Gamification: what the meta-analyses say.

Few design fashions have travelled as far on as thin a diet of evidence as gamification did in its first decade. The word was coined around 2010, and the conference keynotes followed within a year. By the time the first systematic review appeared, points and badges had already been bolted onto banking apps, fitness trackers, and most of the world’s learning platforms. The research has spent the years since catching up with the rollout. The catch-up produced something more useful than a verdict: a fairly precise map of which mechanics move which outcomes, under which conditions, for which people. This article is that map.

Ask a product team whether adding a streak counter, a points balance and a leaderboard will make people learn more, and you will get one of two confident answers. The optimist points to Duolingo. The cynic points to every abandoned corporate LMS with a dusty “badges” tab. The literature agrees with neither. The honest answer is conditional — and the conditions are, by now, fairly well mapped.

The first serious synthesis set the tone. Hamari, Koivisto and Sarsa reviewed the empirical studies available by 2014 and reached a verdict that has aged well. Gamification generally produces positive effects. But those effects depend heavily on the context and on the people using it (Hamari, Koivisto & Sarsa, 2014). They also flagged a problem that still haunts the field. Many early studies measured self-reported enjoyment rather than learning, and could not separate the game elements from the novelty of a new system.

A decade later, we have proper meta-analyses — studies that pool the results of many studies — instead of narrative reviews. They sharpen the picture without overturning it.

What the meta-analyses actually found

Meta-analysis matters more in this literature than in most, because the single studies disagree so much. Individual results range from strongly positive to null to negative. Only pooling shows whether the centre of that spread sits above zero. And only the moderator analyses — the breakdowns by condition — show what drives the spread.

The most cited quantitative synthesis in learning contexts is Sailer and Homner’s 2020 meta-analysis in Educational Psychology Review. Pooling controlled comparisons, they report significant but modest effects that differ by outcome type. Cognitive learning outcomes: a small-to-moderate effect (g ≈ 0.49). Motivation outcomes: smaller (g ≈ 0.36). Behaviour outcomes: smaller again (g ≈ 0.25) (Sailer & Homner, 2020). The pattern matters as much as the numbers: gamification’s clearest win is on learning, not — as the folk theory would have it — on motivation.

Two details keep that finding honest. First, the authors re-ran the analysis on only the most rigorous studies. The learning effect held up; the motivation and behaviour effects became less stable (Sailer & Homner, 2020). The result you would most expect gamification to produce — a motivation bump — is the one that wobbles most under scrutiny. Second, a separate meta-analysis focused on education reached a matching conclusion: an overall medium-sized advantage for gamified over non-gamified instruction (g ≈ 0.50). Its authors also documented, from learners’ own accounts, both why people enjoy these systems and why some come to resent them (Bai, Hew & Huang, 2020).

So the field’s average is real and positive. The trouble — and the interesting part — is the spread around it. A wide spread around a modest mean guarantees that many real deployments sit at zero or below.

How the field graded its own homework

Between the early narrative review and the mature meta-analyses sits a document worth knowing about. Koivisto and Hamari’s 2019 review read several hundred empirical studies of the by-then-sprawling field, and it graded the field’s methods as much as its findings (Koivisto & Hamari, 2019). Their census confirmed education and learning as gamification’s main home.

Their sharper contribution was a catalogue of the field’s recurring weaknesses. Studies too short to separate real effects from novelty. Missing control conditions. Self-report used where behaviour could have been measured. And a persistent habit of studying “gamification” as a bundle rather than isolating which element did what.

That catalogue is the right lens for every effect size in this article. It explains why the rigorous-subset analyses matter more than the headline averages. It explains why single-element experiments like Mekler’s carry weight beyond their size. And it explains why anyone reading a glowing case study should ask two questions first: how long did it run, and what was the comparison? A field that knows its own failure modes this precisely is a field worth trusting — about exactly as far as its better studies go.

The element matters more than the label

“Gamification” is not one thing. Treating it as one is the single biggest source of confused findings. The most useful studies stop asking whether gamification works and start asking which element does what.

Mekler, Brühlmann, Tuch and Opwis ran the cleanest test of the classic trio. In a controlled experiment on an image-annotation task, they compared points, levels and leaderboards against a plain control. The result is quietly important. The game elements significantly increased performance — people did more — but did not significantly affect intrinsic motivation or perceived competence (Mekler et al., 2017). In other words, these mechanics act like external incentives that direct effort — not like magic that makes a boring task feel meaningful. As Mekler and colleagues put it, points, levels and leaderboards by themselves neither make nor break intrinsic motivation in a non-game context.

Sailer and colleagues came at it from the other side. They asked which needs each element feeds, using the lens of self-determination theory — psychology’s standard account of what feeds motivation. Badges, leaderboards and performance graphs strengthened the sense of competence and of meaningful goals. Social elements — avatars, a narrative, teammates — did more for the sense of relatedness (Sailer et al., 2017). The lesson is not that one set is better. Different mechanics pull different psychological levers, so the right element depends on which lever the task needs.

Leaderboards are the clearest case of a mechanic whose value depends on why it works. Landers, Bauer and Callan showed that a leaderboard raised task performance about as much as assigning people a specific, difficult goal did — and that the effect ran through goal-setting, hinging on how committed each person was to the goal (Landers, Bauer & Callan, 2017). Read carefully, that is a warning as much as an endorsement. A leaderboard helps to the extent it works as a clear goal people commit to. Strip out the goal and leave only the social comparison, and there is no reason to expect the same lift.

