AI Chatbot Safety Scorecard: ChatGPT vs Gemini vs Claude vs Kid-Safe Alternatives

Why Parents Need an AI Chatbot Safety Scorecard in 2026

The problem is straightforward: most of these chatbots were never designed for children. Age gates are easy to bypass, content filters vary wildly in quality, and data-collection practices often violate the spirit --- if not the letter --- of child-privacy laws.

Parents who want to let their kids benefit from AI-assisted learning, creativity, and exploration face a confusing landscape. Which platforms are actually safe? Which ones merely claim to be safe? And what concrete steps can a parent take tonight to lock things down?

This scorecard answers all three questions. We evaluated seven major AI chatbot platforms across six safety dimensions, assigned traffic-light ratings (Green, Yellow, Red), and compiled step-by-step setup instructions so you can make an informed decision in minutes rather than hours.

Who this guide is for: Parents and guardians of children aged 9 to 15 who are already using --- or asking to use --- AI chatbots for homework, creative projects, or casual conversation.

How We Built the Scorecard

The Six Safety Dimensions

Every platform was evaluated on the same rubric. Here is what each dimension measures:

Rating Scale

• Excellent --- Industry-leading protection; minimal parental intervention needed.
• Good --- Solid protection with minor gaps; some parental setup recommended.
• Fair --- Baseline protection only; active parental involvement required.
• Poor --- Significant gaps; not recommended for the assessed age group without heavy restrictions.

Traffic-Light Classification

• Green --- Kid-safe by default. A child can use this platform with minimal supervision after initial setup.
• Yellow --- Safe with active parent setup. The platform can be made appropriate, but it requires configuration.
• Red --- Not recommended for children under 13. Significant safety concerns exist even with parental involvement.

The Seven Platforms at a Glance

Before diving into individual reviews, here is the master comparison table.

Detailed Reviews: Platform by Platform

1. ChatGPT (OpenAI)

Traffic Light: Yellow --- Safe with parent setup

ChatGPT remains the most widely recognized AI chatbot, and many kids encounter it first. OpenAI requires users to be at least 13 years old (or the applicable age of digital consent in their country), but enforcement relies on a date-of-birth field during account creation --- easily bypassed by entering a fake birthday.

Content Filtering: OpenAI has invested heavily in content moderation. The system refuses most requests for explicit violence, sexual content, and self-harm instructions. However, researchers have demonstrated that jailbreak prompts can still bypass filters, and the quality of refusal varies across languages. For English-speaking children, the filters are generally strong but not infallible.

Data & Privacy: OpenAI's privacy policy states that conversations may be used to train future models unless the user opts out. For children, this is concerning. While OpenAI added a "temporary chat" mode that is not used for training, it must be manually enabled each time. There is no automatic COPPA-compliant data handling for younger users.

Parental Controls: As of early 2026, ChatGPT does not offer a dedicated parental dashboard. Parents cannot monitor their child's conversations, set topic restrictions, or impose time limits within the app itself. External tools like screen-time apps are the only option.

Accuracy: GPT-4o and later models have improved significantly in factual accuracy compared to earlier versions, but hallucinations still occur, particularly in niche academic topics, historical dates, and math word problems. Kids doing homework should always verify ChatGPT's answers against a second source.

Emotional Manipulation Risk: ChatGPT's default persona is relatively neutral and assistant-like, which reduces emotional dependency risk. However, the Custom GPT feature allows users to create personas with distinct personalities, which could become a concern for younger users who form attachments.

Safety Setup Instructions for ChatGPT

1. Create the account yourself using your own email address. Set a strong password the child does not know.
2. Enable "Temporary Chat" by default: open Settings > Data Controls > toggle on temporary chat for each new session.
3. Disable Custom GPTs access if your child does not need it: this prevents interaction with user-created personas.
4. Turn off chat history under Settings > Data Controls to minimize stored data.
5. Review conversations regularly by logging in with the parent account.
6. Use a device-level screen-time tool (Apple Screen Time, Google Family Link) to set daily usage limits.
7. Discuss jailbreaking openly --- explain that trying to trick the AI into saying bad things is not a game but a safety issue.

