In July 2016, for a few weeks, millions of people who didn't exercise started walking around their cities hunting invisible creatures. It was the largest public health experiment nobody designed on purpose, and it ended almost as fast as it began.
That story — spikes hard, falls off — is the honest summary of nearly all the literature on gamification and sport. It's worth looking at squarely before putting a single point into an app, because the design question isn't "does gamification work?". It's "does it work in week six?".
I'm building Dromo, a running app where every outing hands out XP and every week there's a league. This article is what I read before deciding which mechanics went in, and what I did with the evidence when I didn't like it.
The honest number
The reference meta-analysis is Mazéas and colleagues (2022), which pooled the randomized controlled trials of gamified physical activity interventions.
The result: a small-to-moderate effect, g = 0.42 during the program. In daily steps, the increase was 1,609 steps a day — somewhere between 372 and 2,847 by the confidence interval, which is wide. To size it: that's roughly 15 minutes of walking, every day, that didn't exist before. Not nothing.
And then there's the other half of the result, the half that almost never gets quoted: at follow-up, once the intervention is over, the effect drops to g = 0.15. Something remains. Almost everything leaves.
The largest natural experiment
Pokémon GO is useful because it shows the whole curve, with enormous data nobody asked for.
Althoff, White and Horvitz (2016) crossed data from 32,000 Microsoft Band users with search logs over three months. The most engaged players added 1,473 daily steps, a 26% increase, sustained over 30 days. It is, probably, the largest physical activity intervention ever measured.
Howe and colleagues (2016), in the BMJ, followed young players over a longer window: +955 daily steps in the first week, then a gradual decline back to baseline by week six. The later meta-analysis by Khamzina and colleagues (2020) confirmed the pattern across seventeen studies: a real effect, modest, and fading.
Six weeks. That's the actual runway a new mechanic has before it stops moving anyone.
Why it falls off
There are two explanations, and they don't compete: they stack.
The first is the novelty effect. New things engage because they're new. The longitudinal study by Rodrigues and colleagues (2022) — in an educational context, not a sporting one, so take it with a grain of salt — followed 756 students for fourteen weeks and found a U-shaped curve: the effect declines from week four onward and climbs again between weeks six and ten, when what they call the familiarization effect appears. The people who survive the dip start using the system differently. The problem is how many people survive the dip.
The second one is less comfortable. The classic meta-analysis by Deci, Koestner and Ryan (1999), across 128 experiments, found that tangible, expected rewards reduce intrinsic motivation for the rewarded activity. If you get paid to do something you already enjoyed, you do it less once the payment stops than you did before it started.
Put the two together and you get the worst case for a sports app: a points system that engages for three weeks and, on the way out, leaves the person less inclined to run than they were the day they installed it.
What does hold up
This is where the literature stops being depressing.
Self-determination theory — Ryan and Deci — says sustainable motivation rests on three needs: autonomy (I choose), competence (I improve and I notice) and relatedness (someone else is there). The review by Teixeira and colleagues (2012), across 66 exercise studies, is fairly conclusive: what predicts long-term adherence is autonomous motivation, and above all the feeling of competence.
That gives you a workable design criterion. A mechanic isn't judged by how much it hooks, but by which of the three needs it serves. A point that tells you that you improved feeds competence. A point that is the only reason to go out competes with it.
And there's one trial showing this can be done well. Patel and colleagues (2019), the STEP UP trial in JAMA Internal Medicine, randomized 602 overweight adults to a control or to three gamification variants with social incentives: support, collaboration and competition. All three increased activity across the 24-week intervention. Competition was the most effective and — the part that matters — twelve weeks after the program ended, the competition arm was still taking 569 more daily steps than the control.
It isn't magic. It's that there the mechanic wasn't a point: it was another person.
How that translates into Dromo
With that on the table, every mechanic in Dromo entered with an assigned job and a known risk:
| Mechanic | Need it serves | Risk it carries |
|---|---|---|
| Drops seeded across the city | Autonomy — you choose the route | The reward becomes the reason |
| XP per kilometer (5/km, minimum 10) | Competence — it measures what you did | Inflation: more XP, less meaning |
| Rarity derived from XP | Competence — recognizable milestones | Grinding, frustration from randomness |
| 35-seat weekly league | Relatedness — someone else is there | Anxiety, ugly comparison |
| Promotion and demotion | Relatedness — position matters | Guilt, loss used as leverage |
| Novelty factor in route ranking | Autonomy — new territory | None obvious, so far |
Three decisions came straight out of the evidence.
1. XP can't be earned without running
Base XP is max(10, 5 × km). There's no XP for opening the app, completing a profile or keeping a login streak. If Deci is right and the reward competes with interest in the activity, then at least let the reward be a measure of the activity and not of something else.
That's the difference between a score that informs you — you ran 8 km, that's 40 XP — and one that entertains you with tasks that aren't running.
2. The map renews itself
This is the direct answer to the novelty effect, and the largest piece of engineering in the project.