When it backfires: the overjustification risk

Everything so far has treated the worst case as zero — a mechanic that does nothing. The motivation literature contains a worse worst case. It is old enough and large enough that no gamification design should be signed off without checking against it.

The most important warning evidence predates gamification entirely. In a meta-analysis of 128 experiments, Deci, Koestner and Ryan studied tangible, expected rewards tied to doing a task. Such rewards reliably undermined people’s later free-choice engagement with that task (Deci, Koestner & Ryan, 1999). It is a negative effect on intrinsic motivation, strongest precisely for the reward structures gamification loves to copy. The mechanism, in self-determination terms: a salient controlling reward shifts a person’s felt reason for acting from “I want to” to “I’m being paid to.” The intrinsic reason erodes.

This is the overjustification trap. Gamification walks straight toward it whenever it staples points onto something people already found worthwhile. The practical upshot is uncomfortable for anyone selling “engagement”: the more genuinely interesting a task already is, the more a crude points overlay risks cheapening it. Where gamification is safest is exactly where intrinsic interest is lowest — the repetitive, effortful, easy-to-skip work that learning actually requires. Which is a useful clue about where to aim it.

Two boundary conditions keep the caution in proportion. Deci and colleagues found the undermining effect concentrated in tangible, expected, contingent rewards — rewards tied to doing the task. Verbal feedback and informational signals of competence did not carry the same risk, and could even boost intrinsic motivation (Deci, Koestner & Ryan, 1999). That distinction is the designer’s escape route. A mechanic that reads as information about progress — a mastery bar, a performance graph — sits on the safe side of the line; one that reads as payment for compliance sits on the dangerous side. The same badge can be either, depending on framing. The trap is real, but it is a trap with a posted map.

What the evidence doesn’t show

It is easy to over-read a positive meta-analytic average. Four things this literature does not establish:

• It does not show a large, dependable effect. The pooled effects are small-to-moderate, and the motivational and behavioural ones weaken under high-rigour analysis (Sailer & Homner, 2020). Anyone quoting gamification as a reliable multiplier is quoting the top of a wide, skewed distribution.
• It does not show that the game elements are doing the work. Many primary studies confound the mechanic with novelty, with more time-on-task, or with better feedback introduced alongside it. Reviews have repeatedly flagged short durations and the risk that measured gains are a novelty effect that fades (Hamari, Koivisto & Sarsa, 2014).
• It does not show a motivation boost you can bank on. The canonical points/levels/leaderboards trio moved performance but not intrinsic motivation in a clean experiment (Mekler et al., 2017), and under the wrong framing, contingent rewards can push intrinsic motivation the other way (Deci, Koestner & Ryan, 1999).
• It does not show the effect is uniform across people or contexts. Both the framing reviews and the moderator analyses find that outcomes vary with the user, the task and the competitive structure, and the volume of genuinely mixed results is itself a headline finding (Hamari, Koivisto & Sarsa, 2014). Averages here conceal more than they reveal.

Where the evidence stops

1. 1It does not show a large, dependable effect
2. 2It does not show that the game elements are doing the work
3. 3It does not show a motivation boost you can bank on
4. 4It does not show the effect is uniform across people or contexts

© 2026 FUTURE PROOF™

None of this makes gamification worthless. It moves the question. The right question is never “does gamification work?” — it is “which element, aimed at which action, for which learner, measured against which outcome?” Asked that way, the literature stops being a referendum and becomes a design manual — one whose chapters happen to be effect sizes.

The design rule that follows

Put the strands together — the outcome-level averages, the element-level experiments, the needs mapping, the overjustification boundary — and a single design principle falls out. Game mechanics reliably do one thing: they increase and direct effort. They do not, on their own, make that effort productive — and clumsily applied, they can taint tasks that were fine without them. So the leverage sits entirely in what you attach the mechanic to. Attach a streak, a point or a leaderboard to a low-value action — logging in, watching a video to the end — and you get more of a low-value action. Attach it to an action the learning science already endorses, and the mechanic borrows that action’s validity.

Across almost every review of durable learning, the evidence-backed action is retrieval practice: being made to recall, not merely see the material again. That is the action worth gamifying. There, the extra effort the mechanic buys is spent on the one behaviour most tightly linked to remembering.

The same principle sorts the common mechanics into an honest ranking. A streak that counts days of retrieval protects a habit worth having; a streak that counts logins protects a metric. XP earned per correct recall keeps the incentive pointed at the outcome; XP earned per minute of video watched pays people to leave the tab open. A leaderboard framed as a specific, committed team goal borrows the goal-setting evidence (Landers, Bauer & Callan, 2017); a raw individual ranking borrows only the social-comparison risks. In each pair the mechanic is identical. The difference is entirely in the action it is soldered to — which is why two products with the same feature list can sit at opposite ends of the effect spread.

And the overjustification evidence supplies the final filter: aim mechanics at the tasks people avoid, not the ones they already enjoy (Deci, Koestner & Ryan, 1999). Daily retrieval of half-forgotten material is effortful, unglamorous, and chronically skipped. That makes it at once the highest-value target in learning and the safest one in motivation terms. Gamification’s best use, it turns out, is not making learning fun. It is making the unfun part of learning get done.