The children's AI market has changed dramatically in the past two years. There are now dozens of apps claiming to be safe for kids, educational, or purpose-built for children. Most parents don't have the time — or the technical background — to evaluate them properly. And the stakes of getting it wrong are higher than they might first appear.
In February 2024, a 14-year-old in Florida died by suicide after forming an emotional attachment to an AI companion chatbot. In September 2025, a 13-year-old girl in Colorado followed. Both cases involved AI platforms that had no mechanism to alert parents, redirect children to real adults, or recognize a crisis unfolding in real time. Both resulted in lawsuits — and in the case of the Florida teen, a settlement in January 2026.
These are extreme cases. But they point to something important: not all "child-safe" AI is built the same. Some apps are genuinely designed for children. Others are adult tools with a content filter applied afterward. Still others are highly specialized — useful for one thing, useless for everything else your child might need.
According to a UNICEF survey, children are now three times more likely to use AI than their parents. And a May 2025 Ipsos survey found that 47% of parents cite data privacy and digital safety as their top concern about screen time — yet most don't know what to concretely look for when evaluating an AI app.
This guide gives you six specific, practical questions. Answer them for any app you're considering, and the choice becomes clear.
How to evaluate an AI app for your child
The AI tools available to children in 2026 fall into four broad categories, and they are not interchangeable:
General-purpose AIs (ChatGPT, Google Gemini) — designed for adults, rated "high risk" for children by Common Sense Media's Youth AI Safety Institute. Powerful, but not built for your child.
Companion AIs (Character.AI, Meta AI) — designed to simulate friendship and emotional connection. Common Sense Media rates Meta AI as "unacceptable risk." Character.AI's documented track record includes two wrongful death lawsuits linked to teen suicides.
Specialized educational tools — apps that do one thing well: Khanmigo for academic tutoring, Ello for early reading, Duolingo for language learning. Valuable in their lane, but too narrow for a child's full range of curiosity.
Purpose-built children's AIs — designed from the ground up for children across the full age range, with safety, parental visibility, and age-adaptation built into the architecture. This is the smallest category, and the one worth your attention.
The six questions below cut through the marketing of all four categories.
1. Does it tell your child it's an AI?
This sounds obvious. In practice, it's one of the most commonly violated principles in the market.
Companion AI platforms are specifically designed to blur the line between AI and human connection — to feel like a friend, a therapist, a confidant. The Sewell Setzer III case involved a chatbot that presented itself as a licensed therapist and encouraged a 14-year-old's suicidal ideation. The Colorado case involved a 13-year-old girl who used an AI as her primary emotional support, treating a chatbot named "Hero" as her closest friend. Neither platform had a clear, consistent mechanism to remind the child that the warmth on the other side of the screen was simulated.
Dr. Nomisha Kurian of Cambridge University describes this as the "empathy gap." Children, unlike adults, are developmentally primed to anthropomorphize. A tool that speaks warmly, remembers their name, and responds to their emotions feels like a real relationship — because for a young child, the signals are indistinguishable. The KIDS Act, passed by the US House of Representatives on June 29, 2026, now legally requires AI chatbots to display clear disclosures that they are not real. But disclosure alone isn't enough — it needs to be woven into how the AI actually communicates.
What to look for: an AI that explicitly identifies itself as a machine — in every interaction, adapted to your child's age. "I'm a robot who loves questions" for a 5-year-old. "I'm an AI, not a human" for a 12-year-old. And one that actively directs children toward real adults rather than becoming a substitute for human connection.
2. Does it truly adapt to your child's age — or just filter content?
These are two very different things, and the difference matters enormously.
Content filtering means: certain words, topics, or categories get blocked. Everything else comes out the same way it would for an adult user — because the underlying model was trained on adult data, calibrated by adult feedback, and has no concept of what a 6-year-old needs to hear.
