AI skills and tools for children: what to watch
This is a live Practicum page for Volume 2, Chapter 7. Verified: 2026-08-18. This is not an app ranking or a recommendation to give a child every tool. Check the date, service rules, age limits, data policy, and school requirements.
Use the chapter's main filter: AI should leave some effort to the child. A tool helps when it supports questions, checks a draft, provides practice, and explains mistakes. It becomes a harmful habit when it quietly does all the work.
What to review once a quarter
| Area | What to check | Family question |
|---|---|---|
| Learning with AI | school rules, hint features, review modes, age settings | Does the child understand more, or only submit finished work faster? |
| Companions | age limits, privacy, conversation history, emotional dependence | Is there a real conversation that AI does not replace? |
| Creative work | image, music, video, and presentation generation | What did the child choose, change, and make personally? |
| Projects | tools for research, maps, albums, websites, and prototypes | Did the work leave something tangible, such as a page, album, object, or presentation? |
| Safety | personal data, photos, location, school accounts, payments | What must the child never tell the machine? |
A good sign
After using AI, the child can explain:
- the goal;
- what they did;
- what the machine suggested;
- where the machine was wrong or oversimplified;
- what they could repeat without help.
A warning sign
The answer looks polished, but the child cannot explain the reasoning, defend a choice, or complete a similar task without chat. The answer is not necessarily to ban the tool. Change the mode: use hints, questions, review, and one final attempt by hand.
Related Practicum pages
- Guide: what to teach when answers are cheap
- Family 3C audit
- AI tutor in training mode
- Prompt bank: useful friction instead of a finished answer
Update this page after a material change to school rules, children's modes in AI services, or safe use practices for AI companions.
What the research says, and the part parents usually miss
The evidence in 2026 is more useful than either the panic or the sales pitch, because it is mostly about conditions, not about the tool.
A meta-analysis of 51 studies found that using a chat assistant inside a structured educational setting produced large gains in learning performance and moderate gains in higher-order thinking. The same literature reports the opposite where dependence is heavy and unguided: weaker creativity and weaker independent problem-solving. The variable that flips the sign is not the model. It is whether there is a task, a structure, and an adult somewhere in the loop.
The screen-time research points the same way, and one 2026 result is worth knowing precisely because it is inconvenient: fifteen-year-olds using social apps for more than an hour a day were less likely to score above average for difficulties than those using them less, which the authors read as friendship and support rather than harm. Simple hour-counting does not predict much.
What does predict something, consistently: what the child is doing, on what, at what age, and whether anyone is present. Passive, fast-paced, solo consumption of algorithmic short-form video in early childhood carries the highest documented risk. Balance, structure, and an adult nearby remain the strongest protective factors found.
How to use this at home. Do not negotiate over minutes. Negotiate over the other four things. A child building something with a model beside a parent and a child alone with an autoplay feed are not doing the same activity, and no time limit tells them apart.
Sources (verified 2026-08-18)
- Common Sense Media on AI and children, including reviews and risk ratings: https://www.commonsensemedia.org/ai
- UNESCO on AI in education: https://www.unesco.org/en/digital-education/artificial-intelligence
- Meta-analysis of 51 studies on chat assistants in structured learning, and mixed-methods work on dependence and creativity. Peer-reviewed literature; check the specific study before quoting a number.
- 2026 longitudinal screen-time research reporting that simple hour counts predict little, published via ScienceDaily and the underlying journal article.
Check regulatory facts, including lawsuits involving chat companions and regulator decisions, in the primary source. Keep the tone calm rather than alarming.