Is AI tutoring safe for kids?
It depends entirely on the product, and the differences between them are larger than the category name suggests. Four things separate a tool you can leave a child with from one you cannot: what happens to their data, what the system does when a conversation turns serious, whether a parent can see what happened, and whether the tutor teaches or simply supplies answers. All four are checkable before you sign up.
Data is the first question, not the last
For a product used by a child, the questions are narrow and answerable. What is collected, how long it is kept, whether it trains a model, and whether anything is sold or shared for advertising. A privacy policy that will not answer those plainly has told you something.
In the United States, COPPA sets requirements for services directed at children under 13, including verifiable parental consent. It is a floor rather than a guarantee of good practice, and worth knowing the product's answer rather than assuming compliance.
What happens when a conversation goes somewhere serious
A tutor a teenager talks to daily will eventually receive something that is not about homework. The right behavior is to stop tutoring and route to real help, not to counsel, and not to continue the lesson as though nothing was said.
This is worth asking about explicitly, because it is rarely in the marketing. Ask what the product does if a student expresses distress, and whether a parent is informed.
Teaching or answering
A safety question that does not look like one: a tutor that returns finished answers is not dangerous, but it does quietly remove the work the student was supposed to do. Over a term that shows up as homework grades and test grades moving in opposite directions.
The check is simple. Give it a problem your child is currently learning. If it produces a complete solution immediately, it is an answer service. If it asks a question back, offers a hint, and expects the student to take the next step, it is a tutor.
What a parent should be able to see
Not a transcript of every message, which a teenager may reasonably object to, but enough to know what is happening: what subject the time went into, what skills are improving, and whether anything was flagged.
A product that shows a parent nothing is asking for a lot of trust. One that shows everything creates a different problem, which is a student who stops using it honestly.
The checks, in one list
Worth running against any candidate, including ours.
- Does the privacy policy state plainly what is collected about a child and how long it is kept?
- Is anything sold or shared for advertising? Are there ads in the child experience at all?
- What does it do when a student signals distress, and does it stop tutoring?
- Does it hint and hand the work back, or does it produce the answer?
- What does a parent see, and can they see it without reading every message?
- Is there a way to delete the child's data, and is it findable without contacting support?
Our own answers
SmartScroll is used by children, so it is fair to hold us to the list above. For a child we store a first name, an optional grade, and learning progress. There are no ads and no data selling. If a student signals crisis the tutoring stops and routes to real help, including the 988 line in the United States. The tutor guides with a hint ladder and ends by handing the step back rather than completing the work. Parents get per-skill mastery and a weekly summary rather than a message log.
The reasoning behind those decisions is published in our Algorithm Charter, which exists so the claims on this page can be checked rather than taken on trust.
Questions parents ask
Is it safe to let a 12 year old use an AI tutor?
It depends on the product, not on the age. Run the checks above. A tool with no ads, a clear data policy, a crisis path, and parent visibility is a different proposition from a general-purpose chatbot, even though both get described as AI.
Can an AI tutor replace a human tutor?
For daily practice and unblocking homework it does a lot of what a human does, at a fraction of the cost and whenever the student is stuck. For a student who needs someone to notice they are discouraged, it does not. We would not claim otherwise about our own product.
What about general chatbots for homework?
They are capable and they are not built for this. They will usually produce the finished answer, they were not designed with a child in the room, and their data terms are written for a general audience. That is a different risk profile from a product built for students.
How do I know it is not just making things up?
You do not, entirely, and any product claiming otherwise is overstating it. What you can check is whether the product grounds its teaching in a defined curriculum, whether it shows its reasoning so an error is visible, and whether a parent can see what was covered.
Where to go next
Sources
- The Children's Online Privacy Protection Act (COPPA), US Federal Trade Commission, for the parental-consent requirements referenced above.
- The 988 Suicide and Crisis Lifeline, referenced as the US crisis routing destination.
