← Longevity & Supplement Guides
Is Yoga Really the Best Exercise for Sleep?
A headline has been going round again this month: yoga is the best exercise for sleep. Not helpful for sleep — best. First place, with a prescription attached. High intensity, twice a week, under thirty minutes, for eight to ten weeks.
That last sentence is the one that should make you suspicious. The research underneath it is real, the broad conclusion is probably half right, and the confident part is the part that isn't.
The paper underneath the headline
The source is a network meta-analysis by Li and colleagues, published in Sleep and Biological Rhythms in 2025: thirty randomised controlled trials, 2,576 people with sleep complaints, drawn from more than a dozen countries. Yoga finished top. Walking came next, then resistance work, then tai chi and other aerobic training. The stated optimum was that oddly specific combination — high-intensity yoga, twice weekly, half an hour or less, sustained for eight to ten weeks.
A network meta-analysis is a genuinely clever instrument. It lets you compare things that were never tested against each other by chaining trials together through whatever they shared — usually a control group. Yoga was never raced against Pilates, but if both were raced against "usual care," the maths can estimate the gap between them.
Clever, and fragile. Every link in that chain carries the assumptions of the trials it came from. The authors say so themselves: the evidence base is limited, and higher-quality research is needed before anyone leans on the ranking. That caveat did not travel with the headline.
Three rankings, three different winners
Here is the part that settles it. Run the same exercise on a bigger pile of trials and the crown moves to someone else.
The same question, three times
Three network meta-analyses of exercise for sleep — and three different first places
Li et al., Sleep and Biological Rhythms, 2025
Adults with sleep disorders. Winner: yoga — high intensity, 2×/week, ≤30 min, 8–10 weeks.
Wang, Xin and Pan, BMC Public Health, 2025
Measured by PSQI, ESS and ISI questionnaires. Winner: Pilates (SUCRA 91.7%), aerobic exercise second (69.7%).
Bu et al., BMJ Evidence-Based Medicine, 2025
Adults with insomnia, thirteen approaches compared. Winner: yoga, with walking or jogging and tai chi close behind.
Wang, Xin and Pan pooled eighty-six trials and 7,276 participants in BMC Public Health — almost three times the trials and nearly three times the people of the viral paper. Their winner is Pilates, with a ranking score of 91.7%, and aerobic exercise second at 69.7%. Yoga is not the champion.
Bu and colleagues, in BMJ Evidence-Based Medicine, put yoga back on top across twenty-two trials of adults with insomnia. Elsewhere in this literature you can find reviews that crown combined aerobic-and-resistance training, and reviews that crown tai chi.
When several careful teams sort roughly the same literature and hand the trophy to different practices, that is not a disagreement to be refereed. It is a result in its own right, and the result is this: the gaps between these activities are smaller than the noise between the studies measuring them.
A rank is not an effect
The number that gets quoted as authority in these papers — SUCRA, 91.7% for Pilates — is widely misread. It is not a claim that Pilates is 91.7 percent effective, or 91.7 percent better than the alternatives. It is a probability-weighted position in an ordering: across thousands of simulated re-runs of the analysis, this is how often that option finished near the top.
Which means an option can win by a sliver, or on the strength of two small, enthusiastic trials, and still post an impressive-looking score. A rank order tells you the sequence. It says nothing about the distance between the places. And the distance is the only thing a person deciding what to do on a Tuesday evening actually needs to know.
Nearly all of this was measured by asking people
The instruments doing the work here are questionnaires — the Pittsburgh Sleep Quality Index, the Insomnia Severity Index, the Epworth Sleepiness Scale. A handful of trials added a sleep-lab night or a wrist sensor. Most did not.
And nobody can be blinded to whether they spent ten weeks doing yoga. You know what you were assigned. You know it is meant to help. You fill in the form afterwards.
The cleanest demonstration sits inside the most yoga-friendly of the three reviews. In Bu and colleagues, tai chi increased total sleep time by roughly 52 minutes when measured by sleep diary — and roughly 24 minutes when measured objectively. Same practice, same review, less than half the effect once a sensor was doing the counting.
That gap is not fraud. It is what self-report does, in every field that relies on it. But it reframes the headline figures. Yoga adding nearly two hours of sleep a night — about 111 minutes, the one finding in that review rated moderate certainty — is best read as the ceiling of a measurement that includes how people feel about what they did.