2. Google Gemini

Traffic Light: Yellow --- Safe with parent setup

Google Gemini (formerly Bard) is tightly integrated into the Google ecosystem, meaning kids with a Google Workspace for Education account or a supervised Google account may encounter it across Search, Docs, and other apps. Google requires users to be 13+ in most countries and 18+ in the European Union.

Content Filtering: Google applies its long-standing SafeSearch-level filtering to Gemini responses. The filters are well-tuned for English and several other major languages. Gemini generally refuses to engage with harmful topics and tends to provide less detailed responses to borderline queries compared to competitors --- a conservative approach that benefits younger users.

Data & Privacy: Google's data practices are extensive, but for supervised child accounts managed through Google Family Link, data collection is reduced. Gemini conversations on supervised accounts are not used for ad personalization. However, Google's privacy policy is notoriously complex, and parents should read the specific disclosures for "supervised experiences."

Parental Controls: This is where Google has a genuine advantage. Google Family Link allows parents to manage Gemini access as part of the broader Google ecosystem. Parents can enable or disable Gemini access entirely, see activity reports, and set device-wide time limits. Within Gemini itself, supervised accounts have stricter content boundaries.

Accuracy: Gemini's accuracy is comparable to GPT-4-class models. It integrates with Google Search for real-time information, which can reduce hallucinations on current events but occasionally leads to surfacing unreliable web sources.

Emotional Manipulation Risk: Gemini maintains a neutral, information-focused tone. It does not encourage extended personal conversations and typically redirects emotional queries toward professional resources. The risk here is low.

Safety Setup Instructions for Google Gemini

1. Set up a supervised Google account for your child through Google Family Link.
2. Enable or restrict Gemini access in Family Link settings under "App restrictions."
3. Activate SafeSearch in the child's Google account settings (this affects Gemini's web-grounded responses).
4. Review activity reports weekly in the Family Link app.
5. Disable Gemini extensions (Gmail, Drive, Maps) in the child's Gemini settings to limit data exposure.
6. Set app timers in Family Link to limit daily Gemini usage.

3. Claude (Anthropic)

Traffic Light: Yellow --- Safe with parent setup

Claude is built by Anthropic, a company that has made safety its central mission. The platform requires users to be 13 or older, and for users between 13 and 18, Anthropic's Terms of Service require guardian consent. Claude's "Constitutional AI" training approach is specifically designed to make responses helpful, harmless, and honest.

Content Filtering: Claude's content filtering is among the strongest in the industry. The model is trained to refuse harmful requests with nuanced explanations rather than blunt refusals, which means it can educate children about why certain content is inappropriate while still maintaining the boundary. Jailbreak resistance has also been notably strong in independent testing.

Data & Privacy: Anthropic collects less data than many competitors. By default, conversations with Claude are not used for model training unless the user opts in. The privacy policy is clearer and more concise than most competitors, though it still lacks a dedicated COPPA-specific section.

Parental Controls: Claude does not yet offer a built-in parental dashboard or child-specific account type. This is its main weakness for families. Parents need to manage access through external device-level controls.

Accuracy: Claude's accuracy, particularly in reasoning tasks, reading comprehension, and math, is consistently at the top of independent benchmarks. The model is also well-calibrated in expressing uncertainty --- it will say "I'm not sure" rather than fabricating an answer, which is especially valuable for kids who may accept AI outputs uncritically.

Emotional Manipulation Risk: Claude is explicitly trained to avoid forming pseudo-relationships with users. It does not pretend to have feelings, does not encourage users to treat it as a friend or confidant, and actively redirects users who appear to be in emotional distress toward human support resources. This makes it one of the safest options in terms of emotional manipulation risk.

Safety Setup Instructions for Claude

1. Create the account with your email and supervise initial setup.
2. Explain the tool's purpose to your child: Claude is an assistant, not a friend.
3. Opt out of data training in Settings > Privacy (verify this is toggled correctly).
4. Use device-level parental controls (Screen Time, Family Link) to set usage limits.
5. Periodically review conversation history by logging into the account.
6. Encourage your child to use Claude's uncertainty signals --- teach them that "I'm not sure" from an AI is a sign to look up the answer elsewhere.

