Should children use ChatGPT without adult supervision, and what safeguards should parents put in place?

Start with the child’s age and readiness

OpenAI’s current product rule is a floor, not a measure of readiness. As verified on 2026-09-01, ChatGPT is not meant for under-13s. Users aged 13 through 18 need parental consent. OpenAI also cautions educators to have an adult conduct the actual ChatGPT interaction when it is used educationally with children under 13. Local law, a school’s policy, and a child’s developmental needs may require a more protective approach. OpenAI age guidance

The most useful family question is not “Can my child open the app?” It is “Can my child notice when an answer is wrong, refuse a request for personal information, stop a disturbing conversation, and bring a concern to an adult?” If not, direct adult co-use is the appropriate setting.

This table is a recommendation, not a diagnostic tool. Move a child toward closer supervision if they are impulsive online, easily distressed by content, struggling with peer pressure, coping with trauma or a mental-health condition, or using a chatbot as a primary source of companionship. The American Academy of Pediatrics notes that human-like chatbots can gain a young person’s trust even though they do not think, feel, or provide the care of a real relationship. AAP guidance

What the product safeguards can and cannot do

Current ChatGPT protections as of 2026-09-01

OpenAI is rolling out ChatGPT for Teens to eligible personal accounts identified as under 18. It adds age-appropriate safeguards and learning-oriented features, but its availability can vary during rollout and by region. Do not assume a teen account has the experience until you check the account itself. ChatGPT for Teens

Parents and guardians can link their account to a teen account through Settings > Parental controls. The linked adult can manage selected settings, including Reduce sensitive content, use of conversations to improve models, saved-memory reference, Voice Mode, image generation, study hours, and quiet hours. Some controls, such as browser or coding-network controls, are separate. Some changes take effect only for new conversations. OpenAI parental-controls guidance

Keep Reduce sensitive content on for a linked teen account. It adds safeguards intended to lower exposure to graphic material, viral challenges, risky roleplay, and extreme beauty ideals. It is not a guarantee that every response will be age-appropriate. A parent should report an unsafe response in the product, then discuss it with the child rather than treating the setting as a filter that cannot fail. OpenAI sensitive-content guidance

Parental controls are not a surveillance dashboard. They do not provide conversation transcripts, chat history, or real-time monitoring. In limited serious-safety situations OpenAI may send a safety notification, but it may not identify every concern and explicitly does not replace professional care or emergency services. Either the linked teen or adult can unlink the accounts, with a notification to the parent if the teen does so. This is why an ongoing, non-punitive conversation matters more than a hidden technical control. OpenAI parental-controls guidance

Privacy and data settings as of 2026-09-01

For a consumer teen account, turn off Improve the model for everyone in Settings > Data Controls unless the family has deliberately chosen otherwise. OpenAI says the setting applies across the account’s devices. It prevents conversations from being used to train ChatGPT, but those conversations can still appear in history. A linked parent or guardian can manage the setting for a teen account. OpenAI Data Controls FAQ and OpenAI data-use guidance

Review Settings > Personalization > Memory. When enabled, memory can use context from chats, files, and connected apps to personalize responses. The memory summary may not show every relevant source, so turning memory off is the simplest choice for a young person who does not need personalization. If a family leaves it on for an older teen, review it together and delete inappropriate entries. OpenAI Memory FAQ

Temporary Chat can be a useful privacy option for an ordinary, non-sensitive question that should not appear in history or create memory. It is not a private counseling channel. A non-personalized Temporary Chat does not use existing memory, custom instructions, or plugins, and it does not create memory. OpenAI may keep a copy for up to 30 days for safety purposes. Saving it converts it into a regular chat, and third-party actions can be subject to another provider’s privacy policy. OpenAI Temporary Chat FAQ

The risks adults should name plainly

Hallucinations and confident errors

A hallucination is a fluent answer that is false, invented, or unsupported. It can look persuasive, including when it gives sources, calculations, or a confident explanation. OpenAI’s own Study Mode documentation says the feature can make mistakes and calls for double-checking high-stakes information. The UK government’s school data-protection guidance likewise tells users to fact-check generative-AI results. OpenAI Study Mode guidance and UK school guidance

Teach a child to say, “How can we check this?” rather than “Is this true?” The practical check is to compare the answer with a textbook, a teacher-provided source, a library database, a government or university page, or another source appropriate to the question. For a simple factual claim, find two independent reliable sources. For a calculation, work it on paper or with a calculator. For a quote, find the original source.

