own words, so you can see the thinking rather than take our word for it.

Start your first 7 days for $5

Teaching kids about AI

Part of the series: Teaching kids about AI

Related guides in this series

• How to explain AI to a child — Simple analogies and sample scripts by age (5–12), plus the three things every kid should understand about AI.
• AI activities for kids — 10 free, screen-smart activities to explore how AI works together — no coding, no sign-up, mostly off the screen.
• Fall break activities for kids — A half-day-by-half-day plan for five unstructured days — ten low-prep blocks, screen and off-screen alternating, ending with your child explaining what they made.
• AI tools for kids — A parent’s shortlist of genuinely kid-appropriate AI tools — what to look for, what to avoid, and how to tell a real kid tool from a repackaged chatbot.
• AI tools that teach kids to think, not just get answers — The one testable property that separates a thinking tool from a vending machine, what happened when someone measured the difference, and the ten-minute test you can run tonight on whatever your child already uses.
• Winter break projects a kid actually finishes — Six get started and none get finished, and that is a scoping problem rather than an attention one. Five projects with a finish line you can see, the three sessions that get one done, and why none of it is catch-up work.
• How to teach a kid to write a good AI prompt — A list of fifty prompts teaches nothing, because the skill is producing the question. The three decisions that make a prompt good, a try-it bench on the page, and the one rewrite with a measurement behind it.
• Is it cheating if my child makes art with AI? — Two questions in one coat: a rules question with a clear answer, and an effects question where the best study points both ways — individually better, collectively more similar. What to say, and the three situations that differ.
• The five-minute quiz that beats re-reading — Re-reading raised students’ confidence; being tested raised what they still knew a week later. The four instructions that make an AI run the tested version, and when to come back to it.