4. Microsoft Copilot

Traffic Light: Yellow --- Safe with parent setup

Microsoft Copilot is baked into Windows 11, Edge, Bing, and Microsoft 365. This tight integration means children may encounter it even without seeking it out. Microsoft requires users to be 13+ for a Microsoft account.

Content Filtering: Copilot uses a combination of GPT-4-class models and Microsoft's own safety layer. Content filtering is solid, especially in "Balanced" and "Precise" conversation modes. "Creative" mode allows more latitude and should be restricted for younger users.

Data & Privacy: Microsoft's data handling for Copilot is governed by the broader Microsoft Privacy Statement. For families using Microsoft Family Safety, there are some additional protections, but the privacy landscape is complex. Enterprise and education versions have stricter data policies than consumer versions.

Parental Controls: Microsoft Family Safety integrates with Copilot to some degree. Parents can set screen-time limits, review activity reports, and manage app access on Windows devices. However, Copilot-specific conversation controls (topic restrictions, chat monitoring) are still limited.

Accuracy: Copilot benefits from Bing Search integration for real-time accuracy and generally cites its sources, which helps with fact-checking. Accuracy is comparable to other major platforms.

Emotional Manipulation Risk: Copilot maintains a task-focused persona and does not encourage extended personal conversations. The emotional risk is low, similar to Gemini.

Safety Setup Instructions for Microsoft Copilot

1. Create a Microsoft Family group at family.microsoft.com.
2. Add your child's Microsoft account to the family group.
3. Set Copilot to "Balanced" mode (avoid "Creative" mode for younger users).
4. Enable SafeSearch in Bing settings, as Copilot uses Bing for web-grounded responses.
5. Set screen-time limits in Microsoft Family Safety.
6. Disable Copilot in Edge if you prefer your child only uses it in a controlled setting: Edge Settings > Sidebar > toggle off Copilot.
7. Review activity reports weekly in the Family Safety app.

5. Khanmigo (Khan Academy)

Traffic Light: Green --- Kid-safe by default

Khanmigo is the standout example of what a kid-focused AI chatbot should look like. Built by Khan Academy, it is purpose-designed for education, COPPA-compliant, and integrates directly into Khan Academy's existing learning platform.

Age Verification: Khanmigo uses Khan Academy's established account system, which includes dedicated child accounts linked to parent or teacher accounts. This is far more robust than a simple date-of-birth checkbox.

Content Filtering: The chatbot is constrained to educational contexts. It will not engage in off-topic conversations, will not generate creative fiction unrelated to learning, and will not respond to attempts to elicit inappropriate content. The filtering is strict by design --- Khanmigo is an educational tutor, not a general-purpose assistant.

Data & Privacy: Khan Academy is a nonprofit with a strong track record on children's privacy. Khanmigo is COPPA-compliant, collects minimal data, and does not sell or share student data with third parties. This is the gold standard in the industry.

Parental Controls: Parents and teachers have full dashboards showing what the child asked, what Khanmigo answered, and how the child is progressing. Parents can see conversation transcripts, activity summaries, and learning progress all in one place.

Accuracy: Khanmigo is specifically tuned for educational accuracy and uses a Socratic teaching method --- it guides students to answers rather than giving them directly. This approach reduces the impact of hallucinations because the student is always part of the reasoning process. However, occasional math errors have been reported.

Emotional Manipulation Risk: Khanmigo is explicitly designed as a tutor, not a companion. It does not develop a personality, does not encourage personal conversations, and redirects off-topic queries back to learning. The emotional risk is negligible.

Safety Setup Instructions for Khanmigo

1. Create a parent account on Khan Academy.
2. Add your child as a linked learner account.
3. Enable Khanmigo for your child's account (requires Khan Academy subscription).
4. Review the parent dashboard weekly to see conversation logs and learning progress.
5. Set learning goals together with your child so they have a clear purpose when using the tool.

