A review of 29 randomised trials found app-delivered insomnia therapy works. Where trials put it directly against a human therapist, the human won.
Sleep apps are marketed with a specific implied promise: that the thing in your pocket is a reasonable substitute for a clinician. A review published in npj Digital Medicine in March 2025 pooled 29 randomised trials, which allocate people to one condition or another by chance, covering 9,475 people, and it supports half of that promise clearly while undercutting the other half.
The apps work. Measured against control conditions, fully automated digital cognitive behavioural therapy for insomnia, a structured programme that changes sleep habits and the thinking around them, produced a standardised mean difference of 0.71, a gap measured in units of the spread of scores within the groups. That is a moderate to large improvement in insomnia severity. Where trials compared an app against therapy delivered with a human involved, the therapy came out ahead by 0.61. Both things are true, and most coverage picks one.
How we know
The review restricted itself to randomised controlled trials of fully automated programmes, meaning no therapist contact of any kind, not even messaging. That restriction is the point. Plenty of digital insomnia research involves a clinician somewhere in the loop, and pooling those with genuinely standalone apps would flatter the apps.
One number needs explaining before the rest make sense. The headline analysis pools 32 comparisons from 29 trials, because some trials tested more than one arm and each arm contributes separately. That is standard practice and it is also why the count of comparisons exceeds the count of trials.
That standardised mean difference is used because trials measure the same thing on different scales, which insomnia research does constantly. As a rough guide, 0.2 is small, 0.5 moderate and 0.8 large, though those thresholds are conventions rather than facts.
What these apps are actually delivering
Cognitive behavioural therapy for insomnia is not sleep hygiene advice, and the distinction matters for reading this review. It is a structured programme, usually four to eight sessions, built around a few specific techniques. Sleep restriction deliberately shortens time in bed to match the hours actually slept, then extends it as sleep consolidates. Stimulus control breaks the learned association between the bed and lying awake, by getting people out of bed when they cannot sleep. Cognitive work targets the beliefs that keep people anxious about sleeping, which is often what sustains the problem long after whatever started it has gone.
Those techniques are demanding, and two of them ask people to do something that feels wrong: spend less time in bed while exhausted, and get up in the middle of the night. That is the usual explanation for why a human helps, since a therapist can hold someone through the uncomfortable early weeks. It is also why the apps in this review are not interchangeable with the sleep trackers and relaxation apps that dominate the app stores, which deliver none of this.
What the apps were compared against
The comparison group turns out to matter more than anything else in this review, and the pattern is instructive. Against a waiting list, meaning people offered nothing at all, the apps produced 0.88. Against online sleep education, essentially reading material about sleep, 0.93. Against a placebo condition, 0.98.
Those three are close together, which is mildly reassuring: the benefit does not evaporate when the comparison group gets something rather than nothing. The authors found that the type of control group was the only variable that significantly explained the disagreement between trials, which is a candid thing to report, because it means the bulk of the variation remains unexplained.
Then there is the fourth comparison. Three trials put a fully automated app against therapy involving a human, and those trials favoured the human by 0.61. It is the smallest group of trials in the review and it deserves the most caution. It is also, notably, the comparison where the trials agreed with each other most closely: heterogeneity, the extent to which results disagree by more than chance would explain, was 16 percent there against 91 percent on the main result.
Why it matters
Insomnia therapy has a supply problem. Cognitive behavioural therapy for insomnia is the recommended first line treatment in most clinical guidance, and there are nowhere near enough trained therapists to deliver it to everyone who meets the criteria. That gap is the entire commercial case for the apps, and this review says the case is real: an app is meaningfully better than the nothing most people currently get.
What it does not support is the stronger claim, that automation matches a clinician. On the available trials it does not, though the evidence for that specific comparison is thin enough that a few more trials could move it in either direction.
The heterogeneity is the part a careful reader should hold onto. At 91 percent, the trials are not measuring one stable effect that the review has pinned down. They are measuring something that varies a lot between studies for reasons largely unidentified. A pooled number in that situation is a reasonable summary and a poor prediction of what any particular person would experience.
Heterogeneity was very high, 91 percent on the main result, meaning the trials disagreed with each other far more than chance would explain, and the authors could account for that only by the type of control group used. The comparison against a human therapist rests on just three trials, so it is the least certain number here despite being the most quotable. Adherence was measured inconsistently across trials, quality of life data were too sparse to pool, and because these apps are updated continuously, the software tested is not necessarily the software anyone would download today.
How long the benefit holds: the review pooled follow up as well as end of treatment and found the effect sustained, but its long term window means anything past six months and only four trials reached it. Whether an app helps people whose insomnia sits alongside another condition, which most trials did not isolate. And which parts of the therapy are doing the work, because the trials tested whole programmes rather than their components.
Are sleep apps worth using then?
On this evidence they are better than nothing, and better than reading sleep education material. The review found a moderate to large improvement in insomnia severity across 29 trials. What it does not show is that an app matches therapy involving a person.
How much worse is an app than seeing a therapist?
In the three trials that compared them directly, therapy with a human was ahead by a standardised mean difference of 0.61, a moderate gap. Three trials is a small evidence base, so that figure is the least certain in the review even though it is the most striking.
What does 91 percent heterogeneity mean?
It means the trials disagreed with each other far more than chance would explain, so they are not all measuring one stable effect. The authors could explain the disagreement only by which control group each trial used. A pooled average from such varied trials summarises the research but predicts any individual result poorly.
Does this apply to the sleep app I already have?
Not necessarily. The review tested specific programmes delivering cognitive behavioural therapy for insomnia, not sleep trackers or relaxation apps. The authors also note that these products are updated continuously, so the version tested in a trial may differ from the one available now.
This is research reporting, not medical advice. We describe what studies found. We never tell you what to do. Talk to a doctor before changing anything about your health. Read our full position.
The evidence, read properly, every three days.
We read the paper so you do not have to. Free, and no supplement ads, ever.
Powered by Buttondown