Child-safe rewards: no loot boxes, no timers, nothing lost
What research and regulators say about rewards in kids' games, from loss aversion at age four to the FTC's Epic and Genshin orders, and the rules Peabody's games follow.
Children feel losses strongly from about age four, rewards tied tightly to performance can crowd out a child's own interest, randomized rewards with money value are now restricted for under-16s in the United States, and the pediatricians' 2026 position is that platform design, not hours, is what to watch. A reward system for a children's learning game should therefore take nothing away, pay for effort, keep any learning bonus small, and never roll dice. Peabody (peabody.kids) is a free AI-driven adaptive learning platform for kids with about twenty browser games, free through the end of 2026. This post lays out what the research and the regulators say, then Peabody's own rulebook.
Loss hurts early
The endowment effect, valuing something more once you hold it, is the behavioral fingerprint of loss aversion, and it arrives early. Harbaugh, Krause and Vesterlund (2001, Economics Letters; https://econweb.ucsd.edu/~jandreon/Econ264/papers/Harbaugh%20EconLet%202001.pdf) found it already present in five- to ten-year-olds and found that it does not shrink with age. A 2014 PLOS ONE study (https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0109520) found 78% of four- to six-year-olds showed it on the first trial, and that physical possession mattered more than formal ownership. A 2013 study of decision-making under risk (https://pmc.ncbi.nlm.nih.gov/articles/PMC3548654/) detected loss aversion in young children as well.
Whether the endowment effect is loss aversion by another name is contested, and no direct experiment on loss-framed rewards in children's learning games was found. The safe inference is still safe: children feel losses strongly and early, so a game that takes things away (a pet dies, a streak breaks, coins are deducted, a life is lost) is spending the child's most expensive emotion to keep them playing.
The same team's field evidence is about stopping. Hiniker and colleagues (CHI 2016, University of Washington; https://www.washington.edu/news/2016/05/05/two-minute-warnings-make-kids-screen-time-tantrums-worse/) studied families with one- to five-year-olds and found that transitions off screens went better at natural stopping points, that autoplay and suggested-next were the main triggers of fights, and that two-minute warnings made tantrums worse.
Pay for effort, not for being right
O'Rourke, Haimovitz, Ballweber, Dweck and Popovic (CHI 2014; https://dl.acm.org/doi/10.1145/2556288.2557157) randomized about 15,000 children playing a fractions game called Refraction on BrainPOP to a version that awarded "brain points" for effort, use of strategy and incremental progress rather than for correct answers. The brain-points group played longer, used more strategies and persisted more after challenge, with the effect concentrated in low-performing students. The 2016 follow-up with about 25,000 children (https://learningatscale.acm.org/las2016/uploads/ORourke_BrainPoints.pdf) added a twist: growth-mindset animations and lectures caused many players to quit, while the points structure kept driving persistence. The incentive structure worked; the messaging did not. Both studies measured persistence and time played, not learning.
When rewards crowd out interest
Deci, Koestner and Ryan (1999, Psychological Bulletin; https://depts.washington.edu/techdocs/papers/deciExtrinsicRewardsAndIntrinsicMotivation99.pdf) pooled 128 experiments. Expected tangible rewards reduced free-choice intrinsic motivation (d = minus 0.40 for engagement-contingent, minus 0.36 for completion-contingent, minus 0.28 for performance-contingent rewards), and tangible rewards were more damaging for children than for college students. Positive verbal feedback did the opposite, raising free-choice behavior by d = 0.33. Cameron and Pierce disputed the size of the effect in a long exchange; the best-replicated cell, expected tangible rewards tied to the task in children, is the one that matters here.
Cerasoli, Nicklin and Ford (2014, Psychological Bulletin; https://selfdeterminationtheory.org/wp-content/uploads/2017/06/2014_Cerasoli_Intrinsic.pdf) pooled 183 studies and 212,468 people across school, work and sport. Intrinsic motivation predicted performance whether or not incentives were present, incentives predicted quantity better and intrinsic motivation predicted quality, and when incentives were directly and saliently tied to performance, intrinsic motivation mattered less. That last line is the crowding-out effect, and it is the reason a learning bonus should be small and informational rather than a rate a child can compute.
