Caffeine improves attention by small but real margins, finds largest meta-analysis to date

A new meta-analysis in Psychopharmacology puts the most precise numbers yet on what a cup of coffee does to your attention. Pooling 31 randomised, double-blinded, placebo-controlled trials with 1,455 participants between them, Kasper Kløve and Anders Petersen at the University of Copenhagen report a Hedges’ g of 0.27 for accuracy and 0.28 for reaction time. Both effects are statistically significant. Neither is large.

Cohen’s framework calls a g of 0.27 “small”. On a typical attention task, that puts the caffeine-dosed average roughly 58 to 61 percent ahead of placebo. Not transformative, in other words, but real enough to matter on a dull afternoon. The more interesting result is what happens when you start changing the dose. Reaction time keeps getting faster as you push the dose up. Accuracy follows an inverted U, where it climbs to a peak and then falls off again at very high doses.

What the meta-analysis included

Kløve and Petersen searched PsycINFO, PubMed, and Scopus for every randomised, double-blinded, placebo-controlled trial that pitted pure caffeine against placebo on behavioural measures of attention in rested, healthy adults. Studies using caffeine-containing beverages (coffee, tea, energy drinks) were excluded, since other compounds in those drinks can confound the result, and participants had to be non-sleep-deprived at the time of testing. The final pool came to 31 trials and 1,455 participants.

For each trial, the authors extracted every eligible attention outcome and rolled them into a single composite standardised mean difference, one for reaction time and one for accuracy. Those were then combined in random-effects meta-analyses. Subgroup analyses and meta-regressions tested whether the effect depended on dose, habitual caffeine consumption, task complexity, or which attention network the task taxed.

The inclusion criteria matter because they address a long-standing confound in caffeine research. Older studies often tested caffeine in overnight-abstained participants, who are by definition in caffeine withdrawal at baseline. When those participants improve after a dose, it is unclear whether the caffeine is genuinely enhancing performance above a normal baseline or simply reversing withdrawal. By restricting to rested, healthy adults and explicitly testing habitual consumption as a moderator, the new analysis was designed to speak to that question directly.

The dose-response split

The meta-regression on dose produced the paper’s most practically useful finding. For reaction time, the relationship was linear: more caffeine, faster responses. The effect kept scaling across the range of doses tested, with 200 mg and above outperforming lower doses.

For accuracy, the curve was quadratic. Accuracy improved as the dose moved from low to moderate, then plateaued and declined at the high end. The paper does not pin down a precise inflection point (the included studies used varying dose levels, and the meta-regression models a continuous curve), but the shape is consistent with what the authors describe as a Yerkes-Dodson-type relationship. Arousal improves performance up to an optimum, beyond which more arousal starts to degrade it.

This split goes some way to explaining the noise in older caffeine research. A study using a low dose and measuring accuracy might report a null result. A study using a high dose and measuring only reaction time might report a large one. Both can be right, because the dose-response curve is not the same for the two outcomes.

Habitual use does not erase the benefit

One of the more debated findings concerns habitual caffeine consumption. The meta-analysis found no significant moderation by how much caffeine participants typically drank. The acute attention benefit, in other words, did not differ meaningfully between people who put away several cups of coffee a day and those who rarely touch the stuff.

That finding pushes against the so-called withdrawal-reversal hypothesis, which holds that caffeine’s apparent cognitive benefits are largely an artefact of restoring performance that overnight withdrawal had degraded. If that were the whole story, regular users should show a larger acute benefit than naive users, because regular users are testing from a withdrawal-impaired baseline. The pooled data here does not support that pattern.

Kløve and Petersen are careful not to claim that withdrawal reversal plays no role at all, only that the evidence from 31 trials does not show habitual consumption moderating the effect. A 2013 study in the Journal of Psychopharmacology by Brunye and colleagues, which directly compared non-consumers to withdrawn consumers, reached a similar conclusion. Caffeine improved attention in both groups, and the improvement in the non-consumers was not smaller.

Task complexity and attention networks

The authors went looking for two more moderators and found nothing. They tested whether the boost depended on how demanding the task was. They also tested whether it mattered which of the three classic attention networks (alerting, orienting, or executive control) the task primarily recruited. Neither variable was significant. Caffeine improved performance on tedious monitoring tasks and on cognitively demanding ones, and across all three networks.

