How to build an AI tutor that refuses to do the homework
The same AI that will write your child's essay for them can be turned into a tutor that only gives hints - with nothing more than a few written rules. Here is exactly what those rules say.
A mother asked us, fairly bluntly, why she should let her son near an AI at all if it was just going to write his essays for him. So we built the tutor with her, on the spot, using nothing but the context box and about forty words of rules. Twenty minutes later her son asked it to write his book report. It refused, and asked him what the book was about instead.
Nothing about the AI changed. The same model that will happily write a finished essay for a stranger will refuse to for a child whose context tells it not to. The difference lives entirely in what was written into that context.
How do I stop AI doing my child's homework?
You write context that describes a role, not a request. Instead of hoping a child remembers to ask for hints, the rule sits above every question and applies whether they remember to ask nicely or not.
What it looks like in an actual answer
Student: "Why did the volcano model in my science project not work?" → AI explains the exact chemical reaction and hands over the fix.
Student: "Why did the volcano model not work?" → AI: "What happens to gas when the opening it is escaping through is too narrow?"
Children try to break it, and that is useful
Almost every child, faced with a tutor that will not just answer, tries to talk it out of the rule within the first five minutes - "just this once," "my teacher already told me the answer, I just forgot it," "I'm not really doing homework right now." A well-written rule holds up to this, and watching it hold up is itself a small, useful lesson in what a firm boundary looks like.
The honest limit
A tutor built this way is only as strict as the words in its context. A vaguely written rule - "try to give hints when possible" - gets talked around eventually. A specific one - "never write a full sentence of the final answer" - mostly does not. Writing the rule precisely is, itself, a form of the context engineering the next article in this level pulls apart in more detail.
Questions we get asked
How do I stop AI doing my child's homework?
Write context that gives the AI a role and a firm rule - for example, "you are a tutor who never gives the final answer, only guiding questions" - and place it at the top of every chat. The same AI that will write a finished essay on request will refuse to under rules like these, because the rule applies to every message, not just the polite ones.
Can children talk an AI tutor out of its own rules?
They will certainly try, usually within the first few minutes, with excuses like "just this once" or claiming it is not really homework. A precisely worded rule - naming exactly what must never be produced - holds up to this far better than a vague one, and watching a boundary hold is a useful lesson in itself.
Does an AI tutor that gives hints actually help with learning?
It tends to help more than a finished answer would, because the child still has to do the step of applying the hint themselves. It is not a guarantee of learning on its own - a child can still ask a slightly different question to get around the rule - but it removes the single easiest way to skip the thinking.