Caffeine too late
Peer-reviewed research · 3 sources · Last verified 2026-08-05
Falling asleep is not the test. Caffeine can leave you sleeping less, and shallower, on nights when you had no trouble dropping off - and in a placebo-controlled trial people mostly could not tell. That is the problem: the thing you would judge it by is the thing it moves least. The amount matters as much as the hour, so a small cup late is not the same as a large coffee late. The fix is to pick your cut-off by the clock once, not by how awake you feel.
The one honest fix
Count back eight hours from the time you actually go to bed — that hour is your caffeine cut-off, and nothing after it has caffeine in it
Free · behavioral
What the evidence says
In a systematic review and meta-analysis of 24 controlled crossover trials in healthy adults, caffeine reduced total sleep time by 45.3 minutes compared with the control condition
-45.3 min · 95% CI 29.0-61.5 min
Limitations & how to read this number
The n of 340 counts the participants who contributed to this particular pooled estimate (37 separate effect sizes drawn from 20 trials that reported objective total sleep time), not the whole review. These were short, controlled laboratory-style trials in healthy adults aged 18-65 who were low-to-moderate habitual caffeine drinkers, so the findings may not carry over to teenagers, adults over 65, people who never drink caffeine, or very heavy drinkers. The 45 minutes is an average across a wide mix of doses and timings, from a cup-of-coffee dose to a large pre-workout dose, and from 18 hours before bed down to bedtime itself - it is not what one afternoon coffee does to you. The authors note a 20-minute change in total sleep time is the usual threshold for a clinically meaningful difference, so the average sits well past it. Three of the 24 trials used activity monitors rather than laboratory sleep recording, and those devices tend to overestimate how long people sleep.
SourceGardiner C et al. (2023) · Sleep Medicine Reviews · 10.1016/j.smrv.2023.101764
The same meta-analysis found that caffeine cut deep sleep - the combined N3 and N4 stages - by 11.4 minutes and shifted that time into the lightest stage of sleep, N1, which rose by 6.1 minutes
-11.4 min · 95% CI -4.3 to -18.5 min
Limitations & how to read this number
This is the deep-sleep estimate from 18 trials that recorded sleep stages; the n of 127 is the number of participants contributing to the sleep-architecture models, which is smaller than for total sleep time because staging needs laboratory sleep recording. Deep sleep fell by 11.4 minutes (95% CI -4.3 to -18.5, p = 0.012) and its share of the night by 1.4%, while the lightest stage N1 rose by 6.1 minutes (95% CI 2.3 to 9.9, p = 0.012) and by 1.7% of the night. Two stages showed no significant change: N2 and REM sleep. Losing deep sleep is a change in sleep quality, not a disease outcome - this study measured the structure of one night's sleep, so it cannot tell you what repeating that for years does to your health.
SourceGardiner C et al. (2023) · Sleep Medicine Reviews · 10.1016/j.smrv.2023.101764
To avoid a reduction in total sleep time, the meta-analysis model estimated that a standard cup of coffee containing 107 mg of caffeine should be taken at least 8.8 hours before bed, and a standard serve of pre-workout containing 217.5 mg at least 13.2 hours before bed - for a 10 pm bedtime, that is a coffee cut-off of 1:12 pm
8.8 h
Limitations & how to read this number
This number is a model estimate, not something anyone measured. The authors fitted a meta-regression to the pooled trials using the amount and timing of the final caffeine dose, then read off the earliest time at which the model's 95% confidence interval no longer crossed zero for a given amount. The estimate is only as good as that model: it assumes a statistically significant loss of sleep is also a meaningful one, and where a study gave caffeine several times in a day, only the last dose went into the model, which the authors say may have underestimated the influence of dose and timing. The same model gives no cut-off at all for a 47 mg cup of black tea, and half-life varies from about 2 to 10 hours between healthy adults, partly for genetic reasons, so the right cut-off for you may be earlier or later. The authors' own advice is to treat these times as a starting point and adjust based on how you actually sleep.
SourceGardiner C et al. (2023) · Sleep Medicine Reviews · 10.1016/j.smrv.2023.101764
How to actually do it
Do the arithmetic once, not every day. Take the time you genuinely go to bed - not the time you intend to - and count back eight hours. That hour is your cut-off, and from then on you drink things with no caffeine in them. Eight is a deliberate simplification of the 8.8-hour modelled figure for one cup of coffee, chosen because a round number survives real life; if your last drink of the day is regularly a large coffee, an energy drink or a pre-workout serve, the same model pushes the cut-off several hours earlier still. And if you want to know whether it is working, judge it on how much sleep you get, not on how quickly you fall asleep - falling asleep was the outcome caffeine affected least.
