Quick Answer: Start AI chatbot safety for kids with the service’s age rules, an approved account, private-data boundaries and a clear purpose for each session. Combine device access limits with the chatbot’s own family settings and regular conversation. These controls do different jobs: restricting an app does not check every answer, and a linked parent account does not automatically reveal chat transcripts.
AI can help a child practise a topic, generate ideas or understand an explanation. The useful question is whether the child can use that help without handing over private information, accepting invented answers or treating a chatbot as their main source of emotional support. A short family agreement makes those decisions concrete.
Start with age, account and purpose
Check the exact product, account type and country before creating an account. OpenAI’s age guidance says ChatGPT is not intended for under-13s and requires parental consent for ages 13–18. For educational use with younger children, the adult must conduct the interaction. A parent’s permission does not remove a service’s minimum age.
Google has a different model: eligible supervised accounts below the applicable age can receive parent-enabled Gemini access. Its family guidance says rollout and regional restrictions apply, including supervised-account unavailability in the EEA, Switzerland and UK. Do not transfer the age or permission rules of one chatbot to another.
Write down the approved service, account and activity. “Use this account to ask for hints on the science worksheet” is easier to supervise than “AI is allowed.” Check the school’s assignment policy as well; permission to brainstorm does not necessarily mean permission to submit generated work.
Use three layers of boundaries
Device controls, linked-account settings and family rules complement one another. Start with the layer that matches the problem, then check the limits of the other two. Buying a monitoring subscription is not a substitute for understanding the account.
ChatGPT’s parental-controls guidance separates Quiet hours, which restrict access, from Study hours, which start eligible new chats in study mode and exclude GPT and Project conversations. Either linked person can unlink; linking gives no chat-history or real-time monitoring access. Safety notices can miss concerns. See our ChatGPT parental-controls guide for setup.
At device level, an app restriction and a website restriction may need separate setup. Check the approved account in its app and browser, including another device the child uses. A disabled feature in one account does not establish that another account, website or app is restricted. Discuss alternate access openly instead of treating a dashboard label as proof of universal coverage.
Match Each Rule to an Observable Check
Use a harmless task to check the family setup together. Do not create disturbing prompts or simulate an emergency to test a filter. Ask the child to show the approved account, explain what the rule changes and identify the person they would contact if something went wrong.
| Family objective | A useful check | What it does not prove |
|---|---|---|
| Access only through an approved account | Confirm which account is signed in and whether a family invitation is accepted | A pending invitation is not a completed link; another account can have different settings |
| Stop use at bedtime | Check the access schedule and ordinary app/browser behaviour when it applies | A study-mode schedule is not an access block |
| Keep personal details private | Use an invented homework example and inspect attachments before sending | A history toggle is not a promise that all service processing stops |
| Learn from a reliable answer | Compare the answer with a teacher-approved independent reference | A citation or a second chatbot answer is not confirmation |
If an access check fails, first confirm the account and whether a setting applies to a new conversation or an existing one. Some changes take time to apply. Keep the agreed boundary in place while you resolve the mismatch, rather than repeatedly changing several controls and losing track of which one affected the result.
A family agreement can remain useful when a feature is unavailable: choose an adult-led activity permitted by the service, a non-AI learning resource or a smaller supervised task. Do not assume that buying a subscription removes an age or account restriction.
A practical family AI agreement
- Approved access: Ask before adding a chatbot, browser extension or companion service. Use the agreed account and keep its age information accurate.
- A purpose and stopping point: Agree what the session is for and when it ends, such as understanding one homework question. Fit use around sleep, school and family routines rather than treating a suggested number of minutes as a medical rule.
- Private information stays private: Do not paste passwords, addresses, school timetables, private messages or identifying family documents. Remove names and identifying details from a homework prompt.
- Check important answers: Open the cited source and check that it supports the claim. For schoolwork, compare the explanation with the teacher’s materials or textbook.
- People handle personal problems: A chatbot can produce reassuring language without being a friend, teacher, clinician or trusted adult who knows the child.
- A safe reporting rule: If something feels frightening, sexual, pressuring or confusing, stop and tell a trusted adult. Asking for help should not itself trigger punishment.
Keep the agreement short enough that the child can explain it back. The family online-safety agreement can carry the wider device rules; add the AI-specific boundaries above rather than maintaining conflicting sets of promises.
Protect personal information before the first prompt
Treat a prompt as information given to an online service. A useful homework question rarely needs the child’s full name, school, birth date or a photograph of a complete class document. “Explain fractions using a pizza example” gives the tool enough context without those details.
Review the account’s available history, memory, model-improvement and sharing settings separately. Turning off one does not make every conversation anonymous or guarantee immediate deletion. Google says its Keep Activity setting is unavailable below 13 or the applicable country age; do not look for an adult account’s full set of controls on a supervised child account. Check before connecting an email account, uploading files or making a public sharing link.