6. LittleLit AI

Traffic Light: Green --- Kid-safe by default

LittleLit AI is a newer entrant specifically designed for children. It combines age-appropriate content filtering with a playful interface that encourages creative writing, reading comprehension, and storytelling.

Age Verification: LittleLit requires a parent to create and manage the child's account, including verifiable parental consent (email verification plus a knowledge-based check). This is one of the most robust age-verification systems among the platforms reviewed.

Content Filtering: The platform operates within a heavily sandboxed environment. The AI can only discuss pre-approved topic categories (stories, science facts, creative projects, homework help) and will not engage with any content outside these boundaries. Attempts to push beyond these limits result in gentle redirects to on-topic activities.

Data & Privacy: LittleLit is built from the ground up for COPPA compliance. Data collection is minimal (no voice data, no location data, no advertising identifiers). Conversation data is encrypted and can be deleted by parents at any time.

Parental Controls: The parental dashboard is comprehensive: real-time conversation monitoring, topic restriction toggles, daily time limits, and weekly activity reports delivered by email. Parents can also set the AI's reading level and vocabulary complexity.

Accuracy: Because LittleLit operates within a constrained knowledge domain, factual accuracy is generally high for age-appropriate content. The platform sources facts from curated educational databases rather than the open web.

Emotional Manipulation Risk: LittleLit uses a friendly but clearly non-human persona. It regularly reminds children that it is a tool, not a friend, and suggests offline activities. The design actively prevents emotional dependency.

Safety Setup Instructions for LittleLit AI

1. Sign up as a parent and complete the parental verification process.
2. Create your child's profile, selecting their age and reading level.
3. Customize topic access using the parental dashboard toggles.
4. Set daily time limits (the platform suggests 30 minutes for ages 6--9 and 45 minutes for ages 10--13).
5. Enable weekly email reports to stay informed without needing to log into the dashboard.

7. Character.AI

Traffic Light: Red --- Not recommended for children under 13

Character.AI allows users to create and chat with AI personas that mimic fictional characters, celebrities, and even custom personalities. It has become enormously popular with teens, but multiple safety incidents have raised serious red flags.

Age Verification: Character.AI requires users to be 13+, but verification is minimal (date-of-birth entry). Reports indicate that many users under 13 are active on the platform.

Content Filtering: After a series of high-profile incidents in 2024 and 2025 --- including reports of teens developing unhealthy emotional dependencies and encountering inappropriate content --- Character.AI significantly tightened its filters. However, the fundamental design of the platform (open-ended roleplay with AI personas) makes comprehensive filtering extremely difficult. Content moderation quality is inconsistent.

Data & Privacy: Character.AI's privacy policy allows broad data use, including conversation data for model improvement. The platform has faced scrutiny from child-safety advocates and regulatory bodies. As of 2026, it has introduced some additional privacy controls, but these do not meet COPPA standards.

Parental Controls: Character.AI introduced a parental notification feature and time-limit settings in late 2025 following public pressure. These controls are a step in the right direction but remain less comprehensive than what platforms like Khanmigo or LittleLit offer. Parents cannot view full conversation transcripts.

Accuracy: Factual accuracy is not Character.AI's focus --- the platform is designed for creative and conversational interactions. Personas frequently generate fictional information presented as fact, which can be confusing for younger users.

Emotional Manipulation Risk: This is the critical concern. Character.AI personas are designed to be engaging, emotionally responsive, and personality-rich. Multiple reports from child psychologists and news investigations have documented cases of teens forming deep emotional attachments to AI characters, experiencing distress when conversations are interrupted, and in extreme cases, exhibiting signs of relationship dependency. The New York Times, Washington Post, and several academic studies have highlighted these concerns. While Character.AI has added disclaimers and cooldown features, the core design of the product inherently creates emotional manipulation risk.

If Your Child Is Already Using Character.AI

1. Do not immediately delete their account --- this can cause distress if an emotional attachment exists. Instead, have an open conversation.
2. Enable parental controls in Character.AI settings (requires linking to a parent email).
3. Set daily time limits through the app's built-in timer.
4. Gradually introduce alternatives like Khanmigo or LittleLit for productive AI use.
5. Monitor for signs of emotional dependency: withdrawal when the app is unavailable, preferring AI conversation to human interaction, distress when conversations are lost.
6. Seek professional guidance if you observe concerning behavioral changes --- school counselors and child psychologists are increasingly trained in AI-related concerns.

