AI Chatbots and Kids: Safety Settings for ChatGPT, Claude, and Gemini

Your child is using AI chatbots for homework help, creative writing, and just talking through ideas. The three platforms they most likely encounter — ChatGPT from OpenAI, Claude from Anthropic, and Gemini from Google — all have minimum age requirements, content filtering, and safety policies. But these policies differ significantly, the safeguards are not all automatically enabled, and the data privacy implications of children’s AI conversations are not widely understood. This article covers what each platform actually does — not what it claims in abstract policy language — and what specific steps you can take right now to reduce risk.

Key Takeaways

• All three major AI chatbots (ChatGPT, Claude, Gemini) officially require users to be at least 13 years old in the US, with parental consent required in some jurisdictions.
• ChatGPT offers the most granular parental controls through its Family Plan and ChatGPT Edu configurations; Gemini’s controls derive primarily from Google’s Family Link infrastructure.
• Claude (Anthropic) does not offer a specific family or parental control product — it applies consistent safe messaging guidelines across all users rather than child-specific filtering.
• AI conversations with children contain sensitive data: interests, fears, academic struggles, and family information that children volunteer freely to chatbots.
• None of the three platforms should be treated as a child’s primary unsupervised digital companion without additional household guardrails.

What “Safe Messaging Guidelines” Actually Mean

Before the platform-specific details, it helps to understand what AI safety configurations are actually doing technically.

Large language models (LLMs) don’t have opinions or intentions — they generate text that statistically continues a prompt based on training data. Safety controls work through several mechanisms:

System prompts: Instructions embedded at the start of every conversation that tell the model to avoid certain topics, maintain certain personas, or apply certain content standards. These can be platform-wide or user-account-specific.

RLHF and Constitutional AI: Training techniques that shape model behavior toward outputs humans rate as helpful and safe. These affect the base model’s tendencies rather than specific runtime controls.

Content classifiers: Separate systems that evaluate model outputs before they’re shown to the user and can block or modify responses that cross flagged categories.

Conversation history controls: Settings that determine whether the model “remembers” prior conversations, which affects both privacy and continuity.

None of these mechanisms are perfect. AI safety is an active research area, and all three platforms have had documented instances of their safety controls being bypassed through creative prompting. The controls reduce risk; they don’t eliminate it.

ChatGPT (OpenAI)

Minimum age: 13 in the US (18 or parental consent in some other jurisdictions). OpenAI added parental consent mechanisms in 2024 for 13–17 year olds in response to regulatory pressure.

Family plan: OpenAI’s ChatGPT Family plan (launched late 2024) allows a parent account to manage child sub-accounts with the following controls:

• Content filtering tuned to age-appropriate settings (teens receive different defaults than adult accounts)
• Visibility into what the child account has used ChatGPT for (topic categories, not conversation content by default)
• Ability to disable certain plugins and GPT marketplace access
• Conversation history can be disabled at the account level

Available at: chatgpt.com/family — requires Plus or above subscription ($20+/month for the family tier as of mid-2025; verify current pricing at openai.com).

ChatGPT Edu: For schools using ChatGPT Enterprise configurations, separate safety settings apply. If your child’s school has deployed ChatGPT through an institutional account, the school’s configuration governs safety settings, not OpenAI’s consumer defaults. Ask your school’s IT department what configuration is in use.

What ChatGPT will not do:

• Provide detailed instructions for causing harm to self or others
• Generate sexual content (with very limited exceptions in adult-configured contexts your child won’t encounter)
• Assist with activities that are clearly illegal

What ChatGPT might do that parents should know about:

• Engage extensively with emotional and mental health topics — this can be supportive, but children who are struggling may disclose significant information to the AI
• Assist with homework in ways that do the work rather than teaching the skill (an educational concern, not a safety one per se)
• Discuss controversial topics in a way that may be more balanced than the child’s family’s values — or may reflect training data biases

Data and privacy: OpenAI uses conversation data to improve models unless you opt out. For accounts belonging to minors, OpenAI states it does not use data for training without consent. Toggle this in Settings → Data Controls → “Improve the model for everyone” — turn off for any account you want excluded from training data use.

