Sleep: what the trials show
Ashwagandha, lemon balm, magnesium, glycine, L-theanine and melatonin have all been through randomized trials. The best-evidenced supplement is not the one most people buy, the most popular one works differently than almost everybody thinks, and one item here helps the sleep you did not get.
Last scientifically reviewed: November 22, 2026Editorial review: MDS Labs. External scientific review pending.15 minute read
In brief
- Ashwagandha has the strongest pooled evidence, and the trials are specific: 600 mg a day or more, for 8 weeks or longer, with the clearest effects in people diagnosed with insomnia.
- Lemon balm is the only one here that changed sleep itself rather than opinions about it — about 15% more deep sleep on a monitor — but only as a bioavailability-enhanced extract, which is the whole point.
- Magnesium got people to sleep about 17 minutes faster. Glycine has the most elegant explanation of anything on this page and the thinnest evidence behind it.
- Melatonin is the surprise. In adults with chronic insomnia it was not significantly effective for falling asleep, staying asleep, or sleep efficiency. It is a clock-setting signal, not a sedative, and the curve peaks at just 4 mg.
- Creatine does not help you sleep. It helps when you have not slept — recovering up to 12% of the mental performance that a night of sleep loss takes away.
The four numbers every sleep trial reports
You cannot read this literature without these, and they are why a lot of sleep marketing gets away with vague claims.
Sleep onset latency (SOL) is how long it takes you to fall asleep. Wake after sleep onset (WASO) is time spent awake during the night after first falling asleep. Total sleep time (TST) is actual hours slept. Sleep efficiency is the percentage of time in bed you were asleep.
These move independently, and that is the point. A supplement can help you fall asleep faster without giving you one extra minute of sleep. When a label says "supports better sleep," it is usually declining to say which of these it means.
Three other terms appear below. The Insomnia Severity Index (ISI) is a short questionnaire scored out of 28, where lower is better and a drop of about 3 points is considered a real change. Slow-wave sleep is the deepest stage, the one most associated with feeling physically restored, and REM is the dreaming stage tied to memory and mood. Those last two can only be measured with monitoring equipment, which is why almost no supplement trial reports them.
Start with what works best, because it explains the rest
Cognitive behavioral therapy for insomnia — a structured programme combining sleep restriction, stimulus control, cognitive work and relaxation — was tested across 20 randomized trials in 1,162 adults with chronic insomnia 1.
The results are not what most people picture when they imagine fixing their sleep:
- Time to fall asleep: 19 minutes faster
- Time awake during the night: 26 minutes less
- Sleep efficiency: up 9.91%
- Total sleep time: up 7.61 minutes — not statistically significant 1
The most effective treatment for chronic insomnia barely increases how long you sleep. It compresses your time in bed into actual sleep, so you stop lying awake. The gains held at later follow-up and no adverse outcomes were reported 1.
Keep that in mind throughout. "Better sleep" in these trials almost always means falling asleep faster and waking less, not sleeping longer.
Ashwagandha
Five randomized placebo-controlled trials in 400 adults, and the best pooled supplement result on this page 2.
What the trials show
An overall sleep effect of −0.59, which in this field counts as moderate and genuinely useful 2.
The subgroup findings are what make it actionable. The effect was more pronounced in adults diagnosed with insomnia, at doses of 600 mg a day or more, and with treatment lasting 8 weeks or longer 2. Mental alertness on rising improved too, along with anxiety — which may be part of the mechanism rather than a separate bonus.
What they do not show
Quality of life did not significantly improve 2. Five trials is a small base, and heterogeneity between them was moderate at 62%, meaning they did not all find the same size of effect.
The eight-week minimum is the part most likely to be ignored in practice. A two-week trial of ashwagandha is not a test of what these studies measured.
Lemon balm, and why the formulation is the finding
Lemon balm (Melissa officinalis) is an old remedy with a modern problem, and the trial that solved the problem is the most interesting single study here.
What the trial shows
A double-blind, placebo-controlled, cross-over trial — meaning each participant took both the extract and the placebo, which controls for individual differences better than comparing two groups — tested a standardised extract delivered as a Phytosome, a formulation built specifically to improve how much of the active polyphenol content gets absorbed 3.
- Insomnia Severity Index fell to 6.8, against 9.7 on placebo: a 2.9-point reduction (p = 0.003)
- Slow-wave sleep increased by about 15% and REM decreased by about 10%
- 87% of the treated group reported better sleep, against 30% on placebo (p = 0.0003) 3
That deep-sleep figure makes this the only item on the page with evidence of changing sleep architecture — the actual structure of the night — rather than only how people rate it afterwards.
Lemon balm also has a separate meta-analysis behind it for anxiety (−0.98) and depressive symptoms (−0.47), with no serious side effects 4. Given that lying awake is often an anxiety problem wearing a sleep costume, that may be the route the sleep benefit travels.
What it does not show
This is one trial, against five for ashwagandha and 19 for L-theanine. One good trial is a reason for interest, not a settled case.
And the formulation caveat cuts both ways. The researchers were explicit that lemon balm's effects have not been reliably reproduced in humans, probably because the active compounds are poorly absorbed, which is why they used an enhanced delivery system 3. So this is evidence for a specific engineered extract, not for lemon balm tea or a generic capsule. The anxiety meta-analysis also warns of high heterogeneity between its trials 4.
Magnesium
Three randomized placebo-controlled trials in 151 older adults with insomnia 5.
What the trials show
Sleep onset latency was 17.36 minutes shorter than placebo, with a confidence interval from 7.44 to 27.27 minutes and a solidly significant p-value of 0.0006 5. Falling asleep a quarter of an hour sooner is an effect you would actually notice.
What they do not show
Total sleep time improved by 16 minutes and that result was not statistically significant 5. Faster to sleep, not more sleep.
The honest framing comes from the reviewers themselves: all three trials were at moderate-to-high risk of bias, and they graded the evidence low to very low quality, concluding the literature is substandard for physicians to make well-informed recommendations 5.
They still landed mildly in favour, and the reasoning is worth borrowing. Magnesium is cheap, widely available and safe in ordinary amounts, so the cost of a weak positive being wrong is low 5. That is a different argument from a strong evidence base, and it is reasonable as long as nobody pretends otherwise.
Glycine, the best explanation and the thinnest evidence
Glycine is an amino acid, and its sleep story is unusual: the mechanism is more convincing than the trials.
What the research shows
Taken before bed, glycine improved subjective sleep quality in people with insomniac tendencies. The proposed reason is genuinely satisfying — oral glycine raises glycine levels in blood and cerebrospinal fluid and lowers core body temperature by increasing blood flow to the skin. Falling asleep naturally involves a drop in core temperature, and a low core temperature is maintained through the night, so glycine appears to push a switch the body already uses 6.
There is an aging angle too. A systematic review covering eleven physiological systems found the nervous system responded most positively to glycine, and noted that dietary glycine extends healthy lifespan in model organisms and may reduce inflammation in humans 7.
What it does not show
Now the honest part. That same systematic review found that while longer-term glycine improved sleep in healthy populations, those studies had small sample sizes and a high risk of bias 7. And a 2026 scoping review mapping this entire literature found only three trials had investigated supplemental glycine and sleep outcomes at all 8.
Three small, bias-prone trials and a good mechanism is not the same as proof. A plausible story about why something should work is the most seductive kind of weak evidence, because it feels like understanding.
One practical note from that review: collagen is rich in glycine, and glycine-containing supplements including collagen do reliably raise circulating glycine levels 8. So anyone already taking collagen is getting glycine, whether or not that is why they are taking it.
L-theanine
The amino acid in tea responsible for the calm that coffee does not produce. Nineteen randomized trials in 897 participants 9.
What the trials show
Improvements in subjective sleep onset latency (0.15), subjective daytime dysfunction (0.33) and overall subjective sleep quality (0.43) 9. The daytime result is interesting on its own — how you function the next day is arguably what you actually care about.
What they do not show
Note the word repeated in every outcome: subjective. These are questionnaire scores, not monitored sleep. That does not make them worthless, since how you feel about your sleep matters, but it is weaker than a measured 17 minutes or a measured 15% more deep sleep.
The effect sizes are small, and two of the three intervals nearly touch zero (0.01 to 0.29, and 0.04 to 0.83). The reviewers added a practical caveat: few trials tested pure L-theanine, and the right dose and duration have not been established 9.
Melatonin, and what it is actually for
The most popular sleep supplement in the world, and the one whose evidence is most often misread. Two analyses tell the story together.
What the trials show
A dose-response meta-analysis of 26 double-blind randomized trials, 1,689 observations, found melatonin does reduce sleep onset latency and increase total sleep time — and that the curve peaks at 4 mg a day, beyond which more adds nothing 10.
The more useful finding is what predicted whether it worked. The interval between taking melatonin and the sleep episode significantly predicted how fast people fell asleep, and time of day was the only significant predictor of total sleep time 10. Timing is not a detail here. It is the variable.
Separately, across 31 trials of various supplements, melatonin did significantly improve subjective sleep quality, alongside amino acids and vitamin D 11.
What they do not show
Here is the part that should change what people buy. In a review of 24 randomized trials in chronic insomnia, melatonin in adults was not significantly effective for sleep onset latency, total sleep time, or sleep efficiency. It was significantly effective in children and adolescents, and it improved sleep onset latency in insomnia occurring alongside another condition 12.
So an adult with ordinary chronic insomnia, taking melatonin at bedtime as a sedative, is using it for the one job the trials do not support.
None of this means melatonin does nothing. It means it is a circadian signal — it tells your body what time it is. That is why timing predicts the outcome, why the dose tops out at a modest 4 mg, and why its real strengths lie in problems of clock alignment such as jet lag or a sleep schedule that has drifted later than your life allows. Taking 10 mg at midnight to feel drowsy is not how it works.
Creatine, for the sleep you did not get
This one belongs in a category of its own, because it does not improve sleep at all. It reduces the damage of going without.
What the trials show
Researchers kept volunteers awake for 21 hours and gave them either a single large dose of creatine monohydrate or a placebo, scanning their brains with phosphorus magnetic resonance spectroscopy — a technique that reads the brain's energy chemistry directly rather than inferring it.
At 0.35 g per kg of body weight, creatine shifted those brain energy markers, prevented the fall in brain pH that sleep deprivation causes, and improved cognitive performance and processing speed, partially reversing fatigue-related decline 13.
A follow-up in 29 people tested a lower 0.2 g/kg dose and found it still reduced the deterioration in logical and numerical tasks, language processing speed, and reaction-time vigilance — an improvement of up to 12%, smaller than the high dose but real. Women benefited more than men on several measures 14.
The mechanism makes sense. Sleep deprivation depletes the brain's energy reserves, and creatine is how cells buffer short-term energy demand. You are topping up a tank that being awake too long has drained.
What they do not show
Nothing about sleep quality, sleep duration, or falling asleep. If you sleep badly, creatine does not fix that — it makes the following day cost you less.
And the doses need stating plainly, because they are easy to misread. These were single acute doses scaled to body weight: 0.35 g/kg is about 28 g for an 80 kg person, and even the lower 0.2 g/kg is about 16 g. That is several times the 3 to 5 g daily maintenance dose used in the muscle research, taken once, under a research protocol, during enforced wakefulness. It is a finding about brain energy metabolism, not a recommendation to take a scoop before a night shift. Both trials were also small and conducted in healthy volunteers over a single night.
The doses and durations that were actually studied
| What | What improved | The condition attached |
|---|---|---|
| CBT for insomnia | SOL −19 min, WASO −26 min, efficiency +9.91% | Barely changed total sleep time |
| Ashwagandha, ≥600 mg/day | Overall sleep, SMD −0.59 | 8 weeks or longer; best in diagnosed insomnia |
| Lemon balm, bioavailability-enhanced extract | ISI −2.9 points, deep sleep +15% | One trial; generic extracts have not reproduced it |
| Magnesium, under 1 g per dose | SOL −17.36 min | No significant gain in total sleep; weak trials |
| Glycine, before bed | Subjective sleep quality; lowers core temperature | Only 3 sleep trials, high risk of bias |
| L-theanine | Subjective sleep quality and next-day function | All self-reported; dose undetermined |
| Melatonin, peaking at 4 mg/day | SOL and total sleep time; subjective quality | Not effective in adults with chronic insomnia; timing is the variable |
| Creatine, single 0.2–0.35 g/kg | Up to 12% less cognitive loss when sleep-deprived | Does not improve sleep; acute research doses |
How to read these results
Three checks separate a meaningful sleep claim from a marketing one.
Minutes and percentages beat effect sizes. Magnesium's "17.36 minutes" and lemon balm's "15% more deep sleep" tell you something you can picture. Ashwagandha's "−0.59" needs translating, and as a rough guide 0.2 is small, 0.5 moderate, 0.8 large. When a product cites an effect size rather than a concrete quantity, it is often because the quantity was unimpressive.
Ask whether it was measured or reported. Monitored sleep and a morning questionnaire are both legitimate, but they answer different questions and the questionnaire is far more open to expectation. L-theanine's entire result is subjective, and so is glycine's. Lemon balm's deep-sleep figure, magnesium's minutes and CBT's numbers are not.
Check which of the four numbers moved. Almost everything here improves how fast you fall asleep, and almost nothing reliably increases how long you sleep. A claim of "more sleep" deserves the most skepticism, because even the best treatment on this page could not deliver it.
What this adds up to
There is real, usable evidence here, and it points in a direction most sleep aisles do not.
For the best-supported supplement, it is ashwagandha at 600 mg or more for at least two months — not a quick fix, and specific enough that you can check whether a product matches what was tested. The most intriguing single result belongs to lemon balm, because it is the only one that moved measured deep sleep, with the significant condition that it was an extract engineered for absorption rather than lemon balm in general. For difficulty falling asleep on a budget, magnesium has a believable 17-minute effect resting on admittedly weak trials. L-theanine and glycine are mild, self-reported benefits — glycine with the most elegant mechanism on the page and the least evidence to support it.
If your problem is that you cannot fall asleep at a normal hour, or you are crossing time zones, melatonin is the right tool — low dose, with attention to timing rather than swallowed as you turn out the light. If your problem is chronic insomnia and you are an adult, the trials say melatonin is not your answer, and the thing that works better than anything else here is a behavioral programme you can ask a doctor to refer you to. And if the problem is simply that tonight is already lost, creatine is the one item that addresses tomorrow rather than tonight.
Two patterns are worth carrying away. Better sleep, as measured by the best evidence available, mostly means less time lying awake rather than more hours asleep — products promising extra hours are promising what the research has not delivered. And lemon balm makes the second point for us: with these botanicals, the delivery system can be the difference between a result and a shrug. The same plant failed to show consistent effects for years until someone fixed the absorption.
Keep reading
The lemon balm result is a case study in the idea covered in Bioavailability and half-life — why the same compound can work or do nothing depending on how it is delivered. Sleep and thinking are closely linked, and the memory evidence is in Memory and focus: what the trials show. Creatine's muscle evidence, including the ordinary daily dose, is in Strength and frailty: what the trials show. For the difference between a measured outcome and a reported one, see Understanding evidence.
For a compound often sold for sleep and mood that we have looked at in depth, see 5-HTP: How It Works and What the Trials Show.
Insomnia that persists for months, loud snoring with daytime sleepiness, or sleep that does not refresh you can point to a treatable disorder such as sleep apnoea, and that is worth a doctor rather than a supplement. Ashwagandha is not recommended in pregnancy and interacts with thyroid medication and sedatives; melatonin interacts with blood thinners and some blood pressure drugs; high acute doses of creatine warrant a check with a pharmacist if you have kidney disease.
