A child asks for help writing a story. An AI system can supply characters, a plot and polished sentences. A teacher can supply a question, a picture or a word the child needs. Both responses may help produce a story. They leave very different work for the child to practise.
Primary schools can teach children about AI while protecting the reading, writing, number knowledge and reasoning they are still developing. The central decision is which capability the lesson is meant to build and which support helps the child practise it.
Doing something “alone” should mean without AI performing the intended thinking. It should not mean being isolated, denied teaching or deprived of necessary accessibility support. Children learn with adults, peers, books and appropriate tools. Growing independence is a supported educational goal.
Learning about AI is different from using it
A child can learn that people design AI systems, that generated information can be mistaken and that private information needs protection without operating a personal chatbot account.
The OECD and European Commission's Empowering Learners for the Age of AI, published on 18 June 2026, describes AI literacy through knowledge, skills and attitudes for primary and secondary learners. Its publication overview emphasises understanding, critical evaluation and responsible choices. It does not establish that every child needs direct access to a generative service.
Practical resources also show that AI understanding need not start with individual screen use. Barefoot Computing's primary resource hub describes AI Explorers activities using age-appropriate, unplugged experiences and discussion. Their availability is evidence of a teaching option, rather than proof of a particular learning effect.
Schools therefore have more choices than introducing chatbots everywhere or saying nothing about AI. They can make systems an object of inquiry while deciding separately whether direct use serves a particular lesson.
Begin with ideas children can examine
Primary AI literacy should make a few important distinctions understandable.
A machine-generated answer is not a person who knows the child. A plausible sentence is not automatically a verified fact. A recommendation reflects information and design choices rather than an unquestionable judgment. An unfamiliar digital service is not automatically an appropriate place to share a name, photograph or personal story.
Teachers can explore these ideas through discussion, selected examples and ordinary classroom resources. They need not pretend that a simple game reproduces the operation of a complex AI model.
For younger pupils, recognising a doubtful claim and knowing which adult to ask may be more appropriate than independently researching its provenance. Older primary pupils can examine a little more: what information would help check the claim, which source is relevant and what remains uncertain?
These are suggested teaching directions, not a universal age-by-age curriculum. Schools should align them with their own requirements, learners and safeguarding arrangements.
Protect the skill being learned
The same assistance can be appropriate in one lesson and remove the point of another.
If pupils are practising composing sentences, a generated paragraph may supply the decisions they need to make. If they are discussing a scientific observation, help with transcription may allow them to express their reasoning. The important distinction is the purpose of the task.
Similarly, a tool that supplies a calculation can be useful in some mathematical work. It may be unsuitable when the objective is to understand that number relationship and explain a method.
This is not a claim that children must do every task without assistance. It is a reason to state the target precisely. “Finish the worksheet” tells a teacher little about which help to offer. “Explain how you know these two quantities are equal” makes the intended thinking easier to protect.
Ask what the child needs to practise, what is currently difficult and what help would make the next meaningful step possible.
A story lesson: support the telling without supplying the story
Imagine a fictional primary class using a sequence of pictures to tell a short story. The teacher wants pupils to connect events, explain a character's action and choose language that makes the sequence understandable.
One child cannot begin. A complete generated narrative would solve the production problem. It would also supply the structure and much of the wording that the lesson is meant to develop.
The teacher first checks the obstacle. Does the child understand the pictures? Know what happened? Have an idea but lack a useful word? Need a different way to communicate?
Those answers suggest different support. The teacher might model a sentence using another example, help the child describe the first picture orally, offer a small word bank or provide blank spaces for arranging the events. A peer can listen and ask what happened next.
The child then does the intended composing at an appropriate level. Later, a different picture sequence helps the teacher observe whether the child can reuse the strategy.
This is an illustrative teaching scenario, not a tested intervention. Its point is the choice of assistance: identify the obstacle before giving the finished product.
Make the child's decisions visible
A polished story can hide who supplied its ideas. So can an adult who helpfully suggests every event.
Listen for the child's contribution. Can they explain why an event follows another? Choose a detail that makes a character's action clear? Notice that a listener is confused and add information?
The teacher may still need to teach explicitly. An unsuccessful attempt does not mean the child should be left to struggle indefinitely. It tells the teacher something about what to model, practise or explain.
The Education Endowment Foundation's second-edition guidance on metacognition and self-regulated learning, published on 13 November 2025, emphasises teaching strategies for planning, monitoring and evaluating learning within subject lessons. Its supporting resources include modelling and scaffolding towards greater independence. This is general teaching guidance, not an evaluation of primary-school AI use.
The Learning Coach perspective is to use support deliberately: notice what the learner is trying to do, help with the obstacle and observe which decisions become possible for the learner next.
Number knowledge deserves the same care
Consider children learning different ways to make a number. A correct answer matters, but the relationship matters too.
A pupil might use counters, draw a representation, explain a known combination or check another child's example. These supports can make the mathematical thinking visible. If a digital system supplies every combination before the pupil has considered any, the teacher has less evidence of what the child understands.
