TL;DR. Caffeine is the most widely used performance drug on earth and almost nobody doses it deliberately. The research is better than the advice built on it. The European Food Safety Authority concluded that single doses up to 200 mg give no safety concern and that as little as 100 mg near bedtime can lengthen sleep latency and shorten sleep. The International Society of Sports Nutrition puts the reliably ergogenic dose at 3 to 6 mg per kilogram of body mass, which for a 70 kg adult tops out at 420 mg, 2.1 times the single dose EFSA examined. Those two documents are talking past each other and the gap is where most people live. Meanwhile a 2024 randomised, placebo-controlled crossover trial with in-home polysomnography found that 400 mg significantly reduced deep sleep even when taken 12 hours before bed, while the drinkers only perceived a problem at 4 hours. We took EFSA’s own half-life range and computed residual caffeine at bedtime: an identical 400 mg dose leaves 6.3 mg in a fast metaboliser and 141.4 mg in a slow one, a 22.6-fold spread. That is why “no coffee after 2pm” fails as advice. Below: the verified numbers, our modelled clearance table, and a four-step protocol.
This piece is the input layer under our work on sleep architecture and cognitive performance and energy auditing. Those tell you how to protect and measure the resource. This one covers the single lever most people pull daily without thinking.
The two numbers that do not agree
Start with the conflict, because everything else follows from it.
EFSA’s 2015 scientific opinion, the reference document for caffeine safety across the European Union, concluded that single doses of caffeine up to 200 mg, which it puts at about 3 mg per kilogram for a 70 kg adult, do not give rise to safety concerns, and that habitual consumption up to 400 mg across 24 hours is unlikely to cause harm in healthy adults. The United States Food and Drug Administration cites the same 400 mg daily figure.
The International Society of Sports Nutrition’s 2021 position stand, the most cited synthesis of caffeine and performance, states that caffeine has consistently been shown to improve exercise performance at doses of 3 to 6 mg per kilogram of body mass, most commonly taken 60 minutes before exercise, and that very high doses such as 9 mg per kilogram bring side effects without added benefit.
Put a 70 kg adult in both documents. EFSA’s cleared single dose is 200 mg. The ISSN’s upper performance dose is 420 mg. That is 2.1 times the level the safety opinion actually examined for a single sitting.
Neither body is wrong. They answer different questions: EFSA asked what single dose raises no safety concern, the ISSN asked what dose reliably improves performance, and a dose above the first is not thereby unsafe, it is simply outside what that opinion assessed. But the practical consequence is real. If you take a performance dose you are, by definition, operating past the point the safety review covered, and you should be deliberate about it rather than arriving there by accident at 4pm.
What a single dose actually costs your sleep
The best available evidence on timing is a 2024 randomised, double-blind, placebo-controlled crossover trial published in Sleep, the journal of the Sleep Research Society. Twenty-three men with moderate habitual intake, under 300 mg per day, each completed all seven conditions: placebo, plus 100 mg and 400 mg taken 12, 8 and 4 hours before bed, with 48 hours of washout between them. Sleep was measured with in-home partial polysomnography, not a wearable and not a diary.
Verified data. Effect of 400 mg caffeine versus placebo on measured sleep, by hours before bedtime
| Outcome | 12 hours before | 8 hours before | 4 hours before |
|---|---|---|---|
| Total sleep time | -30.0 min (not significant, p=.060) | -28.7 min (not significant, p=.076) | -50.6 min (p<.001) |
| Sleep efficiency | no significant effect | -6.9 percentage points (p=.001) | -9.5 percentage points (p<.001) |
| Latency to persistent sleep | +15.3 min (p=.020) | no significant effect | +25.4 min (p<.001) |
| Wake after sleep onset | no significant effect | +29.1 min (p=.001) | +26.2 min (p=.002) |
| Deep (N3) sleep duration | -20.6 min (p=.001) | -15.3 min (p=.016) | -29.7 min (p<.001) |
| Perceived sleep quality | no significant effect | no significant effect | -34.0% (p=.006) |
| Any outcome at 100 mg | no significant effect at any timing (p>.05) |
Read the last two rows together, because that is the finding almost nobody reports. Deep sleep was significantly reduced at every single timing tested, including 12 hours before bed. Perceived sleep quality was significantly worse only at 4 hours. The authors state the implication directly: individuals may have difficulty accurately perceiving the influence of caffeine on sleep quality.
Twelve hours before an 11pm bedtime is 11am. A single large coffee at eleven in the morning measurably cut deep sleep that night, and the people it happened to could not feel it.
Original CEOtudent analysis. The detectability gap, derived from the trial’s own results
| Hours before bed | Deep sleep measurably reduced | Sleeper noticed | Gap |
|---|---|---|---|
| 12 | Yes, -20.6 min | No | Damage invisible |
| 8 | Yes, -15.3 min | No | Damage invisible |
| 4 | Yes, -29.7 min | Yes, -34.0% rating | Damage visible |
Two of the three timings sit in a blind spot. If you have ever concluded that afternoon coffee does not affect you because you slept fine, this table is the reason that conclusion is not evidence.