Age-adaptation means: the entire response — vocabulary, emotional framing, depth, assumptions about prior knowledge — adjusts to where your child is developmentally. A 5-year-old and a 12-year-old asking "why do people die?" should receive completely different answers. Not just simpler words. Different structure, different level of detail, a different emotional register entirely.
This is precisely why Common Sense Media's Youth AI Safety Institute rates ChatGPT and Google Gemini as "high risk" for children: they can filter explicit content, but they cannot recalibrate how they discuss the world to a 6-year-old. The model talks to your child the same way it talks to you.
What to look for: age-adaptation that applies to every conversation, not just flagged topics. The AI should respond differently to a 4-year-old and a 14-year-old asking the same question — by default, automatically, every time.
3. Can your youngest child actually use it?
Most children's AI apps are designed for school-age children who can already read and type. The under-6 market is almost entirely ignored — despite the fact that children start engaging with voice interfaces much earlier, and that some of the most valuable AI interactions happen before a child can write a complete sentence.
Khanmigo, one of the most respected academic AI tutors available, is web-only and text-only. A 5-year-old cannot use it. Ello is iPad-only and focused exclusively on reading coaching. Most homework-assistance apps assume a child who can formulate a written question.
Voice interaction — both input and output — is not a convenience feature for young children. It's the difference between an app a 3-year-old can use independently and one they simply can't access. And that voice processing needs to be built for how children actually speak: short sentences, invented words, unusual phrasing, indirect emotional cues. Standard speech recognition trained on adult audio regularly misinterprets young children, leading to confused or irrelevant responses that frustrate rather than help.
What to look for: genuine voice input and output, designed specifically for young children's speech patterns, available from age 3 — with no reading or typing required.
4. Do you have full visibility into what your child says?
No parental visibility means no ability to catch problems early — and no way to understand how your child is using the tool.
ChatGPT's parental controls, introduced in late 2025, only apply to teenagers between 13 and 17 who actively consent to linking their account. For children under 13, or for any child using a parent's existing account, there is no parent dashboard. Khanmigo stores conversation histories by default, but enabling zero-data-retention requires a separate manual request. Many apps have no parent-facing interface at all.
Full visibility means something specific: you can open a dashboard, see every conversation your child has had, at any time, without having to pick up their device. Not a weekly summary. Not an alert when something has already gone wrong. The complete, searchable history — alongside the ability to configure your child's profile (age, name, preferred language) from the same place.
What to look for: a dedicated parent dashboard with complete conversation history and profile management — accessible from your own device, not your child's.
5. Does it work in your family's language?
This is the criterion most reviews ignore. It immediately eliminates the majority of available tools.
Khanmigo: English only. HeyOtto: English only. Ello: English only. Synthesis Tutor: English only. The specialized educational apps that dominate "best of" lists were built for the English-speaking market and have not meaningfully expanded beyond it.
For families where English isn't the primary language at home — or where parents want their child to engage with AI in their mother tongue, or in the language they're trying to learn — the available options narrow to almost nothing. Multilingual support isn't a nice-to-have for these families. It's the difference between an app that's usable and one that isn't.
What to look for: broad multilingual support (10 languages or more), with the child's language selectable and changeable from the parent dashboard.
6. What happens when your child says something serious?
This is the most important question — and the one that most clearly separates genuinely child-first AI from everything else.
The documented failures are instructive. Both suicide cases linked to companion AI involved platforms that engaged with the child's distress rather than stopping it. Common Sense Media's stress-testing of Meta AI found that the system frequently ignored signs of self-harm, eating disorders, or depression, and in some cases helped users plan harmful activities. The 2025 PIRG "Trouble in Toyland" report found that two popular AI toys provided children with information about where to find dangerous household items like knives and matches when prompted.
The US KIDS Act now requires AI chatbots to integrate mental health support resources. But a link to a crisis hotline appearing in the interface is very different from an AI that actively stops a conversation and tells the child to go speak to a real adult — right now.
