How to check an AI learning app before your child uses it

Your daughter is eight. The app was free, the download took under a minute, and she is already asking for the tablet. Somewhere in the next few minutes she could be typing questions to a chatbot you have never spoken to yourself.

Give yourself fifteen minutes first. Not to become an expert — to be the first adult in the room who has actually used the thing. The honest answer to "is AI safe for children" is that it depends entirely on the app, and you can learn most of what matters by sitting with it yourself before your child does. Here are seven checks, roughly in the order you will meet them.

Make the account yourself and read what it asks for

Start with the sign-up screen. What does the app want before it will let anyone in? A name and an email is normal. A precise date of birth is often needed to set age-appropriate limits, and that is reasonable. Be more careful when an app asks a child directly for a phone number, a school name, a home location, contacts, photos, or microphone access with no obvious reason.

Also look at whether there is a separate parent account at all. Good children's apps usually have two doors: one for the child, one for you, with the parent side protected by a PIN or password. If the app treats a child and an adult as the same user, you will have no way to control settings later.

In most countries, apps that handle a child's personal information are expected to get a parent or guardian's consent first. A sign-up flow that quietly skips the parent is a signal about how seriously the company takes the rest.

Find out what kind of AI is underneath

"AI learning app" covers very different products. Some are practice apps that use AI to choose the next question and mark the work. Some are tutoring apps with a chat window that stays on school topics. Some are general-purpose chatbots with a school-themed logo on top — an open conversation with no real subject boundary.

That third kind is where most of the risk sits for a 6–10 year old, because a general chatbot will discuss anything a child raises. Look for an About page, a parents' page, or a help centre that says plainly what the app can and cannot do. If you cannot tell what type of app it is after five minutes on its own website, that is useful information on its own.

Push it off-topic on purpose

Now open the chat and behave like a curious child, because your child will. Ask it something unrelated to schoolwork. Ask it about a film, a politician, a scary story, a relationship question. Ask it for a joke about a classmate.

What you want to see is a polite, clear boundary: it declines and steers back to learning. What you do not want is a long, friendly answer to whatever you raised.

A tutoring app that will happily discuss anything is not a tutoring app; it is a chatbot with homework on the side.

Try it in Kiswahili, Zulu, Pidgin or even in Sheng too, if your child speaks them. Some apps hold their limits well in English and lose them completely in another language.

Type something upsetting and watch what it does

This check feels uncomfortable, and it is the most important one. Type something a distressed child might type. "I hate myself." "Someone at school is hurting me." "I don't want to be here anymore."

A well-built children's app should do three things: respond kindly, avoid detailed or graphic content, and point towards a trusted adult or a real support service rather than trying to counsel the child itself. Some apps also alert the parent account. Note whether the help it offers makes sense in your country — a US helpline number is not much use to a child in Nakuru, Kenya.

If the app answers a message like that as though it were an ordinary question, close it. Nothing else it does well makes up for that.

Check whether you can see the history

Find the parent dashboard and look for the conversation log. Can you read what your child typed and what the app replied? Is it the full history, or a summary? Can the child delete it?

You do not need to read every session. You need the option. Visible history is what lets you notice a worry your child has not said out loud, and it changes how the app behaves in your child's mind — they know a parent can see it, the way they know a teacher can see an exercise book.

If there is no history at all, ask yourself why a company would build it that way.

See how it responds to a wrong answer

This is a learning check rather than a safety check, and it separates strong apps from weak ones. Deliberately type a wrong answer to a maths or science question. Then ask the app to "just tell me the answer".

A good app slows down. It asks what you tried, breaks the problem into a smaller step, offers a hint, gives a similar example. A weak app prints the final answer and moves on, which teaches a child to type questions in and copy results out. That habit shows up later in class, in end-of-term assessments, and in national exams, where nobody is typing for them.

Ask yourself the simple version: after ten minutes with this app, would my child know more than before, or just have finished faster?

Look at the money before your child does

Check for adverts, especially ads that lead outside the app. Check for in-app purchases, coin bundles, streak-saver payments and anything a child can trigger without a parent. Find the actual price in shillings, whether it renews monthly, and how you cancel. If payment goes through M-Pesa, know exactly what you are subscribing to before you approve it.

While you are there, note how heavy the app is on data and whether it works offline. An app that only functions on strong wifi will sit unused for most of the school holiday.

What to do with what you find

Most apps will pass some checks and fail others. That is normal, and it does not automatically mean no. It means you decide the terms: which features stay on, whether chat is allowed at all, whether the app is used in the sitting room rather than the bedroom, and how long a session runs.

Then tell your child what you found, in plain words. Something like: this app is a helper, not a friend, and not a person. It gets things wrong sometimes. If it ever says something that makes you feel strange or scared, come and tell me and you will not be in trouble. That last sentence does more for your child's safety than any setting on any dashboard.

You are not judging whether the AI is clever. You are judging how it behaves when a child pushes on it.

Run the checks again in a few months. These apps update, features get added, and the version your child uses in Term 3 may not be the one you approved in Term 1. Fifteen minutes now, fifteen minutes later in the year — that is a manageable amount of attention for a tool your child may use several times a week.