AI age limits, app by app: the minimum age for every tool your kid asks about

The short version, in one place

Parents almost never ask about age limits in the abstract — they ask because a specific app is already on a specific device. Here is what each one requires today.

• **ChatGPT (OpenAI)** — 13 minimum; users 13–17 need parental permission. Parent-linked teen accounts with quiet hours and content controls are available.
• **Google Gemini** — 13 in most regions via a Google account; younger children can get supervised access through Family Link, with a more restricted teen experience.
• **Claude (Anthropic)** — 18 in the terms of use. There is no teen tier, so a child using Claude is using a parent's account, with a parent responsible for it.
• **Microsoft Copilot** — tied to the Microsoft account age and local rules (13 in the US, up to 16 in parts of Europe); school-issued Copilot accounts follow the district's settings, not yours.
• **Snapchat My AI** — no separate gate beyond Snapchat's 13. If Snapchat is allowed, My AI is on; it is a Snapchat decision, not an AI decision.
• **Character.AI** — states 13 (16 in the EU), but it is a companion product built for engagement. Treat it as 18+ regardless of the printed number.
• **Khanmigo (Khan Academy)** — 13 for a student account, with younger children supported through a parent- or teacher-linked account. The most age-appropriate option on this list.
• **Perplexity, Copilot-style search assistants** — generally 13, sometimes 18. Their real risk is not chat but the open web results they summarise.

Companies revise these terms without announcement. Check the app's own terms page before you rely on any number, including this one.

Why 13 is the number almost everywhere

It is not a developmental judgment. In the US, COPPA imposes strict consent and data-handling duties for users under 13, so the cheapest compliance path is to declare the service off-limits below that age. Similar logic drives the 16 you'll see in parts of Europe under GDPR.

So the age limit tells you when a company stopped being legally exposed. It tells you nothing about whether your particular 12-year-old should be typing questions into a model that answers everything with total confidence.

What the age gate actually checks

A birthdate field. That's it, in nearly every case — no ID, no card, no parent email. A child who can subtract can create an adult account in about eight seconds, which means the enforcement in your house is the only enforcement that exists.

Two practical consequences. First, assume the account age on any app your child set up themselves may be wrong, and check it — a wrong birth year silently disables every teen protection the platform offers. Second, do not treat "it says 13+" as a reason to relax; treat it as the floor beneath the decision you still have to make.

Age limit vs. the right age for your kid

The rule that has held up best across the families we hear from is simpler than any terms page: access follows demonstrated judgment, not birthdays.

• Under 10 — no account of their own on any of these, including the ones that would allow it. Shared use on your device, with you in the room.
• 10–12 — a moderated tutor is fine; general assistants only on a parent-owned account. The goal at this age is the habit of checking an answer, not access.
• 13–15 — their own account where terms allow, linked to yours via parental controls, with homework rules agreed the same day the account is created.
• 16–18 — independence with conversations rather than monitoring. Companion apps stay off the list the whole way through.

The per-age setup for the most common case is covered in how old you have to be to use ChatGPT, and the toggles worth changing are in ChatGPT parental controls.

If your child is already past the gate

Most parents find out about the account after it exists. Deleting it rarely helps — it moves the use to a friend's phone. Fix the account age, switch on whatever controls the platform offers, and put the homework line in writing: family AI homework rules takes about twenty minutes and outlasts any setting.

Then, when work comes back and you can't tell what happened, paste it into our AI homework checker — you'll get a plain-language read on which parts sound like your kid, and the questions to ask them out loud.