Nootropics and Cognitive Performance

01 What this document is, and five things it is not

This article asks one question of a large commercial category: when a product is sold to improve memory, focus, or long-term brain health, what has actually been measured, in whom, for how long, and against what comparator?

The category is unusually easy to write badly about. It contains compounds with genuine pharmacology, compounds that are essential nutrients, compounds whose entire evidence base is a single small trial, and compounds whose evidence base is a mechanism diagram. It also contains, as a matter of documented laboratory fact, unapproved drugs sold as dietary supplements (Cohen, Avula, and Khan, 2022). Treating those as one subject would be the error this document exists to avoid.

Five things follow immediately.

First, this is not a treatment manual. It recommends no compound, amount, route, or schedule for any person, and it does not distinguish between readers on any clinical basis.

Second, it is not an ingredient catalogue. The individual compounds have their own articles in this series — Ginkgo-Biloba, Lions-Mane, Acetyl-L-Carnitine, Pyrroloquinoline-Quinone, L-Theanine, L-Tyrosine-and-NALT, Creatine-Monohydrate, Yerba-Mate, Magnesium — and those remain the deep references. The work of a system article is synthesis: which outcomes matter, which mechanisms are plausible, which claims survive contact with a randomised trial, and which combinations exist only because they appear together on a label.

Third, it is not a defence of the word nootropic. The word is examined in section 02 and then used only in quotation marks or as a market descriptor.

Fourth, it is not a dementia article. Where a compound has been tested in mild cognitive impairment or Alzheimer's disease, that evidence is reported as what it is — evidence in an impaired population — and is never quietly transferred to healthy readers who want sharper focus at work.

Fifth, it is not neutral about measurement. A large part of this file's apparent inconsistency is not biological. It is what happens when small samples are given long test batteries repeatedly, and the interesting results are the ones that get written up. Section 04 treats that as a first-order fact rather than a caveat.

02 The word nootropic

The term was coined in the 1970s for a proposed drug class whose members would improve learning and memory, protect the brain against injury, and be nearly free of the toxicity and stimulant properties of conventional psychoactive drugs. As a definition it has an obvious problem: it specifies a desired outcome rather than a mechanism or a chemical family, so nothing prevents an arbitrary substance from being marketed under it.

What the market now labels "nootropic" spans at least six unrelated things: a stimulant with a well-characterised acute effect (caffeine); amino acids that are ordinary dietary constituents (L-theanine, L-tyrosine, glycine); an essential nutrient with a defined requirement and a deficiency syndrome (choline); standardised botanical extracts with chronic-dosing trial programmes (Bacopa, ginkgo, Rhodiola); an energy-metabolism substrate borrowed from sports nutrition (creatine); and prescription or unapproved drugs sold without disclosure (Cohen et al., 2022).

Those six groups do not share a mechanism, an evidence standard, a time course, or a risk profile. Any sentence beginning "nootropics improve..." is therefore false by construction, whatever follows. This document is organised by evidence structure instead: acute effects under load, chronic-dosing botanicals, nutrient status, and mitochondrial candidates.

03 Which outcomes actually matter

Cognitive research does not have one endpoint, and the endpoints differ in how hard they are to move and how much they matter.

The easiest thing to change is subjective state — self-rated alertness, mental fatigue, or clarity. These respond to expectation, to caffeine, and to being in a study. They are real experiences and they are the weakest possible evidence for a biological claim.

Next are simple speeded tasks — simple and choice reaction time, digit vigilance, rapid visual information processing. These are sensitive to arousal, which is why stimulants move them reliably. Movement here is a statement about arousal, not about memory.

Then memory and learning measures — word-list recall, delayed recognition, paired associates. These are the endpoints most claims are actually about and the hardest to move in healthy people, because a healthy adult's memory is not deficient in anything.

Then executive function — set-shifting, inhibition, working-memory updating, Trail-Making B. Slow, effortful, and sensitive to practice.

Then composite batteries and global scores, which pool many tests into a single standardised score. This is what large modern trials use, because it reduces the multiplicity problem described in the next section (Baker, Manson, Rapp, and colleagues, 2023; Ngandu, Lehtisalo, Solomon, and colleagues, 2015).

Finally, clinical events — incident mild cognitive impairment, incident dementia, conversion from mild cognitive impairment to dementia. These are the outcomes that matter most and are almost never studied, because doing so requires thousands of participants followed for years. Exactly two prevention programmes here reached that standard, and both are reported in section 12.