A useful prompt from an adult could direct attention to a known relationship without completing the reasoning. Some children will need a clear worked example and guided practice first.
The appropriate balance depends on prior knowledge and the objective. There is no rule that every child should receive the same amount of help or that help must always take the form of questions.
The practical concern is whether the learner gets opportunities to retrieve, represent, explain and use the relationship themselves. Faster completion is not a substitute for those opportunities.
Independence must preserve access
“Without AI” can become misleading if adults also remove the support a child needs to participate.
A learner may require an accessible format, communication aid, additional processing time or help recording an answer. Decisions about those supports should reflect the child's needs and the applicable arrangements.
Distinguish the capability being assessed from the means of expressing it. A child explaining a scientific idea may be able to demonstrate understanding through speech, drawing or a supported communication method. Handwriting speed need not become a hidden test of scientific reasoning.
Equally, an accommodation does not make every form of answer generation appropriate for every objective. Teachers and relevant specialists should examine what a support does in the particular task.
The aim is meaningful participation and learning, not an artificial contest over who can manage with the fewest resources.
Do not make children carry the checking burden
Adults sometimes propose a simple safeguard: children can use AI as long as they check its answers.
That expectation needs to match what children know. A pupil may recognise an error in a familiar story but lack the knowledge needed to detect a plausible mistake about an unfamiliar topic.
Primary teaching should provide relevant knowledge and model a manageable checking process. A teacher can select a claim, show where appropriate evidence is found and explain why uncertainty remains. Children can then practise a suitably bounded part.
UNICEF Innocenti's September 2026 Skills for an AI World report summary describes consultations with 97 children aged 9–17 in the Democratic Republic of the Congo, Egypt, Mozambique, Senegal and Rwanda. Participants raised concerns about dependence and valued foundational knowledge and skills.
Those are children's perspectives from specific consultations, not representative estimates, evidence about every primary age or proof of an AI intervention's effect. They do support taking children's questions seriously rather than assuming that confidence with a chatbot means readiness to judge it.
Keep human help easy to seek
A child should know what to do when a digital response is confusing, inappropriate or upsetting: stop and seek help from a trusted adult through the school's established arrangements.
Teaching that route is part of AI literacy. So is understanding that a conversational system does not replace a teacher's responsibility or a human relationship.
In everyday lessons, children should also experience help-seeking as a useful learning action. They can explain what they tried, identify the part that is difficult and ask for suitable support.
This is different from requiring a first attempt before every kind of help. Some learners need an explanation to get started; some situations call for immediate adult intervention. Professional judgment remains essential.
The goal is for children to become more aware of the help they need and more capable of using it, while adults remain responsible for appropriate guidance.
Give families a clear explanation
Families need more than “AI is allowed” or “AI is banned.” Explain the learning purpose and the boundary.
For a story-writing task, the school might want the child's own sequence and sentences, while allowing discussion with an adult or use of an agreed word bank. For a different task, other support may be appropriate. Homework expectations should make that distinction understandable.
Do not assume every family has the same technology, language resources or time. Provide an accessible way to meet the learning objective and a route for asking questions.
Schools must also check their current policies, the suitability of any proposed service and applicable age and data requirements before direct use. A general article cannot provide permission for a particular child, product or school.
For leaders developing those arrangements, our guide to turning AI policy into classroom routines connects broad principles with specific expectations.
Review a primary task before adding a tool
A useful staff discussion starts with an existing lesson rather than a catalogue of applications.
Ask colleagues what children are learning, what difficulty they encounter and which support already helps. Then examine what a proposed AI contribution would supply. Would it make participation easier while preserving the intended practice, or perform that practice for the learner?
Choose evidence that fits the question. A short explanation, a fresh example or an observation of the child's decisions may reveal more than the appearance of the completed work. Interpret it over time and alongside other information.
A school leadership keynote can establish these shared questions. A workshop can apply them to actual primary lessons. Our guide to choosing an AI in education keynote speaker helps organisers connect the event with a practical educational purpose.
The sources reviewed here include a framework overview, a resource hub, general teaching guidance and a child-consultation summary. They do not establish one optimal primary AI curriculum or prove that the illustrative lessons improve outcomes. The classroom proposals require adaptation and review.
Children should learn enough about AI to question it, protect themselves and recognise its limits. They should also receive sustained opportunities to develop the knowledge and skills that make independent judgment possible. The adult's task is to choose help that supports that development.
Planning a primary-school professional development day or leadership event about AI and independent learning? Explore Learning Coach's international keynotes and request a tailored proposal.
References
OECD and European Commission, Empowering Learners for the Age of AI. Published 18 June 2026. Publication overview of the primary and secondary AI literacy framework.
Barefoot Computing, primary teaching resources and AI Explorers. Undated resource hub; describes unplugged activities and discussion.
Education Endowment Foundation, Metacognition and Self-Regulated Learning. Second edition, 13 November 2025; overview and supporting-resource descriptions.
UNICEF Innocenti, Skills for an AI World. September 2026; official summary of child-centred consultations. Sources consulted 29 October 2026.
