In an elementary classroom the defensible use of AI is the teacher’s own, not the children’s. Children in the first years of school are building oral language, decoding, handwriting, counting, and the ordinary skill of talking with another person. None of that is what an open-ended chatbot is for. What belongs in K-5 is a small set of ideas about how machines behave, and children can learn most of them without touching a device. From third grade the machine’s output can come out where children see it, on one shared screen with the teacher at the keyboard.

Where the age-13 recommendation comes from

The age guidance for the elementary grades is a design recommendation rather than an experimentally established threshold, and the evidence on younger students is particularly limited — treat the grade-band defaults as a starting point and adapt them to the children in front of you.

UNESCO recommends 13 as a minimum age for independent conversations with general-purpose generative-AI systems. It pairs that number with national regulation, institutional safeguards, age-appropriate design, privacy protection, and human oversight. The recommendation is guidance rather than a universal legal rule, and individual products may set different terms. It marks the age below which independent conversation with an open-ended system is inappropriate for a child, and says nothing about a teacher using that same system to build Tuesday’s reading passage. A school reading it as a ban applies it to questions it was never aimed at.

A screen minute in these grades has to earn its place, and in the years where children are learning to read faces, take turns, and hold a pencil it rarely can. Elementary AI practice therefore runs mostly on conversation and paper, by design rather than for want of tools.

Kindergarten through grade 2: teacher-facing and unplugged

A teacher drafts a provisional example or adapts a set of directions for a group that needs them worded differently, then reviews the result before any of it reaches a child. In a shared demonstration the teacher types and the class watches, which is as close as the system gets to the children. None of it requires a child to touch the system, and the default here is teacher-facing use and shared demonstrations rather than independent open-ended chatbot use.

The student-facing priorities are oral language, decoding and fluency, handwriting and drawing, counting and physical manipulation, observation and play, talking with other people, and building basic knowledge.

What a K-2 class can learn about machines: none of it needs a device. Machines recognize patterns and still make mistakes. A computer’s answer is not the same as somebody knowing, and people choose the data and the rules a system gets. Pictures and voices can be made by a machine, and a child asks a trusted adult before sharing anything about themselves. All of them still hold the first time that child meets a chatbot.

Grades 3 through 5: teacher-controlled and shared-screen

Upper elementary is where a machine’s confident wrong answer becomes useful classroom material. The teacher still types the prompt and decides what the class sees, and use stays teacher-controlled and shared-screen. What changes from K-2 is that the output now becomes material the children work on directly.

A class can compare an AI summary with a student’s own summary after the class has read the text, hunt for three errors in a synthetic answer, or sort examples of human and machine decision-making. They can also verify an AI claim against a class text, or discuss why fluent language can still be wrong. Teacher-approved feedback on a paragraph a student has already finished is the only activity here that touches a child’s own writing.

Three uses to keep out of these grades are unrestricted research chat, autonomous essay generation, and continuous individual access to a general-purpose system. The first two hand a child the part of the work that builds reading and writing — finding the source, forming the sentence — at the age when that work is the curriculum. Continuous access removes the supervision the whole band rests on.

The rules loosen in the older grades. Supervised, bounded use with disclosure and an initial-attempt requirement arrives in middle school, and graduated autonomy by task follows in high school. The AI in K-12 hub collects the rest of this material.

The five ideas AI4K12 uses to organize AI literacy

AI4K12, the K-12 AI-curriculum guidelines project, organizes AI literacy around five big ideas: perception, representation and reasoning, learning, natural interaction, and societal impact. Representation and reasoning covers how a system stores what it knows and draws conclusions from it, and natural interaction covers how it handles ordinary language. Its grade-band progression charts split at K-2, grades 3-5, grades 6-8, and grades 9-12. An elementary team can read the first two bands to see what a fifth grader should be able to explain about machines.

AI4K12 is stronger on how AI systems work as a technical and social field than on what to permit in a given assignment. An elementary school that wants both will pair it with the assignment-level guidance in the frameworks glossary. The full AI-literacy picture shows where the age-13 threshold sits inside a K-12 progression.

Deciding who operates the tool

The teacher operates the tool and the children work on what it produces. That holds from kindergarten through fifth grade, and it settles most of what a school has to decide about AI in these years.

The Kindred K-12 newsletter is where the rest of this material goes out.