Why kids quit learning games in a minute, and what the good ones do
Our own numbers on how fast kids leave learning games, what Blooket, Gimkit and Roblox do differently, how to get the level right, and the 12 games we're retiring.
Most learning games lose children in under a minute because they ask a question before they show a game, at the wrong level, in a style an 8-year-old finds babyish. The ones kids ask for do the opposite: every right answer buys something they can see (fuel, a blast, a tower, a power), the fun starts within ten seconds, and the questions sit where most of them are wins. Peabody (peabody.kids) learned this the hard way with its own games, and this post is the honest version: our numbers, what the research and the games kids love say, and what we're changing. Peabody is free through the end of 2026.
Our own numbers
We count how long each game is played, with nothing about who played it. Over the three weeks to 27 October 2026, across 482 plays of the 38 games in our Arcade that ask a question:
| What we measured | Signed-in kids | Visitors who aren't signed in |
|---|---|---|
| Half of all plays were shorter than | 91 seconds | 49 seconds |
| Plays that ended inside 30 seconds | 19% | 33% |
The split by game was sharper than the split by child. Six games built as real games first (a pet-hatching world, a battle royale, an obstacle course, a tower defence game, a dungeon crawler and an aiming game) took 71% of all minutes played, and between 16% and 35% of their plays went past five minutes. The games built to teach first did much worse. Our phonics mining game and our potion-shop reading game are the two that visitors land on most from search, and their plays lasted a median 56 and 35 seconds. A farm game's median was 19 seconds. Four games were never played at all once anyone started counting.
So the children aren't the problem. Children play our games for a long time when the game is a real game.
Four ways to put a question in a game
Researchers and game makers sort learning games by where the question sits. The difference matters more than the graphics.
| Where the question sits | What it feels like | Who does it | What we've seen |
|---|---|---|---|
| A gate: answer to pass | A toll in front of the fun | Prodigy, Boddle, most of our older games | The weakest for fun. Prodigy's maths is often described as a tax on a good game |
| Currency or fuel: answer to earn what the game spends | Every answer buys something you can see | Blooket, Gimkit | The ones kids beg teachers for. Gimkit lets players choose when to go and answer for more fuel |
| The move itself: the answer is how you play | The maths is the weapon or the jump | DragonBox, Habgood and Ainsworth's Zombie Division | In Habgood's study, children chose to keep playing the version where maths was the weapon about seven times longer |
| A level-up pick: play fills a bar, then one question, then you choose a power | A reward you look forward to | Blooket's Monster Brawl | Keeps an action game moving, with a question every minute or so |
Most of our learning games used gates. A child got a question, then got to play a bit, then got another question. That's the shape both kids and parents complain about, and it's the one we're moving away from. We wrote more about why the learning has to be the game.
The first ten seconds decide it
Web game portals study this closely because their whole business is the first minute. Poki's guidance to game makers asks one question: can the player start having fun within ten seconds (Defold × Poki)?
Web game portals study this closely because their whole business is the first minute. Poki's guidance to game makers asks one question: can the player start having fun within ten seconds (Defold × Poki)? CrazyGames says its best titles keep most players past the first minute (CrazyGames). One game on Poki went from half of its visitors starting to play to 72% just by cutting its loading time from about 30 seconds to under 4 (Poki).
Count what stood between a child and our games' first fun. A card asking them to type their name. In some games, a menu. In the youngest children's games, five seconds of waiting before the first question. Then, for every visitor who hadn't told us their grade, a kindergarten question, whatever their age. A 4th grader who taps a game and gets "How many dots?" is gone in ten seconds, and they're right to go.
Getting the level right
A game made to teach has to fit the child, and nearly everything fits badly at first. Three things help.
A game made to teach has to fit the child, and nearly everything fits badly at first. Three things help.
- Ask the grade, but treat it as a first guess. Nine in ten parents believe their child is at or above grade level, and only about a third are, so a grade alone puts many children too high. One adaptive program, MobyMax, starts children one grade below and moves them from there (MobyMax).
- Find the level inside the first round, not over weeks. Start a step under, make the first three questions easy wins, move up after three right in a row, move down after two misses, and settle within about a dozen questions. A child should never notice they were placed.
- Keep most answers right. About 75 to 85% right is the band where children keep going and still learn (why most answers should be wins). Big online experiments with tens of thousands of players found that easier versions of a game are played longer but teach more slowly (Lomas and colleagues, 2013). So we treat these as two dials: the questions stay in the 75–85% band, and the child can choose how hard the game is ("Hard road, double loot").
