Cocoa Flavanols and Memory: What COSMOS Actually Found

The claim

The claim, in its usual form, is that flavanols in cocoa improve blood flow to the brain, that this translates into better memory, and that a daily cocoa extract is therefore a way to keep the mind sharp with age. It is often accompanied by an image of dark chocolate and an invitation to feel virtuous about eating it.

The claim is unusually respectable by the standards of this industry. It has a plausible mechanism, real published trials behind it, and a branded ingredient with academic collaborators. It is not a fringe assertion or a fabricated study. It is simply, as tested, wrong on the cognitive endpoint — and it is more useful to explain a well-supported claim that failed than to knock down an obviously bad one.

Where it came from

The early literature genuinely was encouraging. Small trials from Desideri in 2012 and Mastroiacovo in 2015 reported cognitive improvements with cocoa flavanols, and Brickman’s 2014 work reported changes in dentate gyrus function — a hippocampal region associated with age-related memory change — using neuroimaging alongside cognitive testing.

That is a coherent-looking package: a plausible vascular mechanism, imaging that moves in the expected direction, and behavioral results that agree. It is also exactly the profile that has preceded failure in this field over and over. Small trials, short durations, selected populations, outcomes that were not always the pre-specified primary endpoint — these are the conditions under which promising results appear and then do not survive scale.

To the credit of everyone involved, the field did the right thing next. It ran the big trial.

What the evidence actually shows

COSMOS — the COcoa Supplement and Multivitamin Outcomes Study — is a large, long, randomized, placebo-controlled trial with a factorial design that tested cocoa extract and a daily multivitamin-mineral in the same population.³ Its cognitive sub-studies are the definitive test of this claim.

Sub-study	Participants	Intervention	Duration	Cognitive result
COSMOS-Mind	2,262	Cocoa extract, 500 mg flavanols/day	3 years	No effect on cognition
COSMOS-Web	3,500+	Cocoa extract, 500 mg flavanols/day	3 years	No effect at 1 year or across 3 years
COSMOS-Mind, multivitamin arm	2,262	Daily multivitamin-mineral	3 years	Improved global cognition
COSMOS-Web, multivitamin arm	3,500+	Daily multivitamin-mineral	3 years	Improved memory at 1 year, sustained

COSMOS-Mind followed 2,262 participants for three years with telephone-administered cognitive testing. Cocoa extract had no effect on cognition.¹

COSMOS-Web followed more than 3,500 participants for three years with web-based cognitive testing. There was no effect at the primary one-year timepoint and none across three years.²

The dose is the part worth dwelling on. COSMOS used 500 mg of cocoa flavanols per day, including roughly 80 mg of (−)-epicatechin — the compound the mechanism is built around. This was not a homeopathic pinch, not a mislabeled extract, and not a short trial. It is what an adequate test looks like, and the answer was no.

The comparison inside the same trial makes the result harder to explain away. In the multivitamin arm of the identical population, the same investigators using the same instruments found a positive, replicated cognitive effect.¹⁴ The trial was capable of detecting a cognitive benefit. It detected one. It was not the cocoa.

The one surviving finding, in proportion

Participants with low baseline dietary flavanol intake showed a possible benefit. That is the only positive cocoa signal to come out of COSMOS, and it deserves to be described accurately rather than either buried or promoted.

It is a secondary, subgroup finding in a trial whose primary result was null. Subgroup analyses in large trials generate hypotheses; they do not establish effects, because with enough subgroups some will separate by chance. The finding is biologically coherent — it is the same “correcting a shortfall rather than adding a drug” logic that plausibly explains the multivitamin result in the same trial — and coherent is not the same as demonstrated.

What it justifies is a follow-up trial recruiting only low-flavanol-intake participants. What it does not justify is selling cocoa extract to everyone on the strength of a subgroup that most buyers are not in.

What cocoa flavanols do have evidence for

Blood pressure and endothelial function. The cardiovascular literature on cocoa flavanols is reasonably good, and nothing here is meant to dismiss it. That is a cardiovascular finding, and it stays a cardiovascular finding — a change in vascular measures is not a change in memory, and COSMOS is the demonstration of why that step cannot be assumed.

