Do points and streaks make a child want to read?

In 1999 three researchers pooled 128 experiments on what happens to a person's willingness to keep doing something after they have been paid to do it [1]. The measure is simple and hard to fake: leave the person alone with a free choice afterward, and see whether they go back to the activity.

Across 92 experiments, a reward promised in advance for doing an activity reduced that free choice by an effect size of 0.36[1]. Split by what the reward was paid for, the pattern was consistent: 0.40 lower for rewards paid simply for doing the activity, 0.36 for rewards paid for finishing it, 0.28 for rewards paid for doing it well [1]. All three point the same way, and none of them is small.

Children came off worse than adults. For rewards paid for doing the activity, the effect among children was 0.46 lower across 43 studies, against 0.21 for college students [1].

By the rough convention used across this field, an effect size of 0.2 counts as small and 0.5 as moderate, so none of these are rounding errors. If effect sizes are new to you, they are worked through properly, with the phonics numbers as the example, on our science of reading page.

The clearest picture of this in young children is fifty years old. Researchers watched a preschool class in California through a one-way mirror, picked out 51 children aged between 40 and 64 months, so roughly three and a half to five, who already chose to draw with felt tips in their free time, and split them into three groups [2].

One group was promised a certificate with a gold seal and a ribbon if they would draw. One group drew and then got the same certificate as a surprise. One group drew and got nothing. Then the researchers put the felt tips back into the classroom, waited one to two weeks, and watched again [2].

The difference between the three groups was statistically reliable [2]. What separates the two ends of it is not the certificate. Both the first and third groups got exactly the same one. What separated them was the deal: draw, and you will get this.

Fifty-one children in one preschool in 1973 is a small, old study, and one study never settles anything. It matters because the meta-analysis found the same shape across more than a hundred others [1][2].

Second limit: praise is not the easy answer either. Positive feedback did raise free choice overall, by 0.33 [1]. But in the seven studies of children, the effect was 0.11 with a confidence interval running from below zero to above it[1]. In plain terms, on this evidence, praising a child instead of paying them is not a proven substitute. It is just less likely to do harm.

Third limit: unexpected rewards did nothing measurable either way. Nine experiments, an effect of 0.01 [1]. A surprise costs the child nothing, because there was never a deal to notice.

It would be convenient for us to stop here. So here is the strongest published argument against the way we have built this.

A 2020 review pooled the trials of gamified learning: points, badges, levels and the rest, put in front of real learners. The effects were positive. Learning outcomes came out 0.49 ahead across 19 comparisons, motivation 0.36 ahead across 16, and what learners actually did 0.25 ahead across 9 [3]. Effects in school settings were larger than in higher education[3].

Three things stop that being the end of the argument. The learning-outcome effect held up when only the more rigorous studies were counted, while the effects on motivation and on what learners did were less stable [3]. Gamification in these studies is a bundle of design elements, so nothing in the review isolates the one mechanic parents actually ask about, which is a streak that breaks. And the review is measuring the short term, mostly inside a course, not what a child chooses to do with a book years later.

Taken together: points can help someone get through the work in front of them, and promised rewards can cost you their appetite for it afterward. Both are true. Anyone quoting only one of them is selling something.

First: points that cannot be spent are the weakest known design. A review of 24 classroom token-reward systems for young children, every study screened for quality, found that every effective system worked the same way: children saved tokens and traded them for something they chose [4]. In gamified courses, "the points count for nothing" shows up in learners' own words as a reason they stopped caring [6]. Until August 2026 our coins were exactly that: a number that went up. That is why the trade screen now exists.

Second: choice is the most reliable lever in the newer evidence. Letting learners choose, including choosing how their own character looks, carried the largest dependable effect on motivation, 0.64 [5]. Rewards tied to real performance helped, while schemes that mostly paid people for showing up ranged from useless to actively harmful, the worst single result in the whole set [5]. Our coins already pay mostly for mastery and clean runs, the daily bonus stays a small topping, and the spending side is a fixed set of characters the child picks from.

Third: expect less than the headlines. In the one analysis that split by age, the benefit was smallest for children of primary age, 0.31 against 1.02 for secondary students, with the low end of the estimate nearly touching zero [7]. And the effects fade: gains peak in the first months and drop to about zero past a school term, in three separate analyses [5][6]. So coins stay seasoning, the lessons have to carry the motivation, and a parent should expect the trade screen to make practice a bit more fun, not to make a child a better reader. Rewards you could spend made essentially no measured difference to learning itself [6].

Fourth: the one clearly dangerous design is chance. Spending on chance-based reward boxes correlates with problem-gambling symptoms across 15 studies of more than 16,000 people, the link was stronger in adolescents than in adults, and it held for boxes whose contents were only cosmetic [8]. Nobody has studied children as young as ours, which argues for more caution, not less. That is why every number in Plimmery is fixed and public: no mystery boxes, no spinning wheels, nothing on a timer, and coins never touch real money in either direction.

The honest limit over all of it: No study has tested an earned coin with a trade screen, in a reading app, at ages 4 to 9. The findings above are the nearest evidence, not proof. We treat this as an experiment on our own product: we watch whether children keep reading over months, not launch weeks, and we will change or remove the trade screen if the answer is no.

Three of the seven mechanics above pay a child for doing well rather than for taking part, and that is the type the meta-analysis found undermining at 0.28 [1]. We keep them small, we never remove them, and we never rank one child against another. We are not going to claim the research endorses that compromise. It is a judgment, and we would rather you saw it than found it.

Since August 2026 the coins can be traded, for one thing only: cosmetic mascot characters, at fixed public amounts, chosen by the child. The trade is the exchange step the token-economy literature identifies as the mechanism that makes tokens mean anything [4]. What we refused matters as much: no chance in what a trade returns, no timers, no rotating stock, nothing a coin can touch in the lessons themselves, and no bridge to real money in either direction [8].

What we did refuse is anything with a loss in it. No lives to lose, no coins clawed back, no red screen for a missed day, and one missed day a week bridged automatically so an ordinary week of family life does not wipe the number. The evidence does not prove that choice right. It points that way, which is the most anyone can honestly say.

Learning apps with a streak that breaks are easy to find. We went looking for a controlled trial that randomly assigned children to a breakable streak or to no streak and then measured what happened to their interest, and we could not find one.

That is a strange gap. The mechanic is everywhere, the underlying question is fifty years old, and the trial would not be hard to run.

Until somebody runs it, treat every confident claim about streaks as an opinion wearing a lab coat. Including ours. What we can tell you is exactly what the app does, which is on this page, and exactly what the evidence covers, which is on this page too.