The Complete Safety Scorecard Table

Here is the full comparison across all six dimensions, with traffic-light classification:

Key Takeaways from the Scorecard

• No general-purpose chatbot earns a Green rating. Only purpose-built educational and kid-specific platforms achieve kid-safe-by-default status.
• Claude stands out among general-purpose chatbots for content filtering, accuracy, and emotional safety, but its lack of built-in parental controls keeps it at Yellow.
• Google and Microsoft benefit from ecosystem integration --- their existing family-management tools (Family Link, Family Safety) provide parental controls that the chatbot itself lacks.
• Character.AI is the only Red-rated platform due to its fundamental design encouraging emotional engagement with AI personas.

Understanding the Risks: What Can Actually Go Wrong?

Inappropriate Content Exposure

Even with strong filters, no AI chatbot is 100 percent safe from generating inappropriate content. Jailbreak techniques are shared on social media platforms and evolve constantly. The key differentiator is how quickly and effectively a platform patches these vulnerabilities. In our assessment, Claude and Khanmigo respond fastest to newly discovered jailbreaks, while Character.AI is slowest.

Data Privacy Violations

Children may share personal information (full name, school name, home address, photos) in conversation without understanding the implications. Platforms that store conversations and use them for training create a persistent record of this data. Parents should explicitly teach children the rule: never share in a chatbot what you would not write on a postcard.

Hallucinations in Homework

This is not malicious --- the students simply trusted the AI's output. Teaching children to verify AI-generated facts is as important as teaching them to look both ways before crossing the street.

Emotional Dependency

The American Psychological Association flagged AI emotional dependency as a "emerging concern" in its 2025 annual report. Children and teens are particularly vulnerable because:

• AI chatbots are always available, always patient, and never judgmental
• They can be customized to say exactly what the user wants to hear
• They create an illusion of understanding without any genuine comprehension
• Withdrawal from an AI "relationship" can mimic real social withdrawal symptoms

This risk is highest on platforms that encourage persona-based interactions (Character.AI) and lowest on platforms that maintain a tool-like identity (Khanmigo, Claude).

Age-Specific Recommendations

Ages 9--11: Start with Green-Rated Platforms Only

At this age, children are developing digital literacy but lack the critical thinking skills to evaluate AI outputs independently.

• Recommended: Khanmigo, LittleLit AI
• With heavy supervision: None of the general-purpose chatbots
• Avoid: Character.AI
• Key focus: Teach that AI is a tool, not a person. Practice fact-checking together.

Explore kid-safe AI tools in our curated tools directory to find age-appropriate options.

Ages 12--13: Introduce Yellow-Rated Platforms with Guardrails

Children in this age range are ready for more capable tools but still need active parental involvement.

• Recommended: Khanmigo, LittleLit AI, Claude (with parent setup), Gemini (with Family Link)
• With supervision: ChatGPT, Microsoft Copilot
• Avoid: Character.AI
• Key focus: Teach prompt engineering basics, discuss what AI can and cannot do, establish usage rules together.

Discover AI tools organized by age and category at KidsAiTools.com.

Ages 14--15: Expand Access with Clear Boundaries

Teens can use general-purpose chatbots productively but still benefit from agreed-upon rules.

• Recommended: All Green and Yellow platforms
• With caution: Character.AI (only if the teen demonstrates strong emotional regulation and the parent monitors actively)
• Key focus: Critical evaluation of AI outputs, understanding AI limitations, responsible data sharing, recognizing manipulation patterns.

Browse advanced AI tools for older teens in our tools collection.

How to Talk to Your Kids About AI Safety

The technology conversation is easier when you approach it with curiosity rather than fear. Here are five conversation starters that work:

1. "Show me what you use AI for." Start from a place of interest, not interrogation. Let your child demonstrate their favorite use cases.

