My child believes everything the AI says

The uncomfortable finding first

Fisher, Goddu and Keil ran nine experiments on a simple question: what does looking something up do to how much you think you know? The answer is that "searching the Internet for explanatory knowledge creates an illusion whereby people mistake access to information for their own personal understanding of the information". People who had searched rated their own understanding of unrelated questions higher afterwards — and in one condition they even judged their own brains to be more active on an fMRI image. The feeling of knowing went up. The knowing did not.

That is adults, and it is web search rather than a chatbot, so the step across to a nine-year-old asking an AI is an inference and this page will keep calling it one. But the mechanism is the one you are watching in your kitchen. A chatbot is a much purer version of the thing that produced the effect: it does not even return a list of pages to pick between. It returns one fluent paragraph, in the register of someone who knows.

The practical consequence is the part worth sitting with. If having been told inflates the sense of understanding, then a child who has looked something up is not in a neutral state afterwards — they are in a more confident one. And a checking habit, taught as "go and look it up properly", is still a looking-it-up. It can leave your child more certain and no better informed, which is exactly the shape of the thing you are trying to fix.

The verification half: how to check whether an answer is true · The honest version of “is AI safe for kids”

What this page is not going to tell you to do

There is a separate page on this site about how to check an AI answer, and it is built on good evidence: the method that works is reading sideways rather than reading harder, and it is a real skill with a real measurement behind it. Nothing here contradicts it. If you have not taught that yet, go and teach it — it is the more useful hour.

This page is for the situation afterwards. The checks have been explained, your child can even recite them, and they still take the answer at face value. Running the same drill again louder is the obvious move and it is not the one with anything behind it, because the gap is not in the method. It is in the confidence, and confidence is not fixed by technique.

Four things "they believe everything it says" usually turns out to be

These want different responses, and two of them are not credulity at all. Work out which one you have before you change anything — the last column is the whole point of the table.

The calibration habit, in four turns

1. 1. Guess first, and say how sure Before the question is typed: "what do you reckon, and how sure are you out of ten?" Both halves matter and the second one is the one people skip. A guess with no confidence attached cannot be wrong in an interesting way. A guess at eight out of ten that turns out wrong is the most useful thing that will happen all week, and it costs about fifteen seconds.
2. 2. Ask the machine, then go back to the guess Not "was it right?" — "what did you have at eight out of ten, and what did it say?" You are deliberately pointing at the distance between the two, because that distance is the thing your child cannot currently feel. Do this often enough and the number starts coming down on its own, which is the whole objective: not scepticism about AI, calibration about themselves.
3. 3. Make them say the bit the answer skipped Chi's study of students working through physics examples found that the ones who learned "generate many explanations which refine and expand the conditions for the action parts of the example solutions", while the others "do not generate sufficient self-explanations, monitor their learning inaccurately, and subsequently rely heavily on examples". The second sentence is the one to read twice: inaccurate self-monitoring and leaning on the worked example travel together. So ask for the step the answer glossed over — why that, and not the other thing — which is where a fluent paragraph runs out.
4. 4. Once a week, bet on it Pick something you can settle in a minute, both of you commit to a number before asking, and then look. You will lose some. Losing some in front of your child is the active ingredient, because the lesson is not "machines are unreliable" — it is "being sure is not the same as being right", and that one has to be demonstrated on a grown-up to be believed.

Why this lands hardest somewhere around seven and eight

Girouard-Hallam and Danovitch tested children at 4–5 and 7–8 and found the trust split sorting itself out with age: more trust in the voice assistant for facts, more trust in the human for personal things, and in a second study a growing preference to go to the assistant for one and a person for the other. The sorting is normal and it is not finished at seven.

Which puts a child in the middle of our age band at an awkward spot. They are actively granting a machine factual authority — that is the developmental direction — while having had no practice at all in noticing their own certainty. Those two things arriving in the same year is why "it is sometimes wrong, be careful" bounces off. It is information about the machine, offered to a child whose problem is information about themselves.

