Story word problems the parent reads first

The problem

A worksheet of bare numbers is easy to avoid

One of the eight problems on Reward Study's first screen is "Homework is a daily fight," and a chunk of that fight is not about the maths being hard, it is about the maths being flat. Twenty subtraction problems in a row, numbers with nothing around them, ask a child to sit still and grind through them for their own sake. Some children do that without complaint. Plenty do not, and the daily argument is less about the arithmetic than about getting a nine-year-old to start.

Turning the same problem into a story does not make the subtraction easier. It changes what a child is looking at when they sit down: a sentence about a football match losing players to injury reads as a small story with a question at the end, not a wall of digits, and a story is the kind of thing a child will actually start reading.

The instinct to fix this by promising a reward for finishing does not really address it either: a credit at the end of the set does not change how the set feels while a child is in the middle of it. Reward Study already prices every task in credits, so the credit was never the missing piece. What was missing was a way to make the twenty minutes themselves less flat, without touching the actual maths underneath.

What the app does

The numbers come first, the story wraps around them

The mechanism runs in two separate steps, on purpose. A procedural generator produces the actual problem: the numbers, the operation and the exact correct answer, the same way it does for every other math set in the app. Only after that does an on-device Foundation Model, Apple's own on-device AI, take that finished problem and wrap it in a short story, a scoreline, a train timetable, a bag of apples, without changing a single number or the answer underneath. If the story generation fails for any reason, the plain arithmetic problem is what the child sees, never a broken sentence standing in for a question.

The themes

Five presets, a wildcard, and a theme a family picks itself

Five themes are built in: Football, Trains, Apples, Animals and Space. Alongside them sit "Surprise me," which lets the generator pick a theme on its own, and a free-text custom theme a parent can type in directly, a dinosaur dig or a bakery or whatever a particular child is currently obsessed with. Themes are localized along with the rest of the app: a story generated in German is themed around football using the German word for it, not the English one, the same way every other piece of app text follows the device's own language.

A theme sticks for the length of a study set, not just one problem, so a set generated with the Space theme reads like installments of the same mission rather than five unrelated sentences that happen to mention rockets. A child doing five Space-themed subtraction problems in a row is reading a short story broken into five parts, one problem at a time.

The guardrails

What stops a generated story from going wrong

Two checks sit between the generator and the child. First, a story that comes back without a question mark in it is rejected outright and the plain sum is shown instead. No question mark means no actual word problem, whatever the sentence says, and the fallback is the safe number, not a guess. Second, and just as deliberately, a parent reads every generated set before a child ever sees it. An empty set, one where nothing has been reviewed yet, never counts as reviewed on its own. Nothing themed reaches a child's screen without a parent's own read of it first.

Both checks exist because a generated sentence is not guaranteed to be a sentence a nine-year-old should see unreviewed. Rejecting anything without a question mark catches the cases where the story wrapping did not actually produce a question. The parent review catches everything else, which is why the review step is not optional and an unreviewed set is never treated as ready.

Where it works and where it does not

Twelve of the app's 22 topics, not the three sciences

The feature works on 12 of the app's 22 topics, the arithmetic and pre-algebra style topics where a story genuinely fits around a number. It does not run on the three sciences: Physics, Chemistry and Biology work by chapter instead, and it does not run on topics built around a figure or a multiple-choice answer, where a story would have nothing to attach to. It also needs an Apple Intelligence capable device on iOS 26 or later; on anything else, the toggle for it is simply not shown, rather than shown and failing.

The other ten of the 22 topics are excluded for a plain mechanical reason rather than a quality one: a figure-based question about a triangle's angles, or a multiple-choice element symbol question, does not have a natural sentence a story could replace it with.

On the device, and nowhere else

Nothing generated here is sent anywhere

The generation, the story wrapping and the parent's review all happen on the device the app is running on. No prompt, no generated answer and no child's name is ever sent to a server, because there is no server in this picture at all, the same posture covered in full in how the app handles data generally.

A subscription is still what keeps new sets generating: Story word problems is one of the five Pro benefits listed for 2.0, alongside as many children as needed and two-household support. See the full Pro list before turning it on for a whole household.