Do meditation apps work for anxiety? The trials, minus the hype

Some, in some people, against some comparisons. A large review of meditation programs found anxiety scores dropped by a small to moderate amount after eight weeks of training.¹ The catch is what the other group was doing. Against people doing nothing, the drop looks real. Against people doing something else, it shrinks, and in social anxiety trials it disappears.^2,4

What these trials actually compared

Most of the studies behind mindfulness apps and courses are not drug trials. They put one group through a structured program, usually built on the same core as mindfulness-based stress reduction, and compare their symptom scores against another group. What that other group did is the whole story.

A wait-list control means people signed up, were told to wait, and got nothing until the trial ended. Against that bar, a comprehensive analysis pooling more than 200 studies found a moderate benefit, sized at Hedge's g of 0.53.² A separate analysis restricted to healthy adults, not patients, found almost the same number in the same direction.³

An active control is different. It means the comparison group also did something structured: relaxation training, exercise, another kind of therapy. Measured that way, the same body of trials drops to a smaller effect, g of 0.33, and drops again to 0.22 once the comparison is another form of psychological treatment.² The program still helps. It helps less than the wait-list number implies.

Mindfulness effect, by what it was compared withHedges' g

Before vs after, same peopleg 0.55

Against a wait-listg 0.53

Against other active treatmentsg 0.33

Against other psychological treatmentsg 0.22

Where the gap disappears almost entirely

The clearest single example is a meta-analysis of mindfulness-based interventions for social anxiety disorder. Comparing people who did the training against people getting no treatment, the effect was large, g of 0.89.⁴ Comparing the same training against another active treatment, the effect fell to a small negative number, g of -0.19, meaning it performed about the same as whatever it was measured against.⁴ Against cognitive behavioral therapy specifically, it did somewhat worse, g of -0.29.⁴

Inside that same trial set, the within-group change before and after training was large, g of 1.20.⁴ People who did the program felt better than when they started. Was the improvement the mindfulness training itself, or having a scheduled activity, or talking to someone regularly? The between-groups number is built to answer that question, and it answered close to zero.⁴

This is the part that gets lost when a single number is quoted out of context. A before-and-after score and a between-groups score are answering different questions. The first asks whether people changed. Almost any structured program will show some change, because people rarely stay perfectly still between two measurements. The second asks whether the program did something a comparison activity would not have. That is the number that should decide whether an app is worth the time it asks for, and it is the number the marketing rarely leads with.

Stress moves more than anxiety does

Not every outcome behaves the same way. In a review of mindfulness training for nurses, the effect on stress was moderate, at a standardized mean difference of -0.81, though the reviewers rated the certainty very low.⁵ Anxiety, measured in the same trials on the same people, showed no significant effect at all, at -0.30.⁵ The training changed how stressed people said they felt more reliably than it changed a formal anxiety score.

Mindfulness training for nurses: change against comparison groups0 = no effect

Stress-0.81 (-1.11 to -0.52)

Anxiety-0.30 (-0.72 to 0.13)

Depression-0.24 (-0.55 to 0.07)

That split shows up elsewhere too. The large 2014 review found anxiety improved by a moderate amount at eight weeks and a smaller amount at three to six months, alongside only weak evidence for stress and distress specifically.¹ Two reviews measuring roughly the same intervention landed on different outcomes as the strongest one. That is a sign the true effect is being pulled around by which symptom got measured, in which population, more than it reflects one clean number.

Where it held up best

The steadiest results came from people who started out unwell rather than merely busy. In cancer patients, a review of 36 studies found medium effects on both anxiety, Hedges' g of 0.55, and depression, g of 0.43, and those gains were still present at least three months after the program ended.⁶ That combination, a real size and a follow-up that confirms it lasted, is rarer in this literature than the top-line averages suggest.

Healthy, non-clinical adults saw a similar-sized number on paper, a between-group Hedge's g of 0.53 against controls.³ But "healthy individuals" and "cancer patients with genuine anxiety" are not comparable starting points, and a review of otherwise well people is a different bet than one built on people who came in distressed.

That difference matters for anyone deciding whether to bother. A person managing everyday stress at work is closer to the healthy-adult trials. A person managing diagnosed anxiety is closer to the cancer and social-anxiety trials, where the picture is mixed. It was real when checked months later in people who were genuinely unwell, and close to zero when the comparison group also had something structured to do.

What remains unknown

Delivery format is the biggest gap. None of the trials pooled here tested an app against a person leading a course in a room, so nothing in this evidence base can say whether a phone reproduces what an instructor does. The mechanism could be the meditation, or it could be structure, attention, and a scheduled reason to sit still, and only a head-to-head trial against another active format would separate them.

Duration is thinner than it looks. The clearest numbers here run to eight weeks, with one review checking back at three to six months.¹ Little pooled here says what a year of daily use does, in either direction.

What survives is a real, moderate effect against doing nothing, that shrinks as soon as the comparison group is also doing something, and in social anxiety trials fell to about zero. An app that replaces a wait-list is not the same claim as an app that replaces a therapist, and the trials so far have mostly tested the first one.