For a shared family device, identify the signed-in account before starting. Avoid giving a child an adult account as a workaround for age restrictions. Agree who manages recovery without unnecessarily circulating account passwords. Recheck access after an account switch or app installation; the agreement should cover every approved route.
If Personal Information Was Sent by Mistake
Ask what was sent and where before deciding what to do. An address, school document, password and public sharing link create different problems. Stop further sharing, review any public link and use the service’s available deletion or reporting options. Removing a visible chat is not evidence that every retained copy has disappeared.
If a real password or recovery code was exposed, change or replace it at the service that issued it, using the official app or website. If the upload contains another person’s private information, involve the responsible adult, school or organisation as appropriate. Keep the response focused on reducing exposure; blaming the child makes the next mistake harder to report.
Teach verification with an ordinary task
Choose a low-stakes school topic you can check. Ask the chatbot to explain one part, then compare its answer with a textbook, school resource or relevant primary source. If it provides a link, open it together. Does the page exist, refer to the right topic and actually support the sentence? A polished citation is not verification.
Ask the child to explain the answer in their own words and identify what remains uncertain. For a calculation, check the steps independently. For a current fact, check its date. Agree with the teacher how assistance should be acknowledged. The goal is to retain the learning task, rather than simply produce convincing text.
Avoid asking a chatbot to prove itself correct as the only check. A second fluent response can repeat the same mistake. Use a source that is independent of that conversation.
Keep human relationships at the centre
A conversational tool can feel personal because it responds quickly and follows the thread. Talk about that design without embarrassing a child who enjoys using it. Ask what they find helpful and what they would rather discuss with a person.
Look at the effect on daily life: sleep, schoolwork, friendships, activities and whether the child can stop comfortably. A single private conversation or an interest in a fictional character does not establish a disorder. If use is displacing those activities or causing distress, reduce access together and involve a trusted adult or qualified professional as appropriate.
Do not make a chatbot the plan for a serious wellbeing problem. Listen, ask what happened and help the child reach human support. If anyone is in immediate danger, contact local emergency services.
What to do after an upsetting response
- Stop the conversation and move attention back to the child. Ask what they saw and how they are feeling.
- Use the service’s reporting or feedback controls where appropriate. Keep only the information needed to explain the problem; avoid redistributing identifying or intimate material.
- Check the account, enabled features and access route. A problem in one chatbot does not prove another product has the same controls.
- Agree whether to pause that service, narrow its permitted use or arrange further human support. Revisit the decision after the immediate situation has settled.
A reassuring opening is: “You did the right thing by telling me. Let’s work out what to do together.” The response should make future reporting easier, while still setting a clear boundary around harmful content.
Review the agreement when something changes
Revisit the settings when the service introduces a new feature, the child changes account or device, or school expectations change. Ask the child to show how they access it and what happens when a limit applies. Keep the rule connected to a real situation rather than checking a dashboard for its own sake.
For a wider device decision, compare built-in controls and paid apps. Start with the controls that meet the family’s actual need. A tool that restricts access can still be useful without promising to read or evaluate every private conversation.
The Family’s Next Step
Choose one permitted activity, one privacy rule and one trusted adult to contact. Check the account and access route together, then revisit the agreement after a feature, device or school-policy change. Treat any notification as a reason to check in, and the absence of a notification as no assurance of safety.
Frequently asked questions
Can parental controls read every AI chatbot conversation?
No. Access restrictions and activity reports do not automatically expose conversation text. ChatGPT account linking specifically does not provide parental access to chat history or real-time conversations.
Can a parent allow a child below a chatbot’s minimum age?
Parental permission does not override the service’s rules. Check the exact product, account and country. Some services support supervised younger accounts; others require an adult to conduct any interaction with younger children.
What is a sensible time limit?
Choose a limit around the task, sleep and family routine. There is no single minute allowance in this guide that establishes safe use for every child. Make the stopping point clear and check how use affects daily life.
Does a safety filter guarantee appropriate answers?
No. Treat filters as one layer. Keep private-data rules, independent fact checks and a trusted adult available when an answer is uncomfortable or important.
Do families need a paid monitoring app for AI?
Not automatically. Start with the service’s native settings, eligible account supervision and existing device controls. Consider another tool only for a specific supported access or reporting gap.
Sources
OpenAI: age and educational-use requirements; OpenAI: parental-control scope and limitations; Google: supervised Gemini access, availability and privacy guidance.
Share this article
Stuart Kerr Spindlow is Impulsec’s Publisher, Chief Editor and Lead Reviewer. He writes and maintains coverage of consumer cybersecurity and digital-safety software, with a focus on practical suitability, limitations and the evidence behind product claims. Read Stuart’s full profile.