The usual caveat, because this page cites research rather than decorating with it: eighty children per study, from predominantly White upper-middle-class families, using a novel voice assistant in a lab, not a chatbot at home. It is good evidence that the trust split is developmental and moving. It is not a measurement of your child, and the study also found that children's preferences "were not related to… the presence of a voice assistant in children's homes" — so owning a smart speaker is not what did this.

What changes with age, and what does not · If something specific has already happened

What a calibrated answer looks like from the outside

The reason this is hard to run as a parent is that you are not in the room for most of it. What you can see afterwards is whether the explanation came from your child or from the answer.

The research under this page

Three sources, each doing one job: why being told feels like understanding, why the surface of a source is the wrong thing to judge, and why the age band matters. The first is the page's spine and also the source with the longest step between what it measured and what this page uses it for — so it is quoted with that step stated.

1. Fisher, Goddu & Keil, 2015 — Searching for explanations: how the Internet inflates estimates of internal knowledge. Journal of Experimental Psychology: General, 144(3), 674–687. ("Here we show that searching the Internet for explanatory knowledge creates an illusion whereby people mistake access to information for their own personal understanding of the information. Evidence from 9 experiments shows that searching for information online leads to an increase in self-assessed knowledge as people mistakenly think they have more knowledge 'in the head', even seeing their own brains as more active as depicted by functional MRI (fMRI) images.") The load-bearing source. Nine experiments, and the finding is about the FEELING of understanding rather than about understanding: people who searched rated their own knowledge higher afterwards. Two limits this page states in its own prose rather than hiding here — these were adults, and it was web search rather than a chatbot, so the application to a child asking an AI is an inference. It also measures self-assessed knowledge only; it does not show the searching made anyone worse at a later task, and this page does not claim it did. What it does not show: Adults, and web SEARCH rather than a chatbot — so the transfer to a child asking an AI is an inference, and the page states it as one. It is also about self-ASSESSED knowledge: the inflation is in how much people think they know, and these studies do not measure whether the searching made them worse at any later task. Cited for the mechanism that makes the habit on this page worth building, never as a measurement of a child. Crossref carries no abstract for this DOI; the quoted finding is read from the Europe PMC record (hit 1, full abstract) against a positive control that returned in the same call.
2. 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.") Cited for exactly one row of the table — that judging a source by its surface is not a childish error, since PhD historians in this study fell for "official-looking logos and domain names" while fact checkers left the page and opened new tabs. The METHOD this implies is deliberately not taught here; /guide/how-to-check-if-ai-is-right owns it, and duplicating it is the specific thing this page is written to avoid. 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.
3. Girouard-Hallam & Danovitch, 2022 — Children’s trust in and learning from voice assistants. Developmental Psychology, 58(4), 646–661. (Two studies with children aged 4–5 and 7–8. "With increasing age, children showed greater trust in the voice assistant for factual information and greater trust in the human for personal information"; in a second study, "with increasing age, children showed an increasing preference to seek factual information from the voice assistant and an increasing preference to seek personal information from the human". Endorsement of an informant’s statements also predicted later recall.) Why the age band is the hinge: the trust split between machines-for-facts and humans-for-personal-things is still developing at seven to eight. Eighty children per study in a lab with a novel assistant, so it supports the developmental direction and nothing about an individual child. Cited for its null as well — preferences were unrelated to whether there was a voice assistant in the home. What it does not show: Eighty children per study, "from predominately White, upper middle-class families", using a novel voice assistant in a lab rather than Siri at home. CITED FOR ITS NULL AS WELL: children’s preferences "were not related to… the presence of a voice assistant in children's homes", so this is not evidence that owning a smart speaker changes what a child believes.

Questions parents ask

Why does my child believe everything the AI tells them?

