Creatine for the Brain: What Question Are We Actually Asking?
Creatine monohydrate is usually discussed as a supplement for muscle strength and high-intensity exercise. The question examined here is narrower and more demanding: can taking creatine change measurable cognitive performance in adults, and how much confidence should we place in the studies that say it can? The answer matters because a biological possibility can become a public promise very quickly. A supplement that might support brain energy is not automatically a treatment, a memory drug, or a universal way to think better.
This series follows that distinction from the proposed mechanism to the human evidence. It examines what researchers measured, what a 2024 systematic review and meta-analysis reported, and why a 2026 commentary argues that part of the statistical foundation may be weaker than the headline suggests. The practical goal is not to produce a simple yes-or-no verdict. It is to learn how to recognize a plausible, partly positive, but still contested evidence claim.
The supplement and the question
Creatine is a nitrogen-containing compound made in the body and obtained through diet. The form studied in the supplied evidence is creatine monohydrate. That detail matters: evidence about one formulation should not silently become evidence about every product marketed under the broader name “creatine.” The studies also concern adults, not children, and they assess cognitive performance through defined tests rather than through an open-ended sense of feeling mentally sharper.
Those tests cover several domains. Memory asks whether information can be retained and recalled. Attention can be represented by a score or by the time required to complete a task. Processing speed likewise may be expressed as accuracy, completion time, or another test-specific measure. Executive function refers to abilities such as cognitive flexibility, inhibition, and planning, but the particular tests used to represent it are not interchangeable. “Cognition” is therefore not one outcome. It is a family of outcomes, each with its own measurement choices and sources of noise.
That is the first question to ask whenever a supplement is said to improve the brain: improve which measured function, under which conditions, compared with what, and for whom? A result on a memory subtest is not the same claim as better learning at work. A faster reaction-time task is not the same claim as better judgment. The more general the conclusion becomes, the more evidence is needed to connect it to the original measurement.
The comparison also matters. A randomized trial is not simply a before-and-after story in which participants take creatine and report how they feel. It compares an intervention group with a control condition and tries to make the groups differ mainly in the intervention. The review’s question is consequently about average differences between groups on defined tests, not about whether every participant notices a subjective mental change. Even that controlled comparison cannot remove every source of uncertainty: studies can be small, tests can vary in sensitivity, and participants can differ in baseline health, diet, age, and cognitive performance.
The population is part of the result, not a footnote. The review includes adults, mostly healthy people, and a smaller set of participants with specific illnesses. A signal in one of those groups may not travel unchanged to older adults, people with a different condition, or people who were not represented at all. The same rule applies to the intervention: the supplied evidence studies creatine monohydrate, so a conclusion about that form should not be stretched to every formulation, dose, or commercial combination.
Finally, a cognitive endpoint is a choice about what counts as improvement. A test can reward speed, accuracy, recall, inhibition, or flexibility, and a participant can improve on one dimension without improving on all the others. That is why the language of the evidence must stay close to the test. The useful question is not merely whether creatine “works,” but which result appeared, in which participants, under which comparison, and with what uncertainty.
The evidence chain can be kept visible from the beginning:
This chain also prevents a common shortcut. The fact that creatine participates in cellular energy metabolism is a reason to test its cognitive effects, not a result showing that the effects exist. The result still depends on how much of the compound reaches relevant tissue, what task is measured, how the comparison is designed, and how the data are combined.
Why the answer is not a single yes or no
The 2024 review gathered randomized controlled trials in adults and analyzed several cognitive domains. Its headline findings were selective. The pooled result for memory was statistically positive. The review also reported faster completion of attention and processing-speed tasks. But it did not find statistically significant improvements in overall cognitive function, executive-function scores, executive-function time, attention scores, or processing-speed scores.
That mixture is more informative than a single promotional sentence. It suggests that any signal, if real, may depend on the kind of task being used. A time-based outcome asks whether participants complete a task more quickly; a score-based outcome asks something different. Results can move in different directions without contradicting one another, but they cannot be collapsed into the claim that every important aspect of cognition improves.