The good news in the same trial: 100 mg, roughly a small cup, produced no significant effect on objective or subjective sleep at any of the three timings, including four hours before bed. Dose matters more than time of day.
Why cutoff times fail: the half-life you cannot see
EFSA reports the plasma half-life of caffeine in healthy adults as a mean of about 4 hours with a range of about 2 to 8 hours. The ISSN position stand reports the average as 4 to 6 hours with individual values ranging from 1.5 to 10 hours. Over 95% of caffeine is cleared by a single liver enzyme, CYP1A2, whose activity varies widely between people.
Nobody publishes what that variation does to your bedtime exposure, so we computed it. The maths is standard first-order elimination: what remains equals the dose multiplied by one half raised to the power of hours divided by half-life. We used EFSA’s own mean and range.
Original CEOtudent analysis. Modelled caffeine remaining at bedtime, in mg, from EFSA’s published half-life parameters
| Dose taken | Half-life 2h (fast) | Half-life 4h (population mean) | Half-life 8h (slow) |
|---|---|---|---|
| 100 mg, 4 hours before bed | 25.0 | 50.0 | 70.7 |
| 100 mg, 8 hours before bed | 6.3 | 25.0 | 50.0 |
| 200 mg, 4 hours before bed | 50.0 | 100.0 | 141.4 |
| 200 mg, 8 hours before bed | 12.5 | 50.0 | 100.0 |
| 400 mg, 4 hours before bed | 100.0 | 200.0 | 282.8 |
| 400 mg, 8 hours before bed | 25.0 | 100.0 | 200.0 |
| 400 mg, 12 hours before bed | 6.3 | 50.0 | 141.4 |
This is a model, not a measurement: it assumes clean first-order elimination and ignores absorption time, food, oral contraceptives, smoking and pregnancy, all of which shift clearance. Use it for the ratio, not the decimal.
The ratio is the point. The same 400 mg, twelve hours before bed, leaves 6.3 mg in a fast metaboliser and 141.4 mg in a slow one. That is a 22.6-fold difference in exposure from identical behaviour. Set EFSA’s sleep-relevant threshold of 100 mg against the table and the rule falls out: after 400 mg, a fast metaboliser is under it within 4 hours, the average person needs 8, and a slow metaboliser needs 16.
Sixteen hours. There is no time of day at which a slow metaboliser can take 400 mg and be clear by bedtime.
This also explains the trial’s strangest result. At the population mean half-life, 400 mg twelve hours out leaves 50 mg, well under EFSA’s threshold, and yet deep sleep still fell significantly. Either the effect was carried by the slower-clearing participants in that sample of 23, or residuals below 100 mg still disturb sleep architecture. We cannot resolve which from the published data, and the practical instruction is the same either way: your personal half-life, not the clock, sets your cutoff.
Tolerance: the part that undoes the daily habit
This is where the evidence gets uncomfortable for regular drinkers, so we will report it precisely rather than dramatically.
A randomised, double-blind, parallel-groups study published in Neuropsychopharmacology in 2010 tested 379 people: 162 non or low consumers and 217 medium or high consumers. All were kept off caffeine for 16 hours, then given 100 mg followed by another 150 mg ninety minutes later, or placebo on both occasions, and rated anxiety, alertness and headache throughout.
The results, quoted from the authors’ own conclusion: placebo administration in the medium and high consumers decreased alertness and increased headache; caffeine did not increase alertness in the non and low consumers; and with frequent consumption, no net benefit for alertness is gained, as caffeine abstinence reduces alertness and consumption merely returns it to baseline.
In plain terms: for a habitual drinker, the morning coffee is not lifting you above baseline. It is removing an overnight withdrawal deficit that the coffee itself created. The alertness you feel is real, but it is a repair, not a gain. Meanwhile the person who rarely drinks coffee gets no alertness lift from a dose either, which is the other half of the same finding and the half that gets left out of headlines.
We will not overstate this. The ISSN position stand reviews the same withdrawal-reversal question for physical performance and reaches a different verdict there: it cites a study in which 3 mg per kilogram improved cycling performance in trained cyclists regardless of a four-day withdrawal period, and another finding improved endurance in habitual users regardless of 0, 2 or 4 days of abstinence. So the honest summary is that tolerance appears to erode the everyday alertness benefit while leaving much of the acute physical performance benefit intact.
That distinction is the whole decision. Daily caffeine buys you very little on a normal day and costs you deep sleep. Occasional caffeine, deployed against a specific demand, buys you a great deal.