Inappropriate content and manipulation

No family setting can turn a general chatbot into a guaranteed child-safe environment. Sensitive-content reduction is a useful layer, but OpenAI states that no system is perfect. A child can also receive a confusing, prejudiced, frightening, or overly persuasive response without it being obviously explicit. OpenAI sensitive-content guidance

Manipulation does not require a bot to intend harm. A warm voice, apparent agreement, rapid replies, and an invitation to keep talking can make it feel authoritative or friend-like. UNICEF identifies data privacy, inappropriate content, and manipulative design as child risks, and reports concern that young people may prefer AI interaction to turning to people. Treat that as a risk to manage, not proof that every child will become dependent. UNICEF parenting guidance and UNICEF chatbot policy brief

Use direct language with children: “It talks like a person, but it is software. It does not know you, keep secrets like a trusted adult, care whether you are safe, or decide what is good for you.” A child who says “the bot understands me better than anyone” needs connection with a parent, teacher, counselor, coach, or another real person, not more private bot time.

Personal data disclosure

Children often do not recognize that a detail is identifying when it is combined with others. The family rule should prohibit sharing a full name, school, teacher’s name, home address, live location, telephone number, email, passwords, photos with identifying details, family schedules, another person’s private information, or health and therapy details. Do not upload a school record, individualized education plan, medical note, or screenshot containing those details without a parent and the school’s approval.

UNICEF specifically warns that a child may disclose a name, school, routine, friendships, feelings, or health information without realizing how sensitive it is. For schoolwork, UK guidance says schools should know how a tool processes personal data and should fact-check the result. Ask the school which tools and uses are approved. UNICEF parenting guidance and UK school guidance

A good substitute is a generic prompt. Instead of “I am nine, go to Oak Street School, and have a doctor’s appointment tomorrow. What should I say about my anxiety?” use “What are some questions a child could ask a trusted adult about feeling worried?” The second prompt is still not a replacement for clinical care, but it avoids needlessly identifying the child.

Overreliance and homework shortcuts

The learning risk is not only plagiarism. A child can learn to wait for a completed answer instead of trying, explaining their reasoning, tolerating uncertainty, or asking a teacher. UNICEF notes concern that habitual dependence on AI for answers can be difficult to reverse. Schools may also have different rules for collaboration, citation, and graded assessment. UNICEF parenting guidance

ChatGPT’s Study Mode can support a better pattern. It can ask questions, give hints, explain a concept in stages, and check understanding, but it may still give a direct answer and can make mistakes. It does not replace a teacher, tutor, course materials, or a school’s academic requirements. OpenAI Study Mode guidance

Set up the first safe session

1. Decide whether the child meets the age rule. Under 13, do not give independent access. If an adult uses ChatGPT with the child for education, the adult runs the interaction. For ages 13 to 17, get parental consent and create the expectations before use. OpenAI age guidance

2. Use a parent and teen account link where available. In the parent account, go to Settings > Parental controls, invite the teen, and review the settings together after the teen accepts. Keep Reduce sensitive content on. Add quiet hours if late-night or unplanned use is a problem. Use study hours to start eligible new conversations in Study Mode during homework time, knowing that study hours do not block access. OpenAI parental-controls guidance

3. Make privacy the default. Turn off Improve the model for everyone. Consider turning Memory off for younger teens. Do not connect email, cloud storage, or other apps until the family understands why that access is needed. Do not allow browser, agent, or code-execution features without a specific adult-reviewed purpose.

4. Choose a shared place for younger users. For a 13-year-old starting out, a kitchen table or family-room session makes it easy to ask questions without turning every chat into an inspection. For older teens, agree on a periodic check-in instead of demanding transcripts.

5. Agree on a stop-and-show rule. Stop the chat and show an adult if it asks for personal details, makes the child uncomfortable, encourages secrecy, contradicts a trusted source, gives health or safety advice, or seems to know something it should not know. Report an unsafe response through the product if appropriate. OpenAI sensitive-content guidance

6. Make school expectations explicit. A child may use a bot to brainstorm questions, receive hints, practice, or check a draft only if the teacher permits that use. The child should be able to explain the final work in their own words and disclose AI help when school rules require it.

Use Socratic tutoring instead of answer delivery

Ask ChatGPT for a process that keeps the child doing the intellectual work. This is useful even without Study Mode.

• “I am learning fractions. Do not give the answer first. Ask me what I notice, then give one hint and wait for my answer.”