Duolingo's own streak experiments (https://blog.duolingo.com/improving-the-streak/, 2020) are the practical footnote. The controlled gains from streak features were single digits: day-14 retention up 3.3%, daily users up 1%. The big multipliers quoted for streaks are selection effects. What moved behavior most was lowering the bar to "showed up today", which a children's product can copy as a gentle welcome back without the broken-chain framing.
Loot boxes and the regulators
Zendle and Cairns (2018, with a 2019 replication in PLOS ONE; https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0213194) found loot-box spending linked to problem-gambling severity in adult gamers (eta squared of 0.054 and 0.051). Zendle, Meyer and Over (2019, Royal Society Open Science; https://royalsocietypublishing.org/rsos/article/6/6/190049/94826/Adolescents-and-loot-boxes-links-with-problem) found the link more than twice as strong in 1,155 sixteen- to eighteen-year-olds (eta squared 0.120). A 2020 paper found the link held regardless of which features the boxes had. All of this is correlational, and the direction is not settled; the policy line no longer depends on it.
| Action | Year | What it did |
|---|---|---|
| UK Age Appropriate Design Code (Children's Code), ICO | In force 2021 | Fifteen standards: high privacy by default, profiling off by default, no nudge techniques that push children toward weaker privacy, parental controls visible to the child; the ICO's toolkit adds keeping loot boxes to a minimum with odds disclosed |
| FTC staff report, Bringing Dark Patterns to Light | 2022 | Names children's gaming apps that disguise in-app purchases as gameplay |
| FTC v. Epic Games (Fortnite) | December 2022 | $275 million penalty under COPPA plus a $245 million refund order for dark patterns, $520 million in all; voice and text chat off by default for children and teens; express consent before billing |
| FTC v. Cognosphere (Genshin Impact) | January 2025 | $20 million; no loot-box sales to under-16s without parental consent; odds disclosed; anything buyable with virtual currency must be buyable in dollars |
| AAP policy statement, Digital Ecosystems, Children, and Adolescents | February 2026 | Drops hour caps for school-age children; frames engagement-driven platform design as the harm; calls for design standards |
The Epic order is at https://www.ftc.gov/news-events/news/press-releases/2022/12/fortnite-video-game-maker-epic-games-pay-more-half-billion-dollars-over-ftc-allegations, the Genshin order at https://www.ftc.gov/news-events/news/press-releases/2025/01/genshin-impact-game-developer-will-be-banned-selling-lootboxes-teens-under-16-without-parental, the Children's Code at https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/childrens-information/childrens-code-guidance-and-resources/age-appropriate-design-a-code-of-practice-for-online-services/ and the AAP statement at https://publications.aap.org/pediatrics/article/157/2/e2025075320/206129/Digital-Ecosystems-Children-and-Adolescents-Policy.
The vocabulary for all of this comes from Zagal, Bjork and Lewis (2013, Foundations of Digital Games; https://www.semanticscholar.org/paper/Dark-patterns-in-the-design-of-games-Zagal-Bj%C3%B6rk/19a241378b06d868eb5f6b76027172c3aaca86f4), who defined dark game design patterns as ones used intentionally to cause experiences against the player's interests and sorted them into three families: temporal (grinding, playing by appointment, daily-login pressure), monetary (pay to skip, monetized rivalries) and social capital (exploiting the player's friends). A 2023 study of the five most popular free App Store games for ages zero to five found all three families present, and the 5Rights Foundation's Disrupted Childhood report (2018, updated 2023) catalogs the same features aimed at children: endless scroll, popularity metrics, gambling-like features, constant nudges.