Not everything has to be ladybugs
Children don't stay in cute games for long. Among Australian children, the share who play Roblox jumps from about four in ten at ages 6–7 to two in three at 8–9.
Children don't stay in cute games for long. Among Australian children, the share who play Roblox jumps from about four in ten at ages 6–7 to two in three at 8–9. The hits of 2025 were short, simple loops with stakes: a garden that grows while you're away, a base you defend, a monster outside the campfire light. Kids want chases, bosses, rare things to collect, cars, and things that smash.
None of that needs anything a parent would worry about. The best family games have shown how for years. Lego games break enemies into bricks, Plants vs Zombies pelts zombies with vegetables, and in Roblox's Jailbreak people get arrested, not hurt. Rating boards treat cartoonish, bloodless action as fine for ages 10 and up (ESRB on Plants vs Zombies, Roblox's content maturity labels). These are the rules we're writing into every game:
- Smash things, never people. Crates, windows, walls, robot cars and towers are fair game.
- The enemies are zombies, robots, slime, ghosts and monsters. People get "busted" or "bonked". Nobody dies, and there's no blood and no realistic guns.
- Tension is allowed, scares aren't. Night, chases and boss music are fine. Nothing jumps out at the youngest players.
- Losing costs time or coins, never a life for a wrong answer, and a retry takes two seconds.
- No chat with strangers, ever. Children outside a family see each other only by a made-up pen name, on a scoreboard.
Where children's games really go wrong is usually not the action. It's open chat with strangers, pressure to trade, and paid randomness. The US Federal Trade Commission's $20 million settlement with the maker of Genshin Impact over its loot boxes is the line to stay far away from (FTC). At Peabody nothing is for sale, and every egg or chest shows its odds.
What we're changing at Peabody
We're not making any new learning games. We're making the ones we have much better, and letting the weakest go.
We're not making any new learning games. We're making the ones we have much better, and letting the weakest go.
- Twelve games are retired. Together they drew about 5% of plays. Six of them were one game wearing six costumes: tap one of three lit squares, then watch the board play itself. They move to the Arcade's Trash shelf, where they can still be played. Their two good ideas move into games that stay: the words you just read in a library story, and a number line you jump along.
- One card, one tap. Before the first question, a game asks your grade (that part can't be skipped) and offers a name. You can type one or keep the cool pen name it already picked for you, like "Laser Kraken". The game waits while you choose.
- The level is found in the first minute, the way described above. Every end screen has Too easy and Too hard buttons, and one tap fixes it.
- Every answer powers something. Drill fuel that cracks rock, a blaster charge, a nitro boost, a power you pick, bricks for a tower. Answering fast never pays more, so nobody is rushed into guessing.
- Real games around the questions. Every learning game gets a world map, a boss, an upgrade shop, rare things to collect with the odds shown, something that grows while you're away, and a story in which you're the hero, called by your own name or pen name. Your pet follows you from game to game.
- Games that look their age. Bright and spoken for the youngest, louder and faster for the middle grades, and darker and cooler from 6th grade, with three big answer boxes for anyone who needs them.
- A car game. One of our never-played games becomes a car game: drive through the city, smash through the billboard with the right answer for nitro, and outrun the robot cops.
We'll start with the games visitors land on most: phonics mining, the potion shop, the space math game, the dinosaur runner and the stories where you pick what happens. Then we'll measure again. If a rebuilt game still loses children inside a minute, it goes to the Trash too.
How to pick a learning game for your child
The same tests work on any learning game, ours included. Give it two minutes with your child and look for these.
The same tests work on any learning game, ours included. Give it two minutes with your child and look for these.
- Did they get to play within ten seconds? If a quiz, a form or a long tutorial comes first, expect them to leave.
- Does a right answer make something happen in the game? A gate that opens is the weakest. Fuel, a blast or a tower to place is better.
- Did it ask their level or find it? Watch the first few questions. Both "too easy" and "too hard" are reasons children quit.
- Are most answers right? Roughly four in five is about right.
- Is anyone rushing them? Speed bonuses on things a child is still learning reward guessing.
- Can strangers talk to them? Look for pen names and no free chat.
- Is anything for sale, or locked behind a timer? Countdowns and paid boxes are signs a game is built to keep children playing, not learning.
If your child would rather read than play, our library has hundreds of free stories at every level, and do educational games work? covers what the evidence says about learning from games at all.
Checked against how Peabody works on 27 October 2026. Written with help from AI. Peabody is a reading app, not a reading assessment: if you're worried about your child's reading, talk to their teacher. Questions, or something wrong? Write to hello@peabody.kids.