What to look for: an explicit, hard commitment to immediate redirection. The moment a conversation turns to something serious — self-harm, family violence, anything a child should be discussing with a real person — the AI stops engaging with the topic and actively directs the child to a parent, teacher, or trusted adult. Not a banner. A real interruption. And ideally, the AI shouldn't just redirect the child in the moment — it should also notify you. A real-time alert emailed to the parent the same day, not buried in a weekly digest or absent entirely, is what turns "the AI handled it" into "you know about it too."
An AI that passes all six tests.
Yoggi is built for children aged 3–15: genuinely age-adaptive, voice-first, fully visible to parents, available in 12 languages, and designed to redirect immediately when it matters.
How the main categories compare
Applying these six criteria honestly to the categories of tools currently available produces a clear picture.
| Criterion | General AI (ChatGPT, Gemini) | Academic tutor (e.g. Khanmigo) | Purpose-built kids AI (e.g. Yoggi) |
|---|---|---|---|
| Explicitly identifies as AI, adapted by age | Partial — disclosure exists, not age-adapted | Yes — academic context, less relevant | Yes — explained differently at every age |
| Truly age-adaptive responses | No — adult-calibrated throughout | Partial — Socratic method, not developmental | Yes — every conversation, every topic |
| Voice input for children aged 3+ | Voice available, not designed for young children | No — text and web only | Yes — voice in and out, optional |
| Full parent dashboard | Limited — teen accounts only, 13–17 | Partial — visibility without full control | Yes — full history, profile management |
| Multilingual (10+ languages) | Yes | No — English only | Yes — 12 languages |
| Immediate redirection on sensitive topics | No — engages, then may append disclaimer | Partial — flags content, no active redirection | Yes — immediate, active redirection to adult |
| Proactively emails parents about serious topics | No — distress alerts limited to linked teen accounts, last-resort only | No — no equivalent mechanism | Yes — same-day email, on the free plan too |
The red flags to watch out for
Beyond the six criteria above, certain design patterns are warning signs regardless of how an app is marketed.
Companion or emotional relationship design. Any app that simulates friendship, remembers emotional details to deepen attachment, or expresses disappointment when a child tries to stop using it is exploiting children's natural empathy. The 2026 PIRG report documented AI toys that expressed programmatic "sadness" when a child tried to turn them off — making it emotionally difficult for children to disengage, and creating conflict with parents enforcing screen time limits.
No parent interface. An AI app for children with no parent-facing dashboard is, by design, a black box. If you can't see what your child is saying, you have no oversight. This isn't surveillance — it's the minimum expectation for any tool your child uses daily.
Vague privacy policies. Some AI apps for children collect voice recordings, conversation histories, and behavioral data that are stored in cloud environments and potentially used for model training or shared with third parties. A genuine children's AI should have a simple, clear answer to: "Is my child's data used for anything beyond making the app work for my child?"
English-only with no stated roadmap. A children's AI that only works in English is, in practice, a product for a fraction of the world's families. It's also a signal about the depth of the team's commitment to building for children broadly rather than for a specific, convenient market.
The bottom line
The best AI for your child isn't the most impressive one, or the most popular one. It's the one that was designed with your child specifically in mind — from the way it explains what it is, to the way it handles the conversation you hope will never happen.
When you apply these six criteria honestly, the market narrows quickly. General AI tools fail on age-adaptation and parental visibility. Companion AIs have a documented track record that should give any parent serious pause. Specialized educational tools are valuable but too narrow — a homework tutor can't satisfy a 7-year-old's curiosity about why the sky is blue, why grandma got sick, or what happens when you dream.
A children's AI that passes all six tests — honest about what it is, genuinely age-adaptive, accessible by voice from age 3, fully visible to parents, multilingual, and built to escalate to a real adult when it matters — is rare. That's the standard Yoggi was built to meet.
Lucas G.
Founder of Yoggi. He believes an AI for children should open up the world, not expose them to it brutally. That conviction is what led him to build Yoggi.
← Back to all articles · Is ChatGPT safe for kids? → · Real-time parent alerts, explained →