• “Give me a simpler example with different numbers. Then give me one similar question to solve on my own.”

• “Ask me to explain why I chose that step. If I am wrong, tell me which idea to revisit without solving it for me.”

• “Quiz me one question at a time. After each answer, tell me what I understood and what I should check in my notes.”

• “Help me make three questions to ask my teacher about this topic. Do not write my assignment for me.”

OpenAI describes this as a Socratic-style approach. Study Mode can ask what the learner already knows, where they got stuck, and what step they would try next. A parent or teacher should still inspect the first few sessions, because the mode sometimes provides a direct answer and can be wrong. OpenAI Study Mode guidance

Example with a 9-year-old

Hypothetical example. A 9-year-old is stuck on 3/4 + 1/8. Since ChatGPT is not meant for under-13s, the parent sits with the child and conducts the interaction. They enter no name, school, account details, or photo of a worksheet. The parent asks: “Help a child learn to add 3/4 and 1/8. Ask one question at a time, use a drawing idea, and do not reveal the final answer until the child has tried.”

The bot asks the child to name a common denominator. The parent pauses the screen, gives the child paper and fraction strips, and asks the child to explain why fourths need to become eighths. If the bot gives a wrong step or an answer too quickly, the parent says, “Let’s check the workbook example.” The child then solves one new problem without the bot. The takeaway is that the AI supplied a prompt for conversation and practice, while the adult protected privacy, checked correctness, and kept the child as the learner.

A short family AI agreement

Use this as a starting point and adapt the language to the child’s age.

1. We use AI to learn, practice, brainstorm, or create. We do not use it to avoid thinking or to submit work we cannot explain.

2. We do not share names, schools, locations, addresses, passwords, photos with identifying details, health information, or another person’s private information.

3. We check important claims with a textbook, teacher, parent, library, or reliable original source. We do not trust an answer just because it sounds confident.

4. We tell a parent or trusted adult if a bot says something scary, sexual, violent, secretive, upsetting, or confusing. The child will not be punished for showing it.

5. We use a real person for feelings, friendship problems, bullying, body concerns, health, medicine, safety, or serious decisions.

6. We follow our school’s AI policy and name AI assistance when a teacher requires it.

When the bot must stop and a person takes over

ChatGPT should not be treated as a trusted adult, therapist, crisis counselor, doctor, nurse, teacher of record, lawyer, or emergency service. A warm response can be useful as a starting point, but it cannot assess the child’s real environment, notice danger off screen, obtain consent, protect confidentiality in the way a clinician must, or take responsibility for a decision.

Bring in a parent, guardian, teacher, or school counselor for bullying, sexual content, threats, relationship pressure, academic-integrity questions, disturbing media, or a response that seems manipulative or false. Bring in a qualified clinician or local health service for symptoms, medication, eating concerns, mental health, possible abuse, or a child’s ongoing distress. For immediate danger, possible self-harm, threats of violence, overdose, serious injury, or a child who cannot stay safe, contact local emergency services now and stay with the child when safe to do so.

OpenAI’s parental-controls documentation makes the same boundary clear: safety notifications are limited, are not real-time monitoring, may not identify every concern, and do not replace professional care or emergency services. Crisis lines provide real-time trained support and safety planning, but immediate medical emergencies still require emergency services. OpenAI parental-controls guidance and OpenAI crisis-support guidance

Limits and viable alternatives

There is no single setting that makes unrestricted chatbot use safe for every child. A child may use a different account, a different app, a school device, or a friend’s device. Controls also cannot guarantee truth, prevent every unsuitable response, reveal every conversation to a parent, or substitute for a trusted relationship. Use them as one layer in a family plan.

For a younger child, the strongest alternative is adult-mediated use: the adult asks the question, does not provide identifying data, and uses the reply as a discussion prompt. For homework, a teacher-approved tutoring service, school library resource, textbook, or real tutor may be a better fit, especially when a child needs feedback that is accurate, accountable, and tied to the curriculum. UNESCO recommends a human-agency and age-appropriate approach to generative AI in education, including data-privacy protection and an age limit for independent conversations with these platforms. UNESCO guidance

If the issue is loneliness, anxiety, or a wish to talk in private, increase access to people rather than access to bots. UNICEF warns that children and young people may seek advice, support, and relationships from chatbots, which can create distinct risks. A family conversation, teacher, school counselor, pediatrician, mental-health professional, or local crisis service is the appropriate alternative depending on the concern. UNICEF chatbot policy brief