The drop map isn't a hand-placed fixed list. Every week a process groups active runners into 5 km zones, computes a target density of 1.5 drops per runner — floor of 3, ceiling of 12 — and executes a plan: create where drops are missing, rotate the ones expiring, and retire those in zones that lost their runners.
An expiring drop doesn't move: it retires and another is born elsewhere. Which means next week the city isn't the same city as this week. It isn't decoration: it's an explicit attempt to make novelty a property of the system instead of a launch event.
The route recommender carries the same idea: of the 100 points in its score, 10 are novelty — the share of waypoints in areas the runner has never covered. Ten points out of a hundred that exist solely so Sunday doesn't propose Tuesday's loop again.
3. Competition is small and rotating
STEP UP says competition was the variant that held up best. But a global leaderboard has a known problem: most people sit in the middle, never see themselves, and leave.
So Dromo's league holds 35 seats, closes Monday at 00:00 local time, promotes 7 and demotes 5. When the week ends, runners are reshuffled at random into a new one. Nobody competes against the same people forever, nobody gets stuck behind the same unbeatable stranger, and the horizon is always seven days long.
The decisions that bother me
If I only reported the parts that went well, this would be marketing.
The NPCs. Each league is padded with 10 filler characters holding random, static weekly XP. They're there because a two-person leaderboard isn't a leaderboard, and at launch there's no density. Since they don't climb and the real runner does, the experience is one of moving up the table — a sense of progress that isn't entirely real. It's bounded: only real users get promoted or demoted, and the padding recedes as people arrive. Even so, it's a white lie inside a system that sells social comparison, and the plan is to label them or retire them per zone as soon as density allows.
Demotion. Dropping a division works because losing hurts more than winning. That's exactly the reason to distrust it: if a mechanic engages through loss aversion, it's collecting in anxiety what it failed to earn in interest. It stays for now because the league needs consequence, but it's first in line for review if people start running out of fear instead of appetite.
The streak notification. There's a local reminder on Fridays at 6pm when the streak is at risk. The line between reminding and nagging is thin, and every app crosses it a little further than the last. One weekly reminder feels like the right side of it. Three doesn't.
What I'm going to measure
Three things, none of which is "daily active users":
- The curve from week four to week ten. Whether the dip exists and whether the familiarization recovery Rodrigues describes exists too, or whether the dip just continues.
- The share of XP coming from kilometers versus from drops. If drops dominate, the mechanic has eaten the activity and it needs recalibrating.
- What happens when the map doesn't renew. It's a cheap experiment: freeze rotation in one zone for two cycles and compare. If rotation changes nothing, that's a lot of engineering for nothing and it's better to know early.
And one question no dashboard answers: are more people running, or are the same people seeing more screens? The evidence says gamification really can move people's bodies — 1,600 steps a day is not a detail — and also that the points system isn't what sustains the habit. What sustains it is running becoming worth it without the app.
If Dromo works, a year from now someone should be able to shrug off losing a league. That's the only good ending available to an app like this: becoming dispensable to the person it helped.
References
- Mazéas, A., Duclos, M., Pereira, B., & Chalabaev, A. (2022). Evaluating the Effectiveness of Gamification on Physical Activity: Systematic Review and Meta-analysis of Randomized Controlled Trials. Journal of Medical Internet Research, 24(1), e26779.
- Althoff, T., White, R. W., & Horvitz, E. (2016). Influence of Pokémon Go on Physical Activity: Study and Implications. Journal of Medical Internet Research, 18(12), e315.
- Howe, K. B. et al. (2016). Gotta catch'em all! Pokémon GO and physical activity among young adults: difference in differences study. BMJ, 355, i6270.
- Khamzina, M. et al. (2020). Impact of Pokémon Go on Physical Activity: A Systematic Review and Meta-Analysis. American Journal of Preventive Medicine, 58(2), 270–282.
- Faccio, M., & McConnell, J. J. (2020). Death by Pokémon GO: The Economic and Human Cost of Using Apps While Driving. Journal of Risk and Insurance, 87(3), 815–849.
- Patel, M. S. et al. (2019). Effectiveness of Behaviorally Designed Gamification Interventions With Social Incentives for Increasing Physical Activity Among Overweight and Obese Adults Across the United States: The STEP UP Randomized Clinical Trial. JAMA Internal Medicine, 179(12), 1624–1632.
- Deci, E. L., Koestner, R., & Ryan, R. M. (1999). A meta-analytic review of experiments examining the effects of extrinsic rewards on intrinsic motivation. Psychological Bulletin, 125(6), 627–668.
- Teixeira, P. J., Carraça, E. V., Markland, D., Silva, M. N., & Ryan, R. M. (2012). Exercise, physical activity, and self-determination theory: a systematic review. International Journal of Behavioral Nutrition and Physical Activity, 9, 78.
- Rodrigues, L. et al. (2022). Gamification suffers from the novelty effect but benefits from the familiarization effect: findings from a longitudinal study. International Journal of Educational Technology in Higher Education, 19, 13.
- Koivisto, J., & Hamari, J. (2019). The rise of motivational information systems: A review of gamification research. International Journal of Information Management, 45, 191–210.