That null is useful. If you swear caffeine helps you grind through a long stretch of email but does almost nothing when you sit down to write something hard, the explanation probably is not that one task is more difficult than the other. The more likely culprits are expectation, time of day, sleep debt, where you are in the absorption curve from your last cup, and how fast you personally metabolise caffeine. Related work on cognitive supplements has hit the same wall, with effect sizes that depend more on individual variation than on task design (head-to-head analyses of bacopa monnieri and ginkgo show a similar pattern).

What 200 mg actually means

To map the dose-response onto a real day, 200 mg of caffeine is roughly two to three cups of brewed filter coffee, give or take a bean variety. A single espresso shot delivers about 60 to 80 mg. A standard 250 ml energy drink can runs 80 to 150 mg. The trials in the meta-analysis used caffeine tablets, which deliver precise doses and avoid the other compounds (chlorogenic acids in coffee, theobromine in cocoa-derived drinks, L-theanine in tea) that beverages drag along with the caffeine.

The practical takeaway from the dose-response curves: for work that calls for sustained speed, like monitoring a screen, reacting to intermittent signals, or pushing through a queue under time pressure, a higher dose within the studied range is more likely to help than a lower one. For work that demands careful discrimination, error-free performance, or fine judgment, a moderate dose is the safer bet. Very high doses, on the accuracy side of the curve, can actively backfire. Anyone weighing whether to add caffeine to a daily stack should also talk to their doctor before starting any supplement, especially if they have a cardiovascular condition or anxiety disorder.

Caveats

A few caveats are worth keeping in mind. The included trials tested acute effects only, so the analysis has nothing to say about chronic daily use, tolerance, or long-term cognitive outcomes. The participants were healthy adults, which means the findings do not automatically transfer to children, to older adults with cognitive impairment, or to clinical populations. The studies used pure caffeine and not coffee or tea, so the results describe the compound rather than the beverage matrix. And while 31 trials is a respectable haul for a meta-analysis, the dose-response meta-regression has to pool across studies that used different attention tasks, different dose timings, and different participant pools. The fitted curve is a best estimate, not a prescription.

There is also no getting around how small the effect is. A g of 0.27 is real, but it is a quarter of a standard deviation, not a phase shift. Caffeine is not a cognitive panacea. For most people on most days, what the meta-analysis is actually quantifying is the difference between a slightly sharper hour at the desk and an ordinary one. The broader evidence base on common brain-health supplements tells a similar story of modest, dose-dependent gains.

Bottom line

This is the most rigorous look yet at how acute caffeine influences attention, and at what dose. The headline summary: a small but reliable benefit to both speed and accuracy. Speed keeps gaining as the dose rises. Accuracy peaks somewhere in the middle of the studied range, then falls off again. The benefit does not seem to depend on whether you are a habitual coffee drinker, or which type of attention the task taxes. If you reach for an espresso before a hard cognitive block, the data broadly supports the habit. If the work demands precision over pace, more is not better, and the curve says so.

References

1. Kløve K, Petersen A. A systematic review and meta-analysis of the acute effect of caffeine on attention. Psychopharmacology 242(9):1909-1930. 2025. https://doi.org/10.1007/s00213-025-06775-1
2. Rogers PJ, Heatherley SV, Mullings EL, et al. Faster but not smarter: effects of caffeine and caffeine withdrawal on alertness and performance. Psychopharmacology 226(2):229-240. 2013. https://doi.org/10.1007/s00213-012-2889-4
3. Brunye TT, Mahoney CR, Lieberman HR, et al. Acute effects of caffeine on attention: a comparison of non-consumers and withdrawn consumers. Journal of Psychopharmacology 27(1):77-83. 2013. https://doi.org/10.1177/0269881112460112
4. Nehlig A. Is caffeine a cognitive enhancer? Journal of Alzheimer’s Disease 20(Suppl 1):S85-S94. 2010. https://doi.org/10.3233/JAD-2010-091315
5. Cappelletti S, Piacentino D, Sani G, et al. Caffeine: cognitive and physical performance enhancer or psychoactive drug? Current Neuropharmacology 13(1):71-88. 2015. https://doi.org/10.2174/1570159X13666141210215655