Is ChatGPT Safe for Kids? What Parents Should Know About AI Chatbots
PennPaper Team
If your child has a phone, a school laptop or a friend with either, they have probably already talked to an AI chatbot. A Common Sense Media survey published in 2025 found that 72% of US teens had used an AI companion at least once, and over half used one regularly. Pew Research found that the share of teens using ChatGPT for schoolwork doubled in a single year.
So the useful question for parents is no longer "should my child ever touch AI?" It's "which AI, for what, and with what guardrails?"
This article walks through what general-purpose chatbots were designed for, where the real risks are, and what to look for if you want your child to get AI help with schoolwork safely.
The short answer
General-purpose chatbots like ChatGPT, Gemini, Claude and Meta AI are built for adults. They can be used safely by older children with supervision and clear rules, but they were not designed around a child's needs, and their own terms say so.
- OpenAI's terms require users to be at least 13, and users under 18 need a parent's or guardian's permission.
- UNESCO's 2023 guidance on generative AI in education recommended a minimum age of 13 for classroom use.
- In September 2025 the US Federal Trade Commission opened an inquiry into how AI chatbot companies, including OpenAI, Meta, Alphabet, Snap, xAI and Character.AI, measure and limit harms to children and teens.
None of that means a chatbot is dangerous in every conversation. It means the product was not built with your child as the intended user, and the safeguards are being added after the fact.
What parents actually worry about
Parents' concerns tend to fall into five groups. Each is worth taking seriously on its own terms.
1. The chatbot will talk about anything
A general-purpose assistant is designed to be helpful across every topic: recipes, relationships, medicine, politics, fiction. Safety filters block the worst content, but the default posture is "answer the question." A child who starts with fractions can drift anywhere in three messages, and the chatbot will usually follow.
This is the concern behind most of the headlines. Common Sense Media's 2025 risk assessment rated AI companion apps as an unacceptable risk for anyone under 18, largely because they are designed to keep a conversation going rather than to steer it somewhere healthy.
2. It sounds confident even when it's wrong
Language models generate text that sounds right. For math, that is a specific problem: a chatbot can produce a clean, well-formatted solution with an arithmetic slip in step three, and a child has no way to notice. We cover why this happens in why AI chatbots get math wrong.
3. It does the homework instead of teaching it
Ask a chatbot to "solve this" and it will. The answer arrives in seconds, fully worked. A child who copies it has finished the assignment and learned nothing. Research published in PNAS in 2025 found that students who practiced math with unrestricted GPT-4 access did better during practice but 17% worse on a later test taken without it than students who never had access. More on that in does AI make kids worse at math?
4. It collects more than you'd expect
Chat history, uploaded photos, voice recordings and account details can all be stored and, depending on the settings, used to improve the model. Children rarely read privacy settings. See what AI learning apps collect about your child.
5. It becomes a friend
The most emotionally charged concern is attachment. Chatbots are warm, always available and never bored. For a lonely or anxious child, that can become a substitute for people rather than a supplement. This risk is highest with "companion" apps designed around personality and relationship, but general assistants can drift there too.
If your child uses ChatGPT: a practical setup
If your teen already uses a general chatbot for school, banning it outright often just moves the use out of sight. A better approach is structure.
- Use the parental controls that exist. OpenAI introduced parental controls in September 2025 that let a parent link accounts with their teen, reduce sensitive content, set quiet hours and turn off memory and voice. Other providers are adding similar features.
- Keep it in shared spaces. Homework with AI happens at the kitchen table, not behind a closed bedroom door.
- Agree on what it's for. "Explain this concept" and "check my reasoning" are fine. "Do this problem" is not. Write the rule down.
- Ask to see the conversation. Not as a punishment. The same way you'd look over a worksheet.
- Test the understanding, not the answer. Pick one problem and ask your child to explain it to you without the screen. If they can't, the AI did the thinking.
We go deeper on the conversation itself in how to talk to your kids about AI and homework.
What a safer AI for children looks like
The alternative to a general-purpose chatbot is not "no AI." It's AI that was designed around a child from the first line of code. When evaluating any AI tool for your child, look for:
| Question | General chatbot | Purpose-built tutor |
|---|---|---|
| Is it limited to one subject? | No, it answers anything | Should be yes |
| Does it refuse to just give answers? | Only if prompted to | Should by default |
| Can children contact other people? | Varies | Should be no |
| Can parents see every session? | Only with linked accounts | Should be yes |
| Does it store voice or photos? | Often, by default | Should be minimal |
| Is it designed for under-13s? | No | Should be |
This is the gap PennPaper was built to fill. The tutor is limited to math and won't discuss other topics. By default it asks your child to try each step before helping, and doesn't hand over final answers. Children talk only to the tutor, never to other users. Parents can review session transcripts and progress from the parent dashboard. We collect a first name, learning progress and session activity, and voice is processed in real time rather than kept as a permanent recording. We don't use children's data for advertising.
We wrote a longer checklist in how to choose a safe AI tutor for kids.
Frequently asked questions
What age can kids use ChatGPT?
OpenAI's terms require users to be at least 13, with parental permission for anyone under 18.
Is ChatGPT safe for a 10-year-old?
It isn't designed for, or permitted for, children under 13. A tool built for that age group, with a narrow subject focus and parental visibility, is a better fit.
Can my child use AI for homework without cheating?
Yes, if the AI teaches rather than answers. See is using AI for homework cheating?
Should I ban AI entirely?
Most families find that clear rules and the right tool work better than a ban, which tends to push use somewhere you can't see it.
The bottom line
ChatGPT isn't evil, and it isn't a babysitter. It's a powerful adult tool that children are using anyway. The safest path is to know what your child uses, set clear rules for schoolwork, and where you can, give them a tool built for learning rather than for answering everything.
Sources: Common Sense Media (2025), Talk, Trust, and Trade-Offs; Pew Research Center (2025), teen ChatGPT use for schoolwork; OpenAI Terms of Use; UNESCO (2023), Guidance for Generative AI in Education and Research; US FTC (2025), 6(b) orders on AI companion chatbots; Bastani et al. (2025), PNAS