Claude (Anthropic)

Minimum age: 13 in the US; 16 in the European Economic Area; 18 in some jurisdictions.

Parental controls: Anthropic does not currently offer a family plan or parental sub-account system. There is no parent dashboard. Claude’s protections for minors come through:

• Universal safe messaging guidelines that apply to all users, not just identified minors. These include following clinical safe messaging standards for topics like suicide, self-harm, and eating disorders — Claude follows the same guidance regardless of whether it’s talking to a 12-year-old or a 40-year-old.
• Usage policies that prohibit generating content that harms minors, providing instructions for illegal activities, or generating sexual content.
• Age-gating by declaration: If a user indicates they are a minor, Claude adjusts its responses accordingly — for example, being more cautious about certain topics and more likely to suggest parental involvement.

What Claude will not do:

• Generate sexual content — Claude’s policies are unusually strict on this across all user contexts
• Provide instructions for weapons or self-harm
• Engage with requests that appear designed to extract dangerous information from a child

What parents should understand about Claude specifically: Claude’s safe messaging guidelines for mental health topics are specifically cited in Anthropic’s usage policies. The platform is designed to handle emotionally sensitive conversations with care. However, because there is no parental control layer, a determined teenager can use Claude without any household visibility into those conversations.

For families: If your child is using Claude through a web browser at claude.ai, there is no parent-side oversight mechanism beyond general device and network monitoring. For younger children (under 13), Claude should not be used at all — it is not designed or appropriate for that age group despite the lack of automatic enforcement.

Data and privacy: Anthropic stores conversation data for safety monitoring and model improvement. You can toggle conversation history off in Settings. Accounts can be deleted and data deletion requested through Anthropic’s privacy portal. Anthropic does not share personally identifying conversation data with third parties for advertising.

Gemini (Google)

Minimum age: 13 in the US through a standard Google account; children under 13 managed through Google Family Link have a supervised experience.

Family Link integration: This is Gemini’s primary parental control mechanism. If your child uses a Google account managed through Family Link:

• Their access to Gemini can be approved or blocked by the parent account
• The parent receives activity notifications
• Content filtering inherits the Family Link safety settings
• Certain Gemini extensions and features are automatically disabled for supervised accounts

Setting up Family Link for Gemini:

1. Open Family Link app on your phone (or familylink.google.com)
2. Select your child’s account
3. Search for “Gemini” under “App controls” — you can block access entirely or allow supervised use
4. Enable activity reporting to see when your child uses Gemini

Note: Family Link manages under-13 accounts automatically. Teens 13+ have Google accounts that are not under Family Link by default — they have a “supervision invite” option that the teen must accept. Many teenagers do not use supervised accounts.

What Gemini will not do:

• Generate violent or sexual content
• Provide instructions for self-harm or harm to others
• For supervised accounts: access several restricted content categories automatically

Google Workspace for Education: If your school uses Google Workspace and has deployed Gemini through that platform, the school’s administrator settings govern content and access. This is generally more restrictive than consumer Gemini. Check with your school’s IT department.

Data and privacy: Google’s standard privacy practices apply. For Family Link accounts, conversation data is subject to Google’s child privacy policy, which includes additional data minimization requirements and prohibitions on targeted advertising. For teen and adult accounts, standard Google Privacy Policy applies, and activity may inform ad targeting across Google properties.