And sit with the size of it for a moment. Two extra hours a night. If any ordinary, pleasant, freely available behaviour reliably bought two hours of sleep, we would not have needed a meta-analysis to notice.
What the evidence base is actually made of
Risk of bias, as graded by the reviewers themselves
Risk-of-bias judgements across the 22 trials in Bu et al. (2025)
Roughly one trial in five was clean. Two-thirds carried "some concerns," and one in seven was judged high risk. The certainty ratings tell the same story — a single moderate-certainty finding, and low certainty for everything else, including all the numbers that made the rounds.
This is ordinary for exercise research, not a scandal. Small samples, no possibility of blinding, short follow-ups, subjective endpoints. But it does fix the honest summary in place: exercise probably helps sleep, and we cannot yet rank the flavours.
There is one more thing in the raw material worth knowing: almost everyone in these trials slept badly to begin with — the populations are described as having sleep disorders, or diagnosed insomnia. That matters in two directions.
It is where the room to improve lives, so the effects look large. Someone already getting seven and a half unbroken hours has far less available to gain, and none of these numbers were measured on them.
It is also the reason not to import the figures onto yourself if you sleep reasonably well. And if your sleep is genuinely broken rather than merely imperfect, the useful next step is a conversation with a clinician, not a leaderboard.
The one place the measurement was objective
There is a study in this space with no questionnaire in it at all. Leota and colleagues, in Nature Communications in 2025, analysed 14,689 physically active adults wearing a consumer wrist-worn multi-sensor device across 4,084,354 nights.
Their finding: the later in the evening people trained, and the harder they went, the more their sleep suffered. Delayed sleep onset, shorter sleep, lower sleep quality, a higher resting heart rate overnight and lower heart-rate variability. Sessions that finished four or more hours before sleep onset showed no such association.
Now look again at the viral prescription. Its winning ingredient was high intensity. In the largest objectively measured dataset available, intensity is the variable that costs you sleep — when it lands close to bedtime.
The two results are not strictly contradictory, because the ranked trials mostly did not report what time of day people exercised. That absence is the point. The ranking optimised a variable it measured badly while ignoring the variable the objective data says matters most. If you want that thread pulled properly, we did it in the best time of day to exercise.
What every winner had in common
Strip the modality out and look at what the successful arms actually shared.
They were repeated for weeks — eight to twelve, not a fortnight. They landed at a moderate total dose: Wang and colleagues put the overall optimum near 920 MET-minutes a week, along a nonlinear, U-shaped curve, which is a careful way of saying that too little did little and more was not better. And the practices that did best were low-arousal things done in the evening, which is the same lesson the four-hour buffer teaches from the other side.
None of that is a modality. It is a habit, with a dose and a clock.
So the practical read is almost anticlimactic: stop shopping for the winning exercise and pick the one you will still be doing in week ten. That 920 MET-minutes is roughly what mainstream activity guidance already asks of you — we went through those numbers in how much exercise per week really. You are not being asked to take up something new. You are being asked to keep it up, and to finish the hard sessions earlier in the day.
You are the only sample size that applies to you
A between-groups ranking cannot tell you what works for your sleep. It was never designed to.
Two things are worth watching for three or four weeks, and both are things the objective study found actually move: how long it takes you to fall asleep, and your resting heart rate overnight. Shift your hard training an hour or two earlier and see whether either shifts with it. In the Agen app these are trends to notice over weeks — not measurements to diagnose anything, and not worth a verdict from a single night.
Keep the competition honest while you do it. A late coffee or an evening drink will comfortably out-argue your yoga, and both are easier to move than your training schedule — see caffeine and sleep and alcohol, sleep and recovery. Consistency of timing tends to beat heroics of duration, too, which is the argument in sleep regularity versus duration.
The bottom line
Exercise helps sleep. That survives every version of the analysis, and it is not a small thing to know.
Which exercise is best is a question this evidence cannot currently answer — and three teams have now proved that by answering it differently. The ranking is theatre. What is left when the theatre closes is worth more than the trophy: something you will repeat for two or three months, at a dose you already know, finished a few hours before you intend to sleep. Then watch your own numbers, because the leaderboard was never about you.