The four-step caffeine protocol
CEOtudent editorial framework. Managing caffeine as an input rather than a habit
| Step | The rule | The evidence behind it |
|---|---|---|
| 1. Set a ceiling, not a schedule | Cap total daily intake at 400 mg and any single dose at 200 mg unless you are deliberately dosing for performance | EFSA and FDA daily figure; EFSA single-dose no-concern level |
| 2. Find your own half-life before setting a cutoff | Run 400 mg once, 8 hours before bed, and once 12 hours before, and compare. If the 12-hour night is measurably worse you are on the slow side | 22.6-fold modelled spread; EFSA half-life range 2 to 8 hours |
| 3. Spend the big dose on purpose | Reserve 3 to 6 mg per kilogram for a specific demand: a hard training session, a genuinely sleep-deprived day, a deadline that will not move | ISSN ergogenic range, 60 minutes before the demand |
| 4. Protect the deep sleep window | After midday, treat 100 mg as your working unit, not 400 mg | 100 mg produced no significant sleep effect at any timing; 400 mg cut deep sleep at all three |
Two practical notes on step 1. The FDA’s own guidance page says 400 mg is “about two to three 12-fluid-ounce cups of coffee” and, in a table on the same page, gives regular brewed coffee as 113 to 247 mg per 12 fluid ounces. At the top of the FDA’s own range, two cups come to 494 mg, already 23% over the daily figure the same page recommends. The cup heuristic only holds at the weak end of the range, and you cannot see which end you are drinking. Weigh the dose, not the cups.
And on step 3: a performance dose is a withdrawal you will pay for the next morning if you take one daily. That is the trade the tolerance evidence describes, and it is only worth making when the demand is real.
Where a CEO and a student read this differently
A CEO reads this as a resource-allocation problem, which it is. Caffeine is a loan against tomorrow’s sleep, and deep sleep is the input to the cognitive work you are borrowing to fund. Borrowing 200 mg of it to finish something that matters is a good trade. Borrowing it every day at 3pm out of habit is a standing overdraft against an asset you rely on, taken out without a decision.
A student reads it as a measurement problem, which it also is. The single most important finding here is that at two of three timings the damage was real and invisible. Your subjective sense of how you slept did not detect a 20 minute loss of deep sleep. Any protocol built on how you feel will be wrong in the direction that favours the habit. Run step 2, write down what you find, and let the data override the feeling.
Frequently asked questions
How late can I drink coffee?
There is no universal hour, and that is the finding rather than a dodge. The same 400 mg leaves 6.3 mg or 141.4 mg at bedtime twelve hours later depending on your clearance rate, a 22.6-fold spread. What does generalise is dose: 100 mg produced no significant sleep effect at any timing tested, including four hours before bed.
Is 400 mg a day safe?
EFSA and the FDA both cite 400 mg across 24 hours as an amount not generally associated with harm in healthy adults, with 200 mg as the single dose EFSA cleared. Those figures exclude pregnancy, where EFSA notes half-life lengthens considerably, and they are population figures, not personal ones.
Does caffeine actually improve performance or just remove withdrawal?
Both answers are correct for different outcomes. For everyday alertness in habitual drinkers, the 2010 Neuropsychopharmacology trial concluded there is no net benefit, only a return to baseline. For acute exercise performance, the ISSN position stand cites studies showing the effect survives multi-day withdrawal periods in habitual users. Alertness tolerates; physical ergogenics appear to tolerate less.
Should I quit caffeine entirely?
Nothing here supports that. The evidence supports moving from unconscious daily use to deliberate occasional use, so the tool still works when you need it. Note also that non and low consumers in the 2010 trial got no alertness increase from a dose either, so abstinence is not itself a performance strategy.
Does the timing advice change if I sleep badly already?
It matters more, not less. The trial’s largest single effect was a 50.6 minute loss of total sleep at four hours, on top of a 29.7 minute loss of deep sleep, in healthy sleepers with good baseline sleep quality. If your sleep is already short, the same dose eats a larger share of it.
Do wearables tell me if caffeine is hurting my sleep?
Treat them as a trend, not a verdict. The trial’s measurements came from in-home polysomnography, which measures sleep stages directly. Consumer devices estimate them, and we cover what those estimates can and cannot support in our piece on readiness scores.
Sources
- European Food Safety Authority, Panel on Dietetic Products, Nutrition and Allergies, Scientific Opinion on the safety of caffeine, EFSA Journal 2015, volume 13, issue 5, article 4102. Single-dose and daily intake conclusions; the 100 mg statement on sleep latency and duration; plasma half-life mean of about 4 hours with a range of about 2 to 8 hours.
- Guest and colleagues, International Society of Sports Nutrition position stand: caffeine and exercise performance, Journal of the International Society of Sports Nutrition, 2021, volume 18, article 1. Ergogenic dose range of 3 to 6 mg per kilogram; 60 minute timing; CYP1A2 metabolism; withdrawal and habituation evidence.
- Dose and timing effects of caffeine on subsequent sleep: a randomized clinical crossover trial, Sleep, 2024, article zsae230. Twenty-three participants, seven conditions, in-home partial polysomnography; objective and subjective sleep outcomes at 12, 8 and 4 hours before bedtime.
- Rogers and colleagues, Association of the anxiogenic and alerting effects of caffeine with ADORA2A and ADORA1 polymorphisms and habitual level of caffeine consumption, Neuropsychopharmacology, 2010, volume 35, issue 9. Randomised double-blind parallel groups, 379 participants, on tolerance and net alertness benefit.
- United States Food and Drug Administration, consumer update on caffeine intake, last updated August 2024. The 400 mg daily figure and the typical caffeine content table for 12 fluid ounce servings.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