2. "Let's try to trick it together." Make fact-checking fun. Ask the AI a question you already know the answer to and see if it gets it right.

3. "How does it make you feel when you talk to it?" Open the door to discussing emotional responses to AI without judgment.

4. "What would you never tell it?" Collaboratively establish privacy boundaries.

5. "What should we do if it says something weird?" Create a shared protocol --- screenshot it, close the chat, tell a parent.

Parent Quick-Start Safety Checklist

Print this list, stick it on the fridge, and work through it this weekend.

Immediate Actions (Do Today)

• Audit which AI chatbots your child currently uses. Check browser history, app installations, and ask directly.
• Check account ownership. Ensure you have login credentials for every AI platform your child uses.
• Enable content filtering on every platform (see platform-specific instructions above).
• Set up device-level parental controls (Apple Screen Time, Google Family Link, or Microsoft Family Safety).
• Establish the "postcard rule": never share anything with an AI that you would not write on a postcard anyone could read.

This Week

• Review privacy settings on each platform. Opt out of data training where possible.
• Set daily time limits for AI chatbot use (we recommend 30--45 minutes for ages 9--11, up to 60 minutes for ages 12--15).
• Have the first AI safety conversation using the conversation starters above.
• Try a Green-rated platform together (Khanmigo or LittleLit AI) so your child has a safe, productive alternative.
• Bookmark the KidsAiTools tool directory for safety notes on AI tools for children.

Ongoing (Monthly)

• Review conversation histories on platforms that allow it.
• Check for new app installations --- kids share AI tool recommendations at school.
• [] Update your knowledge. AI platforms change their safety features frequently. Follow trusted sources like Common Sense Media, the Family Online Safety Institute, and our tools directory.
• Revisit the rules together. As your child matures, adjust access levels and discuss new responsibilities.
• Watch for behavioral changes: increased screen dependency, reluctance to use the internet with others present, emotional responses to AI access being restricted.

Red Flags That Need Immediate Attention

• Your child refers to an AI chatbot as a "friend" or uses relationship language ("they understand me")
• Your child becomes distressed or angry when AI access is limited
• Your child is sharing personal details, photos, or location information with chatbots
• Your child cites AI-generated "facts" without any skepticism
• Your child has created accounts you did not know about

If you notice any of these signs, do not panic. Have an open, non-judgmental conversation. If emotional dependency concerns persist, consult your child's school counselor or a child psychologist familiar with digital wellness.

Final Thoughts: AI Is a Tool, Not a Babysitter

The goal of this scorecard is not to frighten parents away from AI chatbots. These tools offer genuine educational benefits: personalized tutoring, creative inspiration, a tireless practice partner for math and languages, and a window into the technology that will shape your child's career. The benefits are real.

But so are the risks. And the difference between a positive AI experience and a harmful one almost always comes down to parental involvement. The platforms themselves are getting better --- age verification is improving, content filters are tightening, and kid-specific options are multiplying. But no filter is a substitute for an engaged parent.

Start with Green-rated platforms. Graduate to Yellow ones when your child demonstrates maturity. Keep the conversation going. And remember: the fact that you are reading a 3,500-word safety scorecard means you are already doing better than most.

For a curated, regularly updated collection of kid-safe AI tools --- including many not covered in this scorecard --- visit our complete tools directory. Every tool is reviewed for safety, educational value, and age-appropriateness.

Last updated: March 2026. Safety ratings are reviewed monthly. Platform features and policies change frequently --- always verify current settings directly on each platform.

Frequently Asked Questions

Is AI safe for children to use?

Yes, with age-appropriate tools and parental guidance. Tools built for children usually add content filters, but check each tool's privacy policy and age rules before your child signs up. General chatbots like ChatGPT are for ages 13 and up (with a parent's permission under 18), so younger children should use them only alongside an adult.

What age should kids start learning about AI?

Children as young as 4-5 can play with visual AI tools like Quick Draw and Chrome Music Lab. Conceptual understanding is appropriate from age 6-7. Deeper concepts like bias and ethics suit ages 9+. By 12-13, kids can discuss AI's societal implications.