How should parents and educators explain AI to children aged five to ten?

What to say, by age

Use the child’s actual experience—an AI search summary, a recommendation, a camera effect, an image generator, or a voice assistant—rather than starting with an abstract lecture. Do not say that every computer feature is AI; ordinary animation, a calculator, and a fixed recording may be software without a generative AI system.

An accurate explanation in adult language

“AI” is a broad label for computer systems that perform tasks such as recognizing patterns, recommending content, translating, or generating text and images. A generative AI system produces a new response based on learned statistical patterns. It is useful to call that output a prediction or generated response, not “knowledge.”

That distinction matters: generative systems can produce confident but false content. NIST calls this risk confabulation and cautions that words such as “hallucination” can wrongly suggest consciousness or intent (NIST AI 600-1). An AI system may also retain or process conversation data according to its settings and policy; that is not human memory, and it is another reason not to provide personal details.

Three short scripts for real moments

1. First conversation

For 5–7:

“AI is a kind of computer tool. It learns patterns from lots of examples and then makes a guess. It can be helpful, but it can also get things wrong. People decide what is true and what is kind.”

For 8–10:

“Some AI can write or draw. It does not check every answer against the real world. Before we believe or share something, we find where the information came from and check it together.”

2. When a child sees a surprising image or video

“It might be a real recording, edited, or made with AI. We cannot tell just because it looks real. Let’s pause: who posted it, where did they get it, and can we find a reliable report or original source? If we cannot check it, we do not share it.”

This teaches source-tracing rather than unreliable “spot the AI by its look” tricks. AI-generated material and ordinary edited material can both look convincing.

3. When a tool asks for a name, picture, or secret

“Stop and ask first. Computer tools do not need your full name, school, address, phone number, passwords, or a photo of you or someone else for our activity. We only use information a grown-up has checked is okay to share.”

Avoid making AI sound human

Children naturally use human language for a talking tool. Gently rephrase rather than scolding. The goal is to preserve the child’s curiosity while keeping the model’s limits clear.

An adult can still use ordinary shorthand—“ask the tool”—but should not imply that it cares, keeps secrets, has authority, or can replace a trusted person.

Four simple activities (no child account required)

1. The pattern-and-guess game — 5 minutes

Make a small set of cards: red circles, blue circles, red squares, blue squares. Cover the next card in a pattern and ask the child to predict it. Then deliberately change the pattern and ask what happened.

Say: “AI is good at pattern guesses. A good guess is still not the same as knowing.”

2. Fact, guess, or make-believe? — 10 minutes

Read three prepared statements about a familiar animal or local place: one easy fact, one reasonable but false claim, and one story idea. Let the child sort them into fact to check, guess, and make-believe. Check the factual claims in a book, museum site, library source, or with a teacher.

Learning move: Ask “What would count as evidence?” not “Can you spot a fake?”

3. Two-source detective — 10–15 minutes (best for 8–10)

Choose an adult-prepared, low-stakes question such as “What do monarch butterflies eat?” Find a reputable science, museum, or government source, then a second independent source. Have the child point to the part that supports the answer.

Rule: A fluent answer is not enough. For a claim that matters, look for a named source and check it elsewhere.

4. Make the prompt safe — 5 minutes

Put two prompts on cards:

• “Write a story about [child’s full name] at [school name].”
• “Write a story about an imaginary child who loves space.”

Ask which is safe to use and rewrite the first one. Extend the activity to photos, voice recordings, a sibling’s information, and passwords.

The child’s safety card: “Stop, protect, check, tell”

Keep the rule short enough to recall in the moment.

1. Stop when an answer, image, message, or request feels strange, scary, too grown-up, or too personal.
2. Protect private information: real name, birthday, address, school, phone number, passwords, usernames, live location, family routines, photos/voice, and information about other people.
3. Check important claims with a trusted adult and a reliable source. Do not repost a surprising item just because a tool generated it or a search result summarized it.
4. Tell a trusted adult. It is never “being in trouble” to tell.

In the United States, COPPA provides specific protections around online collection of personal information from children under 13, including parental consent requirements in covered situations (FTC overview; FTC FAQs). That is a legal safeguard, not proof that any individual app is appropriate or that its terms permit a child’s use. Outside the United States, laws differ. Families and schools should check the tool’s current age terms, privacy policy, and local rules before enabling access.

Ask an adult immediately when…

The final row is especially important because children may encounter AI through games, search, messaging, or other apps even if a family never chooses a standalone chatbot.