Partly because it is fluent and confident, and partly because of something that works on adults too: looking information up inflates how much you think you understand. Fisher, Goddu and Keil showed across nine experiments that "searching the Internet for explanatory knowledge creates an illusion whereby people mistake access to information for their own personal understanding of the information". A chatbot is a stronger version of that, because it hands over one confident paragraph instead of a list of sources to weigh. There is also a developmental part: children in the seven-to-eight range are actively granting machines factual authority, while having had almost no practice at noticing their own certainty.

I have already taught them to check. Why has it not worked?

Because checking is a method and the problem is usually confidence. Worse, a check is itself a looking-up, so it can leave a child more certain rather than less. The thing that moves it is calibration rather than technique: have them predict the answer and rate how sure they are out of ten BEFORE the question is asked, then compare. The gap between an eight-out-of-ten guess and the real answer is the only way the confidence becomes visible to the person holding it.

Should I tell my child the AI is often wrong?

It is true and it is not very useful on its own, because it is information about the machine offered to a child whose gap is information about themselves. "It is sometimes wrong" also quietly teaches that the answers are fine when it happens to be right, which is the opposite of the habit you want. Better: let them be sure and wrong once, in a low-stakes bet you both commit a number to. Losing one yourself in front of them does more than any amount of warning, because the lesson is that being sure is not the same as being right.

Is it harmful for a child to trust AI?

Trusting it is not the risk; trusting it without being able to tell when to is. The developmental research suggests the factual channel is the one being handed authority around this age, so that is where the work pays off. What this page would steer you away from is the two overcorrections: banning the tool, which removes the practice along with the problem, and running more verification drills, which addresses a method gap your child may not have. Build the habit of noticing certainty, and the checking they already know how to do starts getting used.

Asks first, answers second

Luchi is built around the move this page is about: a kid makes something, and Luchi asks what they were going for and why it works rather than handing over a finished answer. You get a short weekly note written from what they actually said — their explanation, not the machine's.

Start your first 7 days for $5

And the checking habit itself

Part of the series: Is AI safe for kids?

Related guides in this series

• Is ChatGPT safe for kids? — A straight answer by age — the risks that actually come up, and the five habits that make AI safe when you do let kids use it.
• Screen time and AI — Not all screen time is equal. Why active, creative AI use is different from passive scrolling — and how to keep your child on the right side of the line.
• What is my kid doing on AI? — You can’t read every transcript. Four checks that take under a minute each, what each one tells you, and the one question a ten-year-old actually answers.
• AI apps a parent can actually monitor — Alert, transcript, or record — three different things sold under one word. What seven products say they show a parent, quoted from their own sites.
• Is Google Gemini safe for kids? — Google’s own page for parents is more candid than the reviews — imperfect filters, stated hallucination, "not a person to confide in". What it says, what it leaves out, and the four decisions that are yours.
• The school Chromebook already has AI on it — A school account is managed by the district, not by your Family Link. Who actually holds the switch, what Google’s own documentation says a student account gets, and five questions worth one email.
• Deepfakes: what to actually teach a 10-year-old — The “look closely, count the fingers” advice has been measured, and it does not work — neither warnings nor money improved anyone’s accuracy. What to teach instead, and why a detector app is not the backstop.
• What to ask at a parent-teacher conference about AI — The list you found ends with “ask how you can help with homework” — the one form of involvement a 50-study meta-analysis found was not associated with achievement. Six questions that are, and what each answer tells you.
• Is Siri the kind of AI parents worry about? — Apple has quietly split Siri in two — its own page now says Apple Intelligence “powers an entirely new version of Siri: Siri AI”. How to tell which one your child is talking to, what Apple keeps, and the settings worth changing tonight.
• A family AI agreement you will actually keep — Twenty clauses, signed on Sunday, dead by Thursday. A four-clause version built on what a 94-test meta-analysis found actually makes a plan start — naming the when and the where — plus the one question to ask each week.
• An AI said something it should not have — Google’s own parent page says the filters "are not perfect". The order to do things in — keep the conversation, ask the one question, report from inside the chat, decide one thing — and the point at which this hands you to somebody qualified.