Supporting findings (4)
Caffeine also made it take 9.1 minutes longer to fall asleep, added 11.8 minutes of being awake during the night, and cut sleep efficiency - the share of time in bed actually spent asleep - by 7 percentage points
-7.0% sleep efficiency · 95% CI 4.0-10.0%
Three separate pooled estimates, each from a different set of trials, so they have different sample sizes and should not be added together: sleep onset latency +9.1 minutes (95% CI 3.8 to 14.4, 280 participants), wake after sleep onset +11.8 minutes (95% CI 2.5 to 21.0, 148 participants), and sleep efficiency -7.0 percentage points (95% CI 4.0 to 10.0, 274 participants). The authors judge the first and third to be clinically meaningful and the middle one not. Sleep efficiency is measured as a percentage of time in bed, so it falls both when you lie awake longer and when you sleep less. Time to fall asleep was the one outcome that did not clearly depend on when the caffeine was taken, which is a genuine loose end in the evidence.
SourceGardiner C et al. (2023) · Sleep Medicine Reviews · 10.1016/j.smrv.2023.101764
The amount matters as much as the clock - in a placebo-controlled, double-blind randomised crossover trial in 23 men, a 100 mg dose of caffeine had no significant effect on sleep even when taken 4 hours before bed, while a 400 mg dose delayed sleep initiation and altered sleep architecture whenever it was taken within 12 hours of bedtime
A single small trial: 23 men with an average age of 25, all moderate caffeine drinkers taking under 300 mg a day, each tested through seven conditions with a 48-hour gap between them. No women were included, so it says nothing directly about how this plays out for them. 'No significant effect' from the 100 mg dose is a failure to detect a difference in 23 people, which is not the same as proving there was none - a trial this size can only pick up fairly large effects. The caffeine was given as one single dose, whereas most people spread several drinks across a day, and the paper's own conclusion is that 400 mg taken as one dose within 12 hours of bedtime may harm sleep, with the damage growing the closer to bedtime it lands.
SourceGardiner CL et al. (2025) · Sleep · 10.1093/sleep/zsae230
People are bad at telling when caffeine has hurt their sleep - in that same trial, perceived sleep quality only dropped significantly when 400 mg was taken 4 hours before bed, a 34.02% fall, and was rated as unchanged at 8 and 12 hours even though the recordings showed sleep had been disrupted
-34.02%
This is a self-rating from 23 men in one trial, and the authors describe it as a discrepancy between what the recordings showed and what participants reported rather than a formally tested gap. The p-value for the drop at 4 hours is 0.006; at 8 and 12 hours the change was not statistically significant. Because it is one small study, treat it as a reason to check your own sleep against the clock rather than proof that nobody can feel caffeine's effects.
SourceGardiner CL et al. (2025) · Sleep · 10.1093/sleep/zsae230
The trial that produced the classic six-hour rule gave 400 mg of caffeine - roughly two to three cups of coffee - at bedtime, 3 hours before bed and 6 hours before bed, and every one of those timings significantly disturbed sleep compared with placebo, with about 1.2 hours less total sleep time
~1.2 h less total sleep time
Twelve participants, six men and six women, averaging 29 years old. That is a very small trial, and 400 mg in one hit is a large dose - roughly two to three standard cups of coffee, or one big energy drink. It is the study behind the familiar advice to stop caffeine six hours before bed, and it is one of the 24 trials pooled in the main meta-analysis here, so it is not independent confirmation of it. In this trial the amount of sleep lost was similar whether the caffeine came 6 hours before bed or right at bedtime, which is one reason the pooled evidence on timing is less tidy than the headline suggests.
SourceDrake C et al. (2013) · Journal of Clinical Sleep Medicine · 10.5664/jcsm.3170
Go deeper
Open what you care about
Why "I fall asleep fine" is not evidence
The thing caffeine damages most reliably is not sleep onset - it is the structure and total amount of the night. Pooled across 24 controlled crossover trials in healthy adults, caffeine reduced total sleep time by 45.3 minutes (95% CI 29.0 to 61.5, 340 participants) and sleep efficiency by 7 percentage points versus the control condition, while adding only about 9 minutes to the time it took to drop off. So the experience of getting to sleep normally is entirely compatible with losing three quarters of an hour of sleep. A separate placebo-controlled trial in 23 men made the same point from the other direction: perceived sleep quality only fell significantly when a large 400 mg dose came 4 hours before bed, and was rated unchanged at 8 and 12 hours even though the recordings had already changed.
