Rubric condition met: C, direction mixed: food and drink results do not transfer to extracts
A 2026 GRADE-assessed meta-analysis of 16 RCTs (2,847 participants) found faster Trail Making Test times and better verbal fluency with regular cocoa, and a 2023 meta-analysis of 7 trials found faster executive-function times. COSMOS-Mind and COSMOS-Clinic found no effect on executive function composites in older adults. Evidence is mixed; B would apply on the meta-analysis condition alone, so the grade is C. Re-appraised November 14, 2026: pooled positives come from chocolate and cocoa drinks, while the large extract trials (COSMOS-Mind, n = 2,262, and COSMOS-Clinic, n = 573, 500 mg flavanols) were null on cognition. Results do not transfer across forms.
How sure are we?
Evidence maturity: Emerging Pooled positives come from chocolate and cocoa drinks; the large extract trials (COSMOS-Mind, n = 2,262; COSMOS-Clinic, n = 573) were null on cognition.
Tested as: Cocoa flavanols (foods, drinks or flavanol-dosed extracts)
Who paid for the studies
Mostly industry-funded
Risk of bias
Some concerns. COSMOS: Mars Edge grant and pill donation (counted as industry funding); its null results are not discounted.
Trial registry check
Checked, no issue found. COSMOS registered (NCT02422745).
Outcome measured
A test score, rating scale or performance measure
Who was studied
Mixed healthy and clinical, pooled
Doses used in studies
500 mg flavanols a day in COSMOS
Rules applied
Form does not transfer (logged, no further change): Food and drink results do not transfer to extracts without a subgroup analysis by form
Could rise: Would move to C benefit with an adequately sized RCT of a defined cocoa-flavanol extract, positive on a prespecified executive-function primary.
Could fall: Would move to N with a second large extract trial null on executive function as its primary outcome.
Science is provisional. This grade reflects the evidence we found as of . Next review: .
Maturity sits beside each grade. The grade says what the evidence shows; maturity says how much of it there is and how well it is confirmed.
Established
many trials from several independent teams, including at least one registered in advance
Emerging
two or more trials, or one large registered trial, not yet independently confirmed
Early signal
one trial, small pilots, or results only on secondary outcomes
Mechanistic only
lab, animal or biomarker data, or uncontrolled reports; no controlled human results
each study this grade rests on, by publication year and design. Solid: effect versus the comparator; half-filled: mixed; hollow: no effect; dashed: result not recorded yet. Larger marks had more participants.Show the numbers
Study
Year
Design
Participants
Result
Baker 2023
2023
RCT
2262
result not recorded
Shateri 2023
2023
meta-analysis
not recorded
result not recorded
Vyas 2024
2024
RCT
573
result not recorded
Arabpour 2026
2026
meta-analysis
2847
result not recorded
Effect of cocoa extract supplementation on cognitive function: results from the clinic subcohort of the COSMOS trial.
Vyas CM, Manson JE, Sesso HD. Effect of cocoa extract supplementation on cognitive function: results from the clinic subcohort of the COSMOS trial Am J Clin Nutr, 2024. PMID 38070683
Result not recorded
Design
Randomized, placebo-controlled trial, 2-year in-person cognitive subcohort of COSMOS (COSMOS-Clinic)
Sample
n = 573
Population
Adults aged 60 and over (mean age 69.6)
Funding
NIH grants (NHLBI, NIA, NEI, NCATS, WHI); Crossref funder list also names Mars Edge Cocoa Flavanol Science Hub
Effects of cocoa extract and a multivitamin on cognitive function: A randomized clinical trial.
Baker LD, Manson JE, Rapp SR. Effects of cocoa extract and a multivitamin on cognitive function: A randomized clinical trial Alzheimers Dement, 2023. PMID 36102337
Result not recorded
Design
Randomized, placebo-controlled 2x2 factorial pragmatic trial, 3 years (COSMOS-Mind)
Sample
n = 2,262
Population
Older adults (mean age 73), cognition tested by telephone
Funding
National Institute on Aging (R01AG050657); parent COSMOS trial supported by an investigator-initiated grant from Mars Edge (Mars, Inc.), including study pills
Effects of chocolate on cognitive function in healthy adults: A systematic review and meta-analysis on clinical trials.
Shateri Z, Kooshki A, Hormoznejad R. Effects of chocolate on cognitive function in healthy adults: A systematic review and meta-analysis on clinical trials Phytother Res, 2023. PMID 37211619
Result not recorded
Design
Systematic review and meta-analysis of 7 randomized controlled trials
Population
Healthy adults
Funding
Ahvaz Jundishapur University of Medical Sciences
Effect of Chronic Dark Chocolate or Cocoa Intake on Cognitive Function: A Grading of Recommendations Assessment, Development, and Evaluation (GRADE)-Assessed Systematic Review and Meta-Analysis of Randomized Controlled Trials.
Arabpour Z, Arabpour Z, Jazayeri S. Effect of Chronic Dark Chocolate or Cocoa Intake on Cognitive Function: A Grading of Recommendations Assessment, Development, and Evaluation (GRADE)-Assessed Systematic Review and Meta-Analysis of Randomized Controlled Trials Nutr Rev, 2026. PMID 42641106
Result not recorded
Design
Systematic review and meta-analysis of 16 randomized controlled trials (GRADE-assessed; PROSPERO CRD42024568294)
Sample
n = 2,847
Population
Adults in chronic cocoa or dark chocolate trials (mixed healthy, older and clinical samples per included trials)
Funding
Vice Chancellor for Research and Technology, Iran University of Medical Sciences (grant 30885)
Search and history
Search (): (cocoa[ti] OR flavanol*[ti] OR cacao[ti] OR chocolate[ti]) AND (cognit*[tiab] OR memory[tiab]) AND (meta-analysis[pt] OR systematic review[pt])
Graded C (awaiting approval)
C to C (approved): Editor sign-off November 13, 2026: grade matches the editor-approved final page
C to C (approved) (maturity: emerging): Re-appraisal November 14, 2026, pooled positives come from chocolate and cocoa drinks, while the large extract trials (COSMOS-Mind, n = 2,262, and COSMOS-Clinic, n = 573, 500 mg flavanols) were null on cognition. Results do not transfer across forms. Maturity, bias summary and re-check date added
A, strong: what grade A means
At least one meta-analysis or three independent randomized trials in the relevant healthy population, pointing the same way, with low or some risk of bias, at doses legally available.