The review itself also described uneven certainty. It rated the evidence for memory as moderate certainty, while the evidence for processing speed, overall cognition, executive function, and attention was lower. “Moderate certainty” is not the same as proof, and “low certainty” is not proof of no effect. These labels describe how much trust the reviewers placed in the available body of evidence after considering factors such as bias, consistency, precision, directness, and publication bias.
The later challenge concerns the structure of that body of evidence. The 2026 commentary points out that some trials supplied several cognitive outcomes from the same participants. In the memory analysis, the reported number of observations was larger than the number of unique randomized participants, and individual trials contributed multiple memory subtests. If those correlated outcomes are treated as independent observations, the analysis can behave as though it has more independent evidence than it really does. The apparent estimate may become more precise than the participant-level data justify.
The scale of the discrepancy is visible even before a reader attempts a reanalysis. The 2024 paper describes 16 studies and 492 participants in the review as a whole, yet describes the memory synthesis as involving 24 studies and 1,000 participants. Some of that difference may reflect how outcomes were organized, but it is precisely why a study count, a participant count, and an outcome count must not be treated as synonyms. The later commentary identifies the same underlying concern: several test results can originate from one randomized group of people.
This is also why standard quality checks need to be read for what they actually test. Risk-of-bias judgments, sensitivity analyses, and tests for publication bias can be useful, but they do not automatically answer whether correlated outcomes were entered as independent evidence. A result can look stable when one study is removed and still rely on an incorrect unit of analysis. The statistical question is not an ornamental technicality; it determines how much weight the pooled estimate deserves.
This does not settle the question in the opposite direction. A statistical criticism can show that confidence in a pooled result should be reduced without proving that creatine has no cognitive effect. The provided sources do not contain a corrected reanalysis of the 2024 data, so they cannot tell us what the memory estimate would be after the proposed correction. The responsible description is therefore neither “creatine definitely improves memory” nor “the supplement has been disproved.” It is that a positive result was reported, and its statistical significance and precision are contested.
What a defensible answer must keep separate
Four distinctions will govern the rest of the discussion. The first is biological plausibility versus observed outcome. The proposed brain-energy role explains why researchers might expect an effect. It does not replace a controlled comparison in human participants.
The second is an endpoint-specific result versus general cognitive enhancement. The review’s positive findings concern particular tests or completion times. They do not establish improvements in productivity, academic achievement, decision-making, quality of life, or the prevention of dementia. Those are important outcomes, but they were not supplied by these sources as demonstrated consequences of creatine supplementation.
The third is a reported estimate versus a secure estimate. Meta-analysis can make a scattered literature easier to interpret, but the result is only as sound as the assumptions used to combine the studies. A polished forest plot, a confidence interval, a sensitivity analysis, or a test for publication bias cannot by itself repair a mismatch between the observations and the statistical model. The unit being counted—trial, participant, outcome, or effect size—matters.
The fourth is evidence in a studied population versus advice for an individual. The review included adults and used creatine monohydrate, with mostly healthy participants and a smaller number of studies involving specific illnesses. That is not enough to establish a treatment, a prevention strategy, a best dose, or a universally useful duration. It also does not establish that people who were not represented in the trials should expect the same result.
Keeping these boundaries is not an exercise in excessive caution. It makes the evidence more useful. A narrow claim can be tested and improved: perhaps creatine helps some adults on some demanding cognitive tasks, under some conditions. A broad claim is harder to evaluate because it hides the endpoint, the population, and the assumptions that created it.
It also gives readers a better way to handle disagreement. The 2024 review is evidence of what one synthesis reported, including its methods and its stated certainty. The 2026 commentary is evidence that a consequential assumption in that synthesis deserves re-examination. Neither source, by itself, supplies the missing corrected estimate. Until that estimate exists, confidence should be calibrated rather than forced into a binary verdict. The most honest answer can contain both a plausible signal and a serious reason to doubt how firmly it has been measured.
The next step is to examine the proposed brain-energy mechanism itself. Understanding how creatine is thought to support energy transfer can clarify why certain tasks might be sensitive—and also show exactly where a plausible mechanism stops short of proving a cognitive benefit.