What the 8.8 hours actually is
The headline number from this review is a cut-off time, and it is important to know how it was produced. The authors fitted a meta-regression across the pooled trials using the amount and the timing of the last caffeine dose, then defined the cut-off as the earliest time at which the model's 95% confidence interval stopped crossing zero for a given amount. For a standard 250 mL cup of coffee at 107 mg that lands at 8.8 hours before bed - 1:12 pm for a 10 pm bedtime. For a 217.5 mg serve of pre-workout it is 13.2 hours, or 8:50 am. For a 250 mL cup of black tea at 47 mg there is no cut-off at all. Nobody measured 8.8; it is a prediction, and the authors themselves say consumers should use these times as a starting point and then adjust based on their own sleep.
The dose is as important as the hour
A rule like "no caffeine after 2 pm" quietly assumes every drink is the same size. The evidence says otherwise. In the meta-regression, every extra hour before bedtime shrank the sleep loss by 2.8 minutes, and every extra milligram of caffeine grew it by 0.2 minutes - two levers, not one. That is why the modelled cut-off runs from "any time" for tea to nearly the whole waking day for a pre-workout serve. It also fits the later randomised trial, where 100 mg taken 4 hours before bed produced no significant effect on objective or subjective sleep, while 400 mg taken as a single dose within 12 hours of bedtime delayed sleep initiation and altered sleep architecture. Practically: a small tea late is a different decision from a large takeaway coffee or an energy drink late.
What caffeine does to the shape of the night
Beyond total hours, caffeine reorganises the night into a shallower one. Across the trials that recorded sleep stages, the combined deep-sleep stages N3 and N4 fell by 11.4 minutes and by 1.4% of the night, while the lightest stage N1 rose by 6.1 minutes and by 1.7%. N2 and REM sleep showed no significant change. Mechanistically this makes sense: adenosine accumulates while you are awake and generates the pressure to sleep, and caffeine blocks the A1 and A2A receptors that read that pressure. Deep sleep is where that pressure is discharged most intensely, so muting the signal thins out precisely the stage that repays the debt.
Why your personal cut-off may differ
Caffeine's half-life varies from about 2 to 10 hours across healthy adults, with a typical value of 3 to 6 hours. That spread is not willpower or tolerance; it is largely enzymes and circumstances. Variants in CYP1A2, the liver enzyme that clears caffeine, and in the ADORA2A adenosine receptor change both how fast you clear a dose and how strongly you respond to it, and oral contraceptives, some antidepressants, smoking, alcohol, obesity, impaired liver function, age and ethnicity all shift the half-life too. The included trials were run in healthy 18-to-65-year-olds who were low-to-moderate habitual drinkers, so people who never drink caffeine, very heavy drinkers, teenagers and adults over 65 are outside what this evidence covers.
FAQ
Do I have to give up coffee?
Nothing in this evidence says that. Every one of these trials compared a caffeine condition with a control condition on a single night - none of them tested giving up caffeine, and none measured any health outcome beyond that night's sleep. What the evidence supports is moving your last dose earlier and, if it is a large one, making it smaller. Morning coffee is untouched by all of this.
Why eight hours when the paper says 8.8?
Because 8.8 is a model output with real uncertainty around it, and a round number you can actually remember beats a precise one you will not use. Counting back eight hours is slightly stricter than the paper's estimate for a standard cup of coffee, which is the safer direction to round. If your last drink is a large coffee, an energy drink or a pre-workout serve, the same model puts the cut-off substantially earlier than eight hours.
Does this apply if I have been drinking coffee daily for twenty years?
Partly. The included trials specifically recruited low-to-moderate habitual caffeine drinkers, and the authors flag that the results may not generalise to people with a very high habitual intake or to people who never drink caffeine. There is a plausible tolerance effect, but too few of the trials gave caffeine over several consecutive days for the review to test whether the sleep effect shrinks with repeated exposure - so it is genuinely an open question rather than a settled reassurance.
Sources
- Gardiner C et al. (2023). Sleep Medicine Reviews. 10.1016/j.smrv.2023.101764
- Gardiner CL et al. (2025). Sleep. 10.1093/sleep/zsae230
- Drake C et al. (2013). Journal of Clinical Sleep Medicine. 10.5664/jcsm.3170
Last verified 2026-08-05 · adversarial fact-check + founder science gate.
Corrections are published openly. Educational content, not medical advice — talk to a professional about your own situation.