This is the same distinction we apply everywhere on this site: mechanism is not effect. Resveratrol reproduces cerebral blood flow changes and does not reproduce cognitive benefit. Ginkgo has real vascular pharmacology and failed two enormous prevention trials.⁷⁸ Omega-3 has excellent observational data and randomized trials at 500–2,000 mg/day that found no cognitive improvement in healthy older adults or in established Alzheimer’s disease.⁹ The pattern is the field’s most reliable feature. See how we rate evidence.

Why the claim persists

Four reasons, none of them conspiratorial.

The early studies are still citable. Desideri 2012, Mastroiacovo 2015 and Brickman 2014 were published, peer-reviewed and are not retracted. A marketer can cite them accurately and never mention what happened when the claim was tested at scale. Citing real papers selectively is harder for a reader to catch than inventing one.

Null results travel badly. A trial that finds nothing generates one round of coverage and disappears; the earlier positive results have had over a decade to spread into articles, listicles and product pages that nobody goes back to update.

People want it to be true. A claim that chocolate is good for the brain is the rare health message that asks nothing of anyone. Motivated reasoning does the rest, on both sides of the transaction.

The cardiovascular evidence provides cover. “Studies show cocoa flavanols support healthy blood flow” is true and legally safe, and readers complete the inference to memory themselves — which is precisely what the sentence is engineered to do. Enforcement in this category tends to arrive only when a marketer states the cognitive claim outright, as the FTC’s case against the makers of Prevagen illustrates.¹⁰

What to do instead

If you like chocolate, eat chocolate. Just do not count it as a cognitive intervention. Most chocolate products contain trivial amounts of flavanols anyway — alkalized or Dutch-processed cocoa destroys them — alongside substantial sugar. The bar in the shop is not a low dose of the COSMOS supplement; it is a different product entirely.

If you want the thing COSMOS found that worked, it was the daily multivitamin-mineral: improved global cognition in COSMOS-Mind and improved memory in COSMOS-Web, sustained over three years, in adults over 65, with larger benefits in participants with cardiovascular disease history.¹⁴⁵ The effect sizes are small in absolute terms and the mechanism is not established. It is still the only large randomized cognitive win in this entire field, and it costs a few dollars a month. We compare it directly against a stack in multivitamin vs nootropic stack.

If you want the strongest intervention available, it is not sold in a bottle at all — see exercise and memory. And if you are evaluating any brain supplement, our tier list and the case against brain supplements are the two pages we would read first.

What would change our mind

A randomized trial recruiting exclusively adults with low dietary flavanol intake — the one subgroup COSMOS flagged — powered adequately, running at least three years, with cognition as the pre-registered primary endpoint rather than a secondary analysis. If that trial found a benefit and it replicated, cocoa flavanols would move from “tested and failed” to “works in a defined deficiency-like population,” which is a genuinely different claim. Nothing short of that should shift the conclusion, because the general-population question has already been answered in over 5,700 people.

Frequently asked questions

Is dark chocolate good for memory?

There is no evidence that it is. The trial that tested the active compounds properly, at 500 mg of flavanols daily for three years, found no cognitive effect. Most chocolate contains far less flavanol than that, particularly if the cocoa has been alkalized, plus a lot of sugar.

Did COSMOS use a high enough dose?

500 mg of cocoa flavanols per day including about 80 mg of (−)-epicatechin, for three years, is a substantial and sustained dose — well above what any diet delivers. Underdosing is the usual explanation offered for null supplement trials, and it does not apply here.

What about the studies that found cocoa improved cognition?

They exist and they were not fraudulent. They were small, short and preceded the large randomized test. When a small positive literature meets a properly powered multi-year trial and the trial is null, the trial is the better estimate.

Are cocoa flavanols useless then?

Not useless — just not cognitive. They have reasonably good evidence for blood pressure and endothelial function. That is a cardiovascular claim and it should be described as one rather than being allowed to imply a memory benefit that was tested and did not appear.

Related reading

• A daily multivitamin vs a nootropic stack
• Ginkgo biloba: two enormous trials, two failures
• Prevagen and the apoaequorin verdict
• Do nootropics actually work? An honest tier list
• Exercise and memory: the strongest intervention nobody sells
• Do brain training apps work? The FTC has opinions