Home: choose the smallest useful level of access

Use a shared family rule: “We use AI together for a job; we do not use it alone for secrets, feelings, or important answers.” This is a boundary, not a reason for shame. The American Academy of Pediatrics recommends co-viewing/co-playing as part of a shared media plan; it gives adults a chance to guide rather than merely police online experiences (AAP/HealthyChildren).

A practical 15-minute family routine

1. Notice (3 minutes): Find one AI-adjacent feature the family saw today: a recommendation, search summary, filter, translation, or chatbot.
2. Name (3 minutes): Ask, “What was it trying to predict or generate? What can it not know for sure?”
3. Protect (3 minutes): Name one detail that stays private.
4. Check (4 minutes): Pick one low-stakes claim and verify it together with a reliable source.
5. Reflect (2 minutes): Ask, “When would we ask a person instead?”

Repeat occasionally in ordinary moments rather than relying on one dramatic talk. Adults should model the behavior: say aloud when they check a generated answer or decline to enter a private detail.

Classroom: teach concepts before granting tool access

For ages 5–10, a class can meet the learning goal without every pupil touching a chatbot. Begin with an unplugged pattern activity, source evaluation, an adult-curated example, or a whole-class demonstration. This protects privacy, includes children whose families opt out, and keeps the teacher—not the system—responsible for instruction.

Before any student-facing AI use, a school leader should be able to answer:

• What specific learning problem does this solve that an existing activity does not?
• Which children will use it, for how long, and under whose supervision?
• What student data, prompts, uploads, and logs are collected, retained, shared, or used for training?
• Does the tool’s age policy and the district’s procurement, privacy, accessibility, safeguarding, and parent-notice rules permit this use?
• What is the non-AI alternative for families who opt out or for a failed/unsafe session?

Those questions align with current U.S. Department of Education guidance: use should have a clear instructional purpose, evidence, educator judgment, transparency, and protection of student data (U.S. Department of Education, 2026). UNICEF’s child-centered AI guidance likewise emphasizes safety, privacy, fairness, transparency, well-being, inclusion, and preparing children with skills (UNICEF guidance 3.0).

A safe elementary classroom pattern

1. State the objective: for example, “Students will explain why a generated answer needs evidence.”
2. Use a teacher-owned, pre-reviewed example: no student names, work samples, accounts, photos, or voice data in the prompt.
3. Make pupils think first: draw, predict, write a short answer, or discuss before seeing generated output.
4. Compare and verify: identify a claim in the output and trace it to credible sources. Make “I don’t know yet” an acceptable conclusion.
5. Close the loop: name what the tool did well, what it got wrong or could not establish, and what a person decided.
6. Document and communicate: tell families the purpose, tool, data practice, access route, alternative, and contact person before recurring use.

Do not ask children to upload identifiable writing, artwork, images, behavioral data, individualized education information, or personal stories just to create a classroom demonstration. A school’s approval of an app does not replace a teacher’s duty to follow district policy and local legal advice.

Misinformation: teach verification, not “AI detection”

The practical question for a child is not “Can I identify an AI image by looking at it?” Visual tells are unreliable and change quickly. Teach a repeatable process instead:

This protects against both human-made and AI-generated misinformation. NIST’s description of confident false generated output supports treating a chatbot answer as an unverified claim, even when it includes plausible-sounding explanations or citations (NIST AI 600-1).

Common failure modes—and better moves

Limits, judgment calls, and alternatives

There is no single maturity threshold within a five-year age band. Adjust the language, supervision, and media plan for the child’s reading level, disability, communication needs, anxiety, prior online experience, and family context. A child who can repeat the rules may still need close adult support to use them under pressure.

Evidence about generative AI in elementary education is still developing, and tools, school policies, age terms, and data practices change frequently. Do not infer educational benefit merely because a tool is popular or branded for learning. The cited U.S. guidance explicitly calls for a clear learning purpose, evidence, transparency, and review of effectiveness (U.S. Department of Education, 2026).

Viable alternatives:

• Teach the same literacy with a search result, recommendation list, ad, edited photo, or a deliberately misleading paper card—no generative tool needed.
• Let children write and illustrate their own stories, then discuss how a computer might make a different version; do not outsource story time or creative effort by default.
• For accessibility needs, use established assistive technology and human support first; assess any AI feature with the same privacy and educational scrutiny.
• If a school cannot document privacy, age suitability, instructional value, and an opt-out alternative, use an unplugged lesson instead.

This is educational guidance, not legal, medical, financial, mental-health, or safeguarding advice. For a child’s urgent safety concern, move from the device to a trusted adult and the appropriate local professional or emergency service.