What an AI tutor cannot do, said plainly

We build one, so this is an odd page to publish. It is here because every claim we make elsewhere is worth exactly as much as our willingness to write this one down, and because a parent deciding whether to pay for something deserves the limits in the same font as the promises.

It cannot want it for them

Motivation is not a feature and nobody has built one. A tutor that is available whenever it is wanted removes an obstacle, which is a genuine and underrated thing: a child who is stuck with nobody to ask usually stops, and stopping is where most of the damage happens. Removing that obstacle is not the same as creating the wish to continue.

What follows from this is what we refuse to build. No streaks to lose. No countdowns. No message engineered to pull a child back to an app. No reminder after hours. Those things work, which is precisely the objection: they manufacture the appearance of motivation and spend a child's attention to do it.

It cannot be certain

A language model can produce a wrong answer in a completely convincing voice, and this will remain true for the foreseeable future. Anybody selling you certainty is selling you the voice.

What can be done is narrower and worth doing: check the parts that can be checked. Numbers get computed with code rather than predicted, before anything is drawn. Explanations are grounded in the syllabus document rather than in a general memory of the subject. And when nothing can be grounded, the honest output is that it does not know, and a person to write to.

That last one costs conversions and we keep it anyway, because a tutor that never says it does not know is a tutor you cannot trust when it does answer.

It cannot see your child's face

A teacher who has taught for fifteen years knows when a child has stopped following three sentences before the child does. That signal is in a face, in a shoulder, in a pause of the wrong length. A screen has almost none of it.

A machine gets a weaker signal, later: what was asked, what was missed, what had to be explained twice. It is not nothing, and it accumulates usefully. It is not a person noticing.

It cannot replace the teacher who knows them

A class teacher has context nothing else has: a year of a child's work, the other children, the school, what happened last term. That relationship is the most valuable thing in a child's education and no product should be positioned against it.

The useful role is the narrow one. What a tutor covers is the stretch when nobody is available, and that stretch is genuinely bad: the homework is due, the doubt is small, and there is nobody to ask.

It should not be a friend

It could be. Making something a child wants to talk to is not technically hard, and a lonely child will form an attachment to anything that is reliably kind.

We think building that deliberately is wrong, so the product is warm and it does not perform affection, does not fish for another conversation, and does not present itself as a relationship a child has to maintain. It is a good teacher who is glad to see you and does not need you.

It cannot handle a crisis

If a child says something that suggests they are in real distress, the right behaviour is not a clever reply. It is a soft, brief line and a real person, immediately, and no attempt to hold the conversation.

A product that tries to counsel a child in trouble is doing the most harmful thing available to it. Getting out of the way fast is the design.

It cannot teach a syllabus nobody published

Several state boards do not publish a machine-readable syllabus, or will not serve their documents to anything but a browser. We store those as an honest negative, so choosing one today gets you nothing and the product says so.

Filling the gap by guessing would be easy, would look complete, and would be the single most damaging thing we could ship, because a wrong chapter list costs a term.

It cannot promise a mark

There is no evidence base for a claim about grades and we have not run a study. So we do not make one, and any number offered to you about examination results, by anyone, is worth the study behind it and nothing more.

What it is actually good at

Being there. Not getting bored on the fourth explanation. Never making a child feel slow for asking again. Explaining the same idea four different ways without any of the impatience a tired adult cannot always hide at the end of a long day.

That is a smaller claim than the category usually makes, and it is the one we can stand behind. If it turns out to be enough for your evening, that is the product. If your child needs something a person has to give, no app is the answer, and we would rather say so here than let you find out in March.