Do mental health apps work? The evidence is real, small, and fragile
The largest syntheses find apps beat doing nothing on depression and anxiety by about a third of a standard deviation — but the edge shrinks against real treatment, and one umbrella review found no convincing proof.
Mental health apps do outperform doing nothing, but the effect is small and it does not survive a demanding comparison. The two largest evidence syntheses agree that app-based interventions move depression and anxiety by roughly a third of a standard deviation against an inactive control [s1][s2], and both note that the advantage narrows or vanishes when the app is measured against another active treatment [s1][s2]. That is the honest answer to a question the app stores answer with a download button.
What the pooled trials show
The most cited efficacy synthesis is a meta-analysis of 66 randomised controlled trials of app-supported smartphone interventions [s1]. Pooled against control conditions, the apps significantly improved depressive symptoms (Hedges' g = 0.28, from 54 comparisons), generalized anxiety (g = 0.30, 39 comparisons), stress (g = 0.35), quality of life (g = 0.35), general psychiatric distress (g = 0.40), social anxiety (g = 0.58) and positive affect (g = 0.44) [s1]. By convention a g near 0.2 is small and near 0.5 is moderate, so most of these sit at the small-to-moderate boundary.
The same analysis found where apps did nothing. There was no significant benefit over control on panic symptoms (g = -0.05), post-traumatic stress symptoms (g = 0.18) or negative affect (g = -0.08) [s1]. And crucially, when app interventions were compared not with a waitlist but with an active intervention — face-to-face or computerised treatment — they did not differ significantly, though the authors cautioned that the number of such head-to-head trials was low [s1]. The apps that produced the larger effects shared two features: they were built on cognitive behavioural therapy, and they offered professional guidance and reminders to keep people using them [s1].
The sceptical reading
A stricter appraisal reached a colder conclusion. A 2022 meta-review pooled 14 meta-analyses covering 145 randomised trials and 47,940 participants, and graded each pooled effect using umbrella-review rules that ask whether the evidence is large, consistent and free of publication bias [s2]. Judged that way, the authors "failed to find convincing evidence of efficacy" for any effect — the top tier of evidence was empty [s2].
Eight effects reached the next tier, "highly suggestive," including smartphone interventions beating inactive controls on psychological symptoms and quality of life, with effect sizes of 0.32 to 0.47 [s2]. But the pattern underneath was the important part: the magnitude and the strength of evidence "tended to diminish as comparison conditions became more rigorous" — from inactive to active controls, from non-specific to specific ones [s2]. Publication bias was rarely even assessed for the representative effects, and adverse effects were not reported at all [s2]. An intervention studied 145 times that has almost never been checked for harm is a data gap, not a clean bill.
What it means for a reader
Three things follow, and none of them is "apps do not work." First, the benefit over nothing is real but modest, and it is largest for depression, generalised anxiety and stress — not for panic or PTSD [s1]. Second, an app is not equivalent to therapy: against active treatment the measured advantage disappears [s1][s2], and the apps proven in trials are frequently not the ones a person can actually download — about half of trialled self-guided apps are publicly available, with evidence quality having no bearing on which reach a store. Third, guidance matters: the trials that attached a human, even lightly, produced the larger effects [s1], a pattern that recurs elsewhere — a separate synthesis found digital therapy for panic disorder worked far better with a clinician attached than alone.
The reasonable position is that a well-designed, CBT-based app is a low-intensity option for someone who cannot access standard care, which is roughly what the efficacy analysis concluded [s1] — not a replacement for it, and not, on the current evidence, a proven benefit once a real alternative is on the table.
Sources
- [s1] The efficacy of app-supported smartphone interventions for mental health problems: a meta-analysis of randomized controlled trials — World Psychiatry (2019)
- [s2] Mobile phone-based interventions for mental health: A systematic meta-review of 14 meta-analyses of randomized controlled trials — PLOS Digital Health (2022)
- [s3] Identifying what works in mental health apps through meta-regression analyses of 169 trials — npj Digital Medicine (2026)
Sources
- The efficacy of app-supported smartphone interventions for mental health problems: a meta-analysis of randomized controlled trials — World Psychiatry , September 9, 2019
- Mobile phone-based interventions for mental health: A systematic meta-review of 14 meta-analyses of randomized controlled trials — PLOS Digital Health , January 18, 2022
More on
The mental health app features people use most are not the ones that work
A meta-regression of 169 trials coded 34 therapeutic ingredients. The most common — psychoeducation, relaxation, mindfulness — were not the ones that best predicted getting better.
Mental health apps aimed at teenagers are mostly commercial, and barely tested for harm
A survey of seven countries found the youth mental health app market dominated by paid journaling and meditation apps. The AI-powered ones show early promise on symptoms — and none had checked for harm.
Meditation apps do a little, even without a teacher — and mostly for stress
Across 34 trials, mindfulness apps produced small effects on stress, anxiety and depression, and none on general distress. A 2026 review confirms they can help without a facilitator — at low certainty.
Half the mental health apps proven in trials cannot be downloaded
A review of 110 studies found that 52% of trialled self-guided apps were publicly available — and that evidence quality had no bearing on which ones made it to an app store.