The warning for learning games specifically is Prodigy. In February 2021 Fairplay and 21 other groups complained to the FTC (https://fairplayforkids.org/feb-19-2021-advocates-to-ftc-prodigy-math-game-preys-on-kids-and-families/) that the free-at-school math game showed 16 membership ads in 19 minutes against four math problems at home, and that its efficacy claims were unsubstantiated. No public FTC action followed, but school districts de-listed it. The lesson is that a learning product can be judged on its selling as much as its teaching.
Peabody's rulebook
| Rule | What it means in the games | Where the evidence points |
|---|---|---|
| Nothing dies | No pet, creature or character is lost to a wrong answer or a missed day | Harbaugh 2001; Hiniker 2016 |
| Nothing lost | Wrong answers cost tempo, never a life; a miss pays nothing and takes nothing; passed rungs are saved per child and apply in every game | Brain Points 2014 |
| No leaderboards | No scores shown to children, no percentiles, no comparison with other children | Zagal 2013 social-capital patterns |
| No countdown timers on new items | Speed is measured, never demanded; a new item never has a clock | Deci 1999 on controlling rewards |
| Coins come from the world | Just play pays from the world itself; the game must play whole with the questions off | Ryan, Rigby and Przybylski 2006 on autonomy |
| Learning modes pay more, a little | The learning bonus exists and is kept modest rather than a rate to grind | Cerasoli 2014 on crowding out |
| Nothing is bought, nothing is random | No shop, no ads, no loot boxes; no loop rewards children at random | Genshin order 2025; Zendle 2019 |
| Parental controls off by default | Time limits, lights-out, learn-first and earn-more-play are there if you want them; learning is never capped; reading, math and making stay open at lights-out | ICO Children's Code |
| Chat off by default | Chat with other kids is off until a kid turns it on from their own profile, and it is tap-only phrases with kids who are here; the one AI chat, typed Talk with Peabody (3rd grade and up in a family account), is separate, short and screened | Epic order on chat defaults |
The questions themselves come from the practice brain, ordinary code: a round is about 85% things the child already knows, a miss comes back three questions later so the round ends on a win, and two misses in the last five stop new items for the round. AI writes the stories; code decides the rest. The full rulebook is in how Peabody picks the next question.
What this doesn't show
- The loot-box evidence is correlational and mostly in adults and adolescents; the regulation is what settles the question for a product aimed at four- to thirteen-year-olds.
- No experiment has tested loss-framed rewards in a children's learning game directly; the inference comes from the endowment-effect and transition studies.
- The practice brain does pay per answer: one point for a first-try right answer, a quarter after a hint or miss, nothing for a wrong one. That is a small, flat rate, and it is still the kind of performance-contingent reward Deci and Cerasoli say to watch. Peabody keeps it modest and has not measured its effect.
- Peabody has not been tested in a trial and makes no claim about outcomes.
What a parent can do
Open any game your child plays and check four things: whether there is a timed chest or a come-back-in-three-hours bonus; whether the "keep playing" button is bigger or brighter than "stop"; whether chat is on by default; and whether anything is sold inside the child's space. The ICO's standards and the Epic order turned on exactly those four. On Peabody, the games are open without an account, nothing is sold, and the library is one tap from any of them. Start free if you want to see the rulebook running.
Questions parents ask
Does Peabody have loot boxes or a shop?
No. Nothing is bought, there are no ads, and no loop in the platform rewards children at random. Coins come from the world and from answering, and they buy nothing with money value.
Why are there no timers? Don't kids need to get fast?
Speed matters for math facts, and Peabody measures it: an answer is "quick" if it comes within about three seconds for adding and two and a half for times tables. But a new item never has a clock, and a slow right answer is still a right answer. A planned Training Center for math-fact speed will time only facts a child already knows.
What about streaks?
Peabody has none. Duolingo's own experiments put the causal gain from streak features in single digits, and the loss framing is the part children feel most. Progress is saved, nothing breaks, and coming back after a week finds everything where it was.
Are the parental controls on by default?
No. They are off until a grown-up turns them on. The exceptions that apply to everyone are chat quiet hours from 8 pm to 7 am and the fact that learning is never capped by any control.
Checked against how Peabody works on 16 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.