The claim ladder therefore runs from "felt more alert" to "did not develop dementia," and a product may sit six rungs below where its packaging implies.

04 Why cognitive testing is easy to fool

Three methodological facts explain more of this literature's disagreement than any biological hypothesis.

Practice effects. People get better at cognitive tests by taking them. Goldberg, Harvey, Wesnes, and colleagues examined practice effects arising from serial cognitive assessment and set out the implications for preclinical Alzheimer's disease randomised trials, where repeated testing is unavoidable (Goldberg et al., 2015). Aschenbrenner, Hassenstab, Wang, and colleagues took the argument a step further, asking whether prevention trials should try to avoid practice effects or use them as an outcome in their own right (Aschenbrenner et al., 2022). Both papers are established as methodology. Their consequence for this article is blunt: in an uncontrolled or single-arm study, improvement across visits is the expected result of testing, not evidence of a compound.

Multiplicity. A typical supplement trial administers a battery of eight to fifteen measures, each yielding several outcomes, at two or three timepoints. Without a pre-specified primary endpoint and correction, the probability that something reaches conventional significance approaches certainty. This is why a study can be described as positive while its own abstract reports movement on one measure out of a dozen.

Sample size. Many of the trials in this field enrolled fewer than fifty people. At that size, only large effects are detectable, so a published significant result in a small trial is either a large effect or an artefact — and small studies with null results are published less often.

None of this means the field is worthless. It means that when a large, pre-registered, adequately powered trial with a single composite primary endpoint disagrees with a cluster of small positive studies, the large trial is not one more data point to be averaged in. Section 12 is the clearest example in this document.

05 Three populations that are not one population

Almost every dispute in this field dissolves once the population is named.

People with a deficiency. If a nutrient is genuinely lacking, correcting it can improve function, and this says nothing about supplementation in people who are replete. Choline is the clean case: it has a defined requirement and a demonstrable deficiency state (Zeisel and da Costa, 2009; NIH ODS, 2026).

People with impairment. Mild cognitive impairment and dementia populations have room to improve on clinical scales, different biology, and different placebo dynamics. Most of the positive trials here — acetyl-L-carnitine, α-GPC, Hericium erinaceus, phosphatidylserine's original programme — were conducted here (Montgomery, Thal, and Amrein, 2003; Sagaro, Traini, and Amenta, 2023; Mori, Inatomi, Ouchi, and colleagues, 2009; Crook, Tinklenberg, Yesavage, and colleagues, 1991).

Healthy people who want to be sharper. This is who buys the products, and it is the population with the least supporting evidence, because there is no deficit to correct and performance is already near ceiling on most measures.

Transferring a result across these three groups is the single most common error in commercial cognitive claims. The population is named in every reporting sentence.

06 Caffeine, and the withdrawal-reversal problem

Caffeine is the only compound here whose acute effects on attention are not seriously disputed, and it is also the compound whose evidence base contains the field's most instructive confound.

Einöther and Giesbrecht reviewed the assumptions behind caffeine's reputation as an attention enhancer and found several of them weaker than commonly stated (Einöther and Giesbrecht, 2013). The central problem is that habitual consumers are, in most studies, tested after a period of abstinence. Their baseline is therefore a mild withdrawal state, and caffeine's apparent benefit may be the reversal of that state rather than enhancement above normal function.

Rogers, Heatherley, Hayward, and colleagues examined caffeine and caffeine withdrawal on mood and cognitive performance degraded by sleep restriction, one of the designs capable of separating the two accounts (Rogers et al., 2005). Heatherley reviewed what the withdrawal literature does and does not establish about sleepiness and driving performance, and concluded that the research supports less than is commonly claimed (Heatherley, 2011).

The honest summary is narrower than the marketing sentence. That caffeine improves speeded attention tasks acutely, in the state most people are in when they take it, is strongly supported. That it raises cognitive function above a genuine, non-withdrawn baseline is plausible and not settled. That it improves memory or executive function in a way that persists is not supported.

For this article the practical consequence is that any "nootropic" blend containing caffeine or a caffeine-bearing botanical — guarana, yerba maté, green tea, kola — has a built-in active comparator problem. Its acute effect is attributable to caffeine unless the trial included a caffeine-matched arm, and most did not. The existing Yerba-Mate article covers the botanical caffeine sources; it is not repeated here.

07 L-theanine, and what the combination trial actually found

L-theanine is an amino acid found in tea, catalogued as C7H14N2O3 (PubChem, 2026d). It is sold as a calming agent and, more often, as caffeine's partner in "focus" formulations. The combination claim has a real primary source, and that source is more interesting than its citations suggest.

