A child may interpret a fluent answer as more authoritative than it is. This session gives a caregiver a practical way to introduce three rules: check claims, protect personal information, and involve an adult when the topic is sensitive or consequential. It does not establish that a child is ready for independent use.
The caregiver should operate the account and device while the child observes or suggests prompts. Follow the provider’s minimum-age, parental-consent, and account-sharing rules; do not create an account or misstate a child’s age to bypass them. Keep school records and the child’s identity out of the session. See AI literacy by age for a broader readiness discussion.
Before you open anything
Three setup rules, non-negotiable:
- Caregiver-controlled account. Use your own login, on a device you control the settings for. Do not create a separate account for the child for this first session.
- Nothing identifying goes in. No school name, no home address, no photo of the child’s face, no full name paired with any other detail. Practice with a made-up scenario or a genuinely generic question instead.
- Short and structured, not open-ended. End while the child can still explain the rules back. The right duration depends on age, attention, access needs, and the child; the suggested times below are facilitation estimates, not developmental standards.
Treat this like teaching a child to swim in a pool you can see the bottom of, not the open sea. If your child ever wants to explore further, that happens later, on a separate, deliberately-decided account — not by extending this session.
The script
Step 1: name what the tool actually is (2 minutes)
Use words and a communication mode the child can access: “This program makes answers from patterns in information. It can sound sure and still be wrong.” Ask them to explain or demonstrate the idea in their own way. If repetition is difficult, use an example, picture, or choice question instead; verbal recall is not the only evidence of understanding and is not a readiness test by itself.
Step 2: ask something real together (3 minutes)
Pick a genuine, low-stakes question the child actually wants to know — “how do octopuses see,” “why does bread rise,” whatever is real for them. Type it together (or let them type it, with you watching). Read the answer out loud. Ask: “does this match anything you already know?”
This ordinary first question lets the child see a useful answer before practicing verification. It is a facilitation choice, not a proven developmental sequence; adapt it if the child needs a different order.
Step 3: choose a source before checking (3 minutes)
Choose a factual question and an authoritative, child-readable source before asking the model. The exercise must work whether the model is right or wrong:
- Ask for one precise fact stated in the book or official webpage you selected.
- Ask the model to identify what source would verify its answer.
- Predict what part of the source would answer the question, but do not compare yet.
Do not try to trick the model into inventing a falsehood or pretend a correct answer is wrong. The lesson is that verification is a normal step, not that every answer fails.
If the model gives no inspectable source, say that this limits how much weight you can place on the answer. For an older child, why AI sometimes produces confident wrong answers explains the mechanism.
Step 4: compare the claim with the source (5 minutes)
Open the source together and compare the exact claim, date, population, or table with the model’s answer. Discuss why the source is suitable: who published it, how current it is, and whether it directly addresses the question. If the answer matches, say: “This claim matched this source.” One matching page does not make every model answer reliable. If it differs, identify the mismatch without dramatizing it. The lesson is not “AI is bad”; it is “checkable claims get checked against an appropriate source.”
Step 5: the privacy pause (3 minutes)
Before ending, ask the child: “if you wanted to ask it something about you—your school, your friends, where we live—or if it showed something sexual, frightening, threatening, or secret, what should you do?” Build toward: stop, do not share more, and tell a trusted adult. Practice one hypothetical school-project prompt and let the child identify details to leave out.
Do not use this session to enter any real identifying detail about the child “just to see what happens.” The demonstration works better with a hypothetical than with your child’s actual data, and hypothetical mistakes are much easier to walk back than real ones already submitted to a provider’s systems.
Step 6: closing reflection (2-3 minutes)
End with three short questions, in the child’s own words:
- “What is this thing good at?”
- “What is it bad at, or where does it need checking?”
- “What would you never type into it without asking me first?”
Write down their answers if they’re old enough to appreciate that, or just listen. The goal of this closing step is retrieval, not new information — if they can answer these three questions unprompted a day later, the lesson held.
Adjusting the script by age
The steps stay the same; who does the typing and how much you explain changes.
- Early readers: the caregiver types and reads the answer aloud; the child helps find the matching sentence in the source.
- Preteens: the child can suggest wording while the caregiver operates the account, then attempts the source check with guidance.
- Teenagers: involve them in choosing the source and explaining the mismatch or limitation. Independent account use remains subject to provider rules, school policy, maturity, and household boundaries.
A common misconception this session corrects
Do not assume experience with search or voice assistants automatically creates generative-AI literacy. Some chat tools show sources and some search tools summarize answers, so the interfaces increasingly overlap. The durable rule is product-independent: identify the claim, open a suitable source, and compare.
What this session is not
It is not a one-time inoculation. A single facilitated conversation does not make a child safe to use AI unsupervised afterward — it establishes a shared vocabulary and one memorable demonstration you can refer back to later. Plan on repeating some version of Steps 3 through 5 periodically as your child’s actual usage grows, not just once.
It is also not a substitute for deciding, separately and deliberately, whether and when your child gets their own account. AI literacy by age covers that decision; this script covers what to do in the room before you make it.
Try it today
Pick one real, low-stakes question your child has asked and run Steps 1 through 6. Do not skip the source comparison or privacy pause. Repeat the rules as products and the child’s use change; one lesson is not evidence of readiness or safety.
The child’s-first-AI-conversation lesson plan turns this script into a printable facilitator card with the exact prompts to use for each step, phrased for three different age bands.
Evidence and review boundary
- UNICEF Guidance on AI and children, version 3 emphasizes safety, privacy, transparency, development, and children’s best interests.
- UNICEF: When AI becomes a friend (2026 policy brief) addresses relational AI risks and child-rights safeguards.
- APA’s health advisory on AI and adolescent well-being recommends developmentally appropriate safeguards and repeated reminders that AI health information may be inaccurate.
- UNESCO guidance for generative AI in education supports age-appropriate, human-centred use and data protection.
- Google’s Gemini sign-in requirements illustrate why age and account rules must be checked for the specific product; other vendors have different terms.
This is a conservative educational activity, not an individualized recommendation from a child psychologist, paediatrician, school, or safeguarding professional. Adjust it to the child’s development, access needs, school rules, and safeguarding context.