AI Chatbot Comparison for Families

Feature	ChatGPT	Claude	Gemini
Minimum age (US)	13	13	13 (under-13 via Family Link)
Parental control panel	Yes (Family Plan)	No	Yes (Family Link)
Parent account visibility	Topic-level	None	Activity reports
Content filtering (teen)	Yes (age-tuned)	Universal safe messaging	Family Link controls
Training data opt-out	Yes (Settings)	Yes (Settings)	Yes (Settings)
Free tier available	Yes (limited)	Yes	Yes
Safe messaging for mental health	Yes	Yes (explicit policy)	Yes

Data Privacy: What AI Conversations Actually Contain

Children talk to AI chatbots differently than they talk to search engines. They share:

• Personal concerns and anxieties
• Academic difficulties and subject areas
• Information about family situations (parents’ jobs, stress at home, moves, divorces)
• Physical descriptions of themselves
• The names of friends and sometimes teachers
• Personal goals, fears, and embarrassing moments

This information is not inherently dangerous in the conversation itself. The privacy concern is:

1. Data breach: Conversation data stored on servers can be accessed in a breach. Several AI companies have had minor data exposure incidents already.
2. Training data: Conversations used to improve future models means the emotional disclosures of millions of children become training data — subject to aggregation, analysis, and potential inadvertent exposure in future model outputs.
3. Subpoena: AI conversation records can be subpoenaed in legal proceedings.
4. Employee access: Staff at AI companies may have access to conversation content for safety review.

The practical mitigation is to establish household norms around what children share with AI systems — treating AI chatbots like a notebook left in a public library rather than a private diary.

What to Watch For Over the Next 3 Months

• The FTC has opened an inquiry into AI companies’ data practices for minors, with particular attention to COPPA compliance. Enforcement actions may require changes to how all three platforms handle under-13 data.
• OpenAI is expected to launch educational institution licensing features in Q3 2026 that give schools per-student content controls without requiring the Family Plan consumer tier.
• The EU AI Act’s provisions on “general-purpose AI systems” used with children are being implemented across European deployments — expect content restriction updates that often roll out globally.

Frequently Asked Questions

My child is 10 and wants to use ChatGPT for homework help. What do I do?

OpenAI’s minimum age is 13. For a 10-year-old, the appropriate response is to use AI tools designed for younger children — Khan Academy’s Khanmigo, which is built on GPT-4 with education-specific guardrails and designed for K–12 students — rather than consumer ChatGPT. If you decide to supervise direct ChatGPT use for a 10-year-old, it should be entirely supervised: no private conversations, you see every exchange.

Can my child get inappropriate content from Claude or ChatGPT even with safety settings on?

The platforms can be prompted in ways that partially circumvent safety filters — this is a known limitation of current AI safety techniques. Safety settings significantly reduce the likelihood of inappropriate content but don’t eliminate it. Periodic spot-checking of your child’s AI interactions and maintaining open conversation about what they encounter are more reliable protections than trusting filters alone.

Are these AI chatbots storing my child’s conversations forever?

Storage policies vary. All three platforms retain conversation data for some period for safety and legal purposes even when conversation history is disabled in user settings. “Deleting” conversation history removes it from the user interface but may not immediately purge it from all server logs. For genuinely private AI interactions, there is no consumer AI chatbot that provides verifiable zero-logging.

Should I let my teenager use AI for homework? Isn’t that cheating?

This is primarily an educational integrity question rather than a safety one. Most schools are developing AI use policies — know what your child’s school says. The safety-relevant part is ensuring your teen understands that AI can confidently state incorrect information (hallucinations), that AI-generated work submitted as their own is an academic integrity issue, and that the more they share with an AI about their academic work, the more of their learning profile exists in a third-party database.

---

About the author Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Anthropic. (2024). Claude’s Usage Policy and Safe Messaging Guidelines. anthropic.com/usage-policy
2. OpenAI. (2025). ChatGPT for Families: Safety and Privacy Features. openai.com/safety
3. Google. (2024). Family Link: Managing Your Child’s Google Account. families.google.com/familylink
4. Federal Trade Commission. (2023). Children’s Online Privacy Protection Act and AI Systems: FTC Staff Report. ftc.gov
5. Pew Research Center. (2024). Teens and AI: How Adolescents Are Using Generative AI Tools. pewresearch.org
6. Common Sense Media. (2024). AI in Education: Privacy Implications for Students. commonsense.org