The science behind Mentalizi
Mentalizi builds on techniques whose first meta-analyses date back to the 1980s (Feltz & Landers, 1983). Here is what research measures about them, and what it does not prove yet.
These numbers measure an average gap between the people who did the exercise and the people who didn't. Not a promise about you. Around 0.5: pick one person at random from each group, and the one who trained comes out ahead about six times out of ten, and behind four times out of ten.
The detail: g, d and SMD all say the same thing, a gap counted in standard deviations. By convention 0.2 is small, 0.5 moderate, 0.8 large. The bracket that follows is the interval: the range of values the data can't rule out. The wider it is, the less we know. ⚠ A 95% confidence interval does not say "there's a 95% chance the true effect is in there". That is the definition of a credible interval, and only one of our four references publishes one.
⚠ These figures measure techniques, not our app. Mentalizi has not been through a controlled trial to date. We use this work to design our protocols; we cannot claim its results as our own.
The four measures behind our protocols
Pre-performance routines
Three estimates, not one. Pre–post: SMC = 0.31, 95% CI [0.18; 0.44], a small effect. Experimental, no pressure: g = 0.64, 95% CI [0.45; 0.83]. Under pressure: g = 0.70, but with a very wide interval, 95% CI [0.24; 1.16]: the true effect may be no more than small. We show the most favorable figure in large type; here is the most cautious one too.
No moderator comes out significant: not age, not gender, not competitive level, not routine complexity. Read that carefully: a moderator test that finds nothing does not prove the routine works the same for everyone; mostly it means the data can't settle the question. For us that's a reason to offer one protocol, not a guarantee of identical results.
⚠ And note what "routine" means in this literature: a few-second sequence immediately before a single attempt (a putt, a serve, a free throw), not a session repeated over weeks. Some of the pooled effects even come from techniques we don't offer (left-hand dynamic handgrip, quiet eye).
Rupprecht, Tran & Gröpel (2021 online / 2024 in print), International Review of Sport and Exercise Psychology, 17(1), 39–64. The literature search closes in 2020: citing “2024” alone would lend this work a freshness it doesn’t have. DOI
Mental imagery
One of the largest syntheses to date on rehearsing a movement in your head. Moderate effect, 95% credible interval [0.34; 0.67]. This is a Bayesian meta-analysis, not a confidence interval. The Bayes factor is 12.4: strong evidence, not decisive. And the evidence base is narrower than "motor performance" in general: agility, muscle strength, tennis, football.
⚠ Between-study heterogeneity reaches I² = 66%: the pooled studies are not quite measuring the same thing.
Liu, Zhao, Zhang, Zhang, Liang & Ning (2025), Behavioral Sciences, 15(5), 685. DOI
Self-talk
ES = 0.48, 95% CI [0.38; 0.58]. The effect is stronger when self-talk is taught rather than merely suggested, and on fine or newly learned motor tasks. In other words: you need a guide, not a slogan.
That moderator has a consequence worth stating: on a movement you already own, involving the whole body, the expected effect is smaller than on a fine skill you are still learning.
This is our most solid reference. An independent Bayesian replication (Corcoran & Steele, 2023), across 34 studies and 128 effects, lands on almost exactly the same number: 0.47 [0.39; 0.56]. Of our four references, it's the only one that has been replicated.
Hatzigeorgiadis, Zourbanos, Galanis & Theodorakis (2011), Perspectives on Psychological Science, 6(4), 348–356. DOI
Mental preparation, in controlled trials
The historic reference of the field, on sport performance measured at post-test. 95% CI [0.22; 0.92], across 35 randomized trials and 997 athletes.
⚠ One caveat we owe you: in that same meta-analysis, the largest effects show up when the intervention is delivered by the coach. None of the 35 trials evaluated an app. These trials measure protocols delivered by professionals and coaches. We take our cues from them, we don't stand in for them.
Brown & Fletcher (2017), Sports Medicine, 47(1), 77–99. DOI
What research does not prove yet
The most recent meta-analysis in the field is severe, and we would rather write it ourselves.
Reinebo and colleagues report moderate effects (psychological skills training: g = 0.83, 95% CI [0.21; 1.45]; imagery: g = 0.75, 95% CI [0.14; 1.36]; mindfulness: g = 0.67, 95% CI [0.01; 1.32]): three intervals that come close to zero before the sensitivity analysis even starts. Then they write that when non-randomized trials and subjective performance outcomes were removed in sensitivity analyses, the overall estimates of the effect size were no longer significant in any of the syntheses
. Their conclusion: these effects are not stable
.
⚠ We drew an argument from that sentence that wasn't in it. Reinebo and colleagues are not faulting protocols for being delivered inconsistently: they are faulting the trials for being non-randomized and for measuring performance by self-report. A standardized product fixes neither. What we take from it is smaller: we deliver the same protocol to everyone, which is what will make our own results checkable the day we run a controlled trial. We have not run one yet.
Three caveats we'd rather not write, and are writing anyway. Not one of the 25 studies synthesized by Reinebo was preregistered, so the results were not protected against analyses picked after the fact. Between-study heterogeneity is high: the studies are not quite measuring the same thing. And in nearly all these trials the participant knows they are getting the intervention: part of the effect may be expectation.
Reinebo, Alfonsson, Jansson-Fröjmark, Rozental & Lundgren (2024), Sports Medicine, 54(2), 347–373. 111 studies reviewed, 25 synthesized. DOI
How this translates into the product
- Short. A three-minute session actually gets done on a Tuesday night. That is a usage choice, not a research finding: the one reference of ours that tests dose, Liu et al. (2025), puts the optimum nearer ten minutes, three times a week, over about a hundred days. We would rather have a short session that happens than a long one that never starts. But that is a bet on adherence, not evidence.
- Repeatable. The same protocol, in the same order, every time. We take our cues from the pre-performance routine studies without claiming our format is the one they measured.
- Measurable. We track usage indicators and self-reported states, presented as a logbook, never as a verdict or a health score.
Mentalizi is a mental preparation and wellbeing tool for athletes. It is not a medical device, and it does not replace advice, follow-up or care from a health professional. If you are going through a difficult period, speak to a professional; in France, 3114 is free and available 24/7.
Literature last reviewed: 2 September 2026 · Written and checked in-house · No external funding. Every figure on this page was checked against the published abstract of the study cited; the DOI links lead to the source.