Haskell, Kennedy, Milne, and colleagues ran a randomised, placebo-controlled, double-blind, balanced crossover study of L-theanine at 250 mg and caffeine at 150 mg, alone and combined, with salivary caffeine monitored (Haskell et al., 2008). Three results deserve separating.

Caffeine alone produced faster digit-vigilance reaction time, improved accuracy on rapid visual information processing, and attenuated self-reported mental fatigue. That is the expected stimulant profile.

L-theanine alone increased headache ratings and decreased correct serial-seven subtractions (Haskell et al., 2008). Taken by itself, in this study, it made one measure worse. This is rarely mentioned in material promoting theanine.

The combination improved rapid visual information processing accuracy and mental-fatigue ratings, and additionally produced faster simple reaction time, faster numeric working-memory reaction time, and improved sentence-verification accuracy, with reduced "headache" and "tired" ratings and increased "alert" ratings; there was a significant caffeine × theanine interaction on delayed word-recognition reaction time (Haskell et al., 2008). The authors' own conclusion was appropriately modest: beverages containing both may have a different pharmacological profile from those containing caffeine alone.

A different profile is not a larger effect, and it is not a memory claim. The finding is strongly supported as an acute single-dose result in that sample. Mancini, Beglinger, Drewe, and colleagues systematically reviewed green tea's effects on cognition, mood, and brain function, which is the closest thing to a whole-beverage synthesis (Mancini et al., 2017). Kahathuduwa and colleagues extended the combination to sustained attention and inhibitory control in children with attention-deficit/hyperactivity disorder as a proof-of-concept study, which is a clinical population and a separate question. The dedicated L-Theanine article carries the compound-level detail.

08 L-tyrosine under load

L-tyrosine is the precursor of dopamine and noradrenaline, and its evidence base has an unusually clear shape: it does something when catecholamine turnover is high, and little otherwise.

Jongkees, Hommel, Kühn, and colleagues reviewed tyrosine supplementation in clinical and healthy populations under stress or cognitive demand and concluded that benefit is contingent on the depleting condition rather than general (Jongkees et al., 2015). Attipoe, Zeno, Lee, and colleagues conducted a rapid evidence assessment for military use — the population most often exposed to cold, noise, and sleep loss — and reached a comparably conditional conclusion (Attipoe et al., 2015). Lieberman's earlier synthesis of nutrition, brain function, and cognitive performance placed tyrosine in the same conditional frame (Lieberman, 2003).

The pattern is strongly supported as a boundary condition: tyrosine's effects appear under acute stressors that deplete catecholamines, and are small or absent in unstressed people performing ordinary tasks. That is a genuinely useful scientific finding and a poor basis for a daily "focus" product sold to people sitting at a desk. Compound-level detail sits in L-Tyrosine-and-NALT.

09 Creatine, the most surprising entry

Creatine (C4H9N3O2; PubChem, 2026e) belongs to sports nutrition, and its presence here is a result rather than an assumption. The brain is metabolically expensive, creatine participates in cellular energy buffering, and supplementation can raise brain creatine content — so the hypothesis is mechanistically reasonable rather than post hoc.

Prokopidis, Giannos, Triantafyllidis, and colleagues conducted a systematic review and meta-analysis of randomised controlled trials of creatine and memory in healthy people. From twenty-three initially identified trials, ten met inclusion criteria and eight entered the meta-analysis. Overall, creatine improved measures of memory relative to placebo, with a standardised mean difference of 0.29 (95% CI 0.04 to 0.53; I² = 66%; p = 0.02). The subgroup finding is the substantive one: a significant improvement in older adults aged 66 to 76 (SMD 0.88; 95% CI 0.22 to 1.55; I² = 83%; p = 0.009) against essentially nothing in younger participants aged 11 to 31 (SMD 0.03; 95% CI −0.14 to 0.20; I² = 0%; p = 0.72). Dose, duration, sex, and geographical origin did not influence the findings (Prokopidis et al., 2023).

Read carefully, that is a small overall effect with substantial heterogeneity, concentrated in older adults, resting on eight trials. Xu, Bi, Zhang, and colleagues published a broader systematic review and meta-analysis of creatine and cognitive function in adults (Xu et al., 2024), subsequently issued a corrigendum (Xu et al., 2025), and drew a published commentary from Citherlet (Citherlet, 2026) — a sequence that is itself evidence about how fragile pooled estimates in this field can be. