The nightcap

This is not survey data about people who drink. It is a pooled analysis of 27 controlled experiments in healthy adults aged 18 to 70, published in Sleep Medicine Reviews in 2025 and pre-registered before it was run. In 25 of the 27, each person slept in the laboratory twice - once after a measured dose of alcohol and once in an alcohol-free control condition - so each participant is largely their own comparison, which removes most of the confounding that makes drinking research hard to read. Sleep was recorded by polysomnography, meaning electrodes measuring brain activity, eye movement and muscle tone through the night, which is how sleep stages are actually determined rather than estimated. Across the 27 studies, 37 different alcohol conditions were administered, from roughly one standard drink to well over six. The honest caveats are these. The reviewers judged that 21 of the 27 studies carried some risk of bias - most commonly because participants met the alcohol and control conditions in a fixed order rather than a randomised one, and in five studies because there was no proper gap between conditions. Sample sizes are small, in the dozens to low hundreds per outcome, because overnight brain recording is expensive. And a laboratory night with a weighed dose is not a Friday at home.

Gardiner C et al. 2025

The nightcap's entire promise is that it gets you to sleep faster. Pooled across every trial that timed it, it did not - the difference was -2.0 minutes with a range from -4.3 to 0.2, which includes no difference at all (34 effect estimates, 271 adults). That is the finding most people find hardest to believe, and it deserves a fair hearing on both sides. Against it: the review's dose-response analysis does show a real shortening, of 6.4 minutes for every extra gram of alcohol per kilogram of body weight, and that shortening becomes statistically clear at around five standard drinks. So alcohol can hurry sleep onset; it just takes a lot. Also against it: several of these trials asked people how fast they felt they had fallen asleep, and three of five found a reported improvement where the recording showed none. The feeling is real even when the clock disagrees. What survives all of that is the shape of the trade. The amount of alcohol that would genuinely shorten your time to sleep is roughly twice the amount that already measurably costs you REM sleep. The nightcap is not a small good thing with a small cost; at ordinary amounts it is the cost without the good thing.

Gardiner C et al. 2025

Three separate measurements in this review point the same way. The first REM period arrived 18.0 minutes later after alcohol than in the control condition (95% CI 8.8 to 27.1, from 29 effect estimates in 212 adults). REM sleep made up 2.8 percentage points less of the night (95% CI -3.9 to -1.7, from 32 effect estimates in 255 adults) - the review's most precisely estimated REM finding, with a p-value below 0.001. And in the smaller set of trials that reported REM in minutes, REM duration fell by 11.3 minutes (95% CI -20.8 to -1.8, from 13 effect estimates in 96 adults). These are three views of one phenomenon rather than three independent findings, and they should be read together, not added up. Two things they do not establish. They do not show any consequence: nobody in these trials was followed into the next day, so the reasons to value REM sleep come from other research. And they do not show a change in deep sleep across the night as a whole, despite the folk belief that alcohol deepens it - the pooled figures for stage three sleep, in minutes and as a share of the night, showed no measurable difference. That belief is not baseless, though, and the review is fair about it: the two trials that reported the night in halves found deep sleep increased in the first three hours and then reduced, along with REM sleep, in the second half. The reviewers described those two rather than pooling them, so there is no dependable number to attach to it. What shifted instead was the balance between REM and lighter stage two sleep, which rose by 1.8 percentage points.

Gardiner C et al. 2025

The doses in these trials are measured in grams of alcohol per kilogram of body weight, so the reviewers translated them using a person of 62 kg and a standard drink of 10 g. On that basis, the delay to the first REM period became statistically clear at about 0.35 grams per kilogram, or roughly two standard drinks; the loss of REM sleep at about 0.50, roughly three; and the shortening of time to fall asleep not until about 0.85, roughly five. Note that the paper's own abstract and discussion print "two standard drinks" for the 0.50 threshold, which its own figures and its own methods section contradict - 0.50 times 62 kg is 31 grams, which is three drinks of 10 grams, and the methods section says exactly that. The safe statement is that REM disruption is detectable somewhere in the range of two to three drinks. Two further cautions. A standard drink is not a fixed quantity: it is 8 grams in the United Kingdom, 10 in Australia, about 14 in the United States, so a British reader should treat these counts as generous and an American reader as cautious. And the share of the night spent in REM fell without depending on dose at all, which means there is no drink count below which that particular effect has been shown to disappear.

Gardiner C et al. 2025

Quantifying the timing of alcohol intake was one of the four stated aims of this review, and it is the aim it failed. Timing moved almost nothing: of every outcome tested, only REM sleep duration showed any timing effect at all, and it was both tiny and barely distinguishable from chance - not on time to fall asleep, not on the delay to REM, not on the share of the night spent in REM, not on total sleep time, sleep efficiency or time awake in the night. The reviewers explain why, and the explanation matters more than the result: almost every trial in the literature gave people alcohol within about three hours of bedtime, so there was barely any variation in timing to analyse. This is an absence of evidence, not evidence of absence. The three-hour rule is a physiological argument - alcohol clears from the blood at a broadly predictable rate, so leaving a gap should mean less on board at lights out - and that argument is reasonable. It has simply never been put to the test in a sleep laboratory. Which leaves an uncomfortable but honest position: the one variable this literature has actually pinned down is how much, and the answer there starts low. Building a rule out of the variable it could not pin down would be inventing a permission the evidence has not granted. Two things follow from that for anyone looking for a workaround. Do not substitute a timing rule for the change: the advice to finish drinking three or four hours before bed is untested, and the only review that set out to measure the effect of timing could not establish it, so nobody can tell you that an earlier drink is a safe drink. And if the reason you reach for it is that you cannot switch off, that is the thing worth addressing - it is a different problem from this one, one this review did not study, and this page gives no advice on it.

Gardiner C et al. 2025

That is too strong. What the pooled recordings show is no measurable head start at ordinary amounts, and a real one at around five standard drinks. What several of the same trials also show is that people often report falling asleep faster even when the recording says otherwise, so the feeling is not imaginary. The finding is about a measured outcome in a laboratory, in healthy adults aged 18 to 70, over a single night - not about whether alcohol makes anybody drowsy, which it plainly does.

This review cannot tell you that it is. The delay to the first REM period became detectable at around two standard drinks and the loss of REM sleep at around three, using the reviewers' assumption of a 62 kg person and a 10 gram drink - and the fall in REM's share of the night did not depend on dose at all, meaning no threshold was found below which it stopped. A single glass sits below the tested thresholds for two of the three REM measures, which is not the same as being shown to be harmless. The honest answer is that less is better evidenced than earlier, and that nobody has established a safe amount for sleep.

Talk to a doctor before you make a sudden change. Everyone in these trials was a healthy adult without alcohol dependence, and the reviewers say directly that their results may not generalise to people with alcohol use disorder or a diagnosed sleep disorder. Stopping abruptly after regular heavy drinking can be medically dangerous and is not something to do on the strength of a sleep finding. If drinking is how you get to sleep most nights, that is worth raising with a clinician for both reasons at once - the drinking and the sleep.