Caffeine can quietly cut your deep sleep, even when you feel fine

Coffee has a reputation problem that is easy to dismiss: everyone knows a person who drinks it at 9 p.m. and sleeps fine. A pair of 2026 publications on caffeine and sleep EEG make the case that "sleeps fine" is not the whole story, because the part of sleep caffeine disrupts most is not the part people can feel.

The finding

A systematic and mechanistic review published in the journal Nutrients, by James Chmiel and Donata Kurpas, pulled together decades of human sleep-EEG research on caffeine and asked what it shows once total sleep time and self-reported rest are set aside. The most consistent finding across the studies it covers is suppression of slow-wave activity, the low-frequency brain waves (roughly 1 to 4.5 Hz) that define the deepest, most restorative stage of non-REM sleep. Caffeine also tends to increase faster, lighter-sleep brain wave activity, a shift the review describes as pushing sleep toward more wake-like cortical dynamics even when someone stays asleep the whole night.

A separate, more tightly controlled body of evidence backs this up with harder numbers. A meta-analysis of 22 controlled crossover trials with 956 participants, published in Sleep Medicine in late 2025, pooled studies where caffeine or a placebo was given under lab conditions and sleep was measured by polysomnography. Compared with placebo, caffeine reduced total sleep time by a mean of 34.67 minutes, cut sleep efficiency by 4.74 percentage points, reduced the proportion of slow-wave sleep by 1.01 percentage points, and delayed sleep onset by a mean of 8.35 minutes. Effects were larger in younger adults and at higher doses. In one earlier study the review draws on, of habitual caffeine users aged 10 to 16, slow-wave activity at the start of the night ran about 20% lower, give or take 9 percentage points, than in non-users.

What this design can and cannot tell you

The 956-participant meta-analysis is built from crossover trials: the same people were given caffeine and placebo on different nights, in a lab, with sleep measured directly rather than self-reported. That design supports causal language, caffeine reduced slow-wave sleep and total sleep time in these trials, rather than the weaker "linked to" that observational data would call for.

The broader EEG review is a different kind of evidence. It synthesizes many smaller studies that used different doses, timing, populations and recording methods, which is why its own language stays at the level of "the most consistent finding" rather than a single pooled number. The 20% figure specifically comes from one included study of adolescents and should not be read as caffeine's effect on adults in general, it is cited here as an illustration of how large the effect can look in EEG terms, not as a global average.

Neither piece of evidence says caffeine wakes people up in a way they would notice. That is closer to the opposite of what both found: total sleep time and sleep efficiency moved by amounts most people would not flag on their own, and the deepest sleep stage was thinner without any obvious daytime complaint attached to it in the underlying trials.

How much caffeine is in what you drink

Content varies by brand, bean, brew strength and serving size. Amounts below are per realistic serving, ordered by density:

• Cold brew coffee, 16 oz: 200 to 260 mg
• Energy shot, 2 oz: about 200 mg
• Drip coffee, brewed, 8 oz: 96 mg
• Energy drink, 8.4 oz can: 70 to 100 mg, some larger cans run higher
• Matcha green tea, 8 oz: 60 to 70 mg
• Espresso, single shot, 1 oz: 63 mg
• Instant coffee, 8 oz: 62 mg
• Black tea, brewed, 8 oz: 47 to 48 mg
• Yerba mate, 8 oz: 40 mg
• Cola, regular or diet, 12 oz can: 30 to 45 mg
• Green tea, brewed, 8 oz: 29 mg
• Dark chocolate, 60 to 69% cacao, 1 oz: 24 mg
• Bottled iced tea, 8 oz: 18 to 25 mg, brand-dependent
• Semisweet chocolate, 1 oz: 12 mg
• Milk chocolate, 1 oz: 6 mg
• Decaf coffee, 8 oz: 2 to 7 mg

A single 16 oz cold brew can already sit at or above half the general daily ceiling described below, before anything else in the day is counted.

How much you need

Caffeine is not an essential nutrient, so there is no RDA or AI for it, only intake guidance:

• Healthy, non-pregnant adults: up to 400 mg a day, about four 8 oz cups of brewed coffee, is the general FDA and Mayo Clinic ceiling for most people.
• Pregnancy: the American College of Obstetricians and Gynecologists recommends staying under 200 mg a day.
• Children up to age 12: Health Canada's guidance caps intake at 2.5 mg per kilogram of body weight a day.
• Anyone sensitive to caffeine's effect on sleep, anxiety or heart rhythm: lower than the general ceiling, and earlier in the day, matters more than the total.

Too little, and too much

Cutting back is not a deficiency, since the body needs no caffeine, but habitual users who reduce or stop often get withdrawal: headache in roughly half of people who quit, along with fatigue, low mood, irritability and trouble concentrating. Symptoms typically start 12 to 24 hours after the last dose, peak within the first one to three days, and resolve within two to nine days.

Too much, above the roughly 400 mg general ceiling, is associated with nervousness, irritability, sleeplessness, increased urination, abnormal heart rhythms and dyspepsia. Below that ceiling, the sleep-specific effects described above still show up: the crossover trials in the Sleep Medicine meta-analysis measured real reductions in slow-wave sleep at doses and timings well within what many people drink on an ordinary day, which is a separate question from whether a given amount is "safe" in the FDA sense.

Tracking it in AIM

AIM does not track caffeine as standard, so add it as a custom metric: give it a name, a unit (mg) and a daily target of under 400 mg, and it is estimated on every meal from then on, pinned to your cards or left behind "more". Logging afternoon and evening coffee separately from a morning cup makes it easier to see how close a given day's timing runs to the 8-to-12-hour buffer sleep researchers recommend before bed.

Estimates from a photograph carry real uncertainty for micronutrients, because content varies with soil, cultivar, storage and cooking in ways no image contains. Treat the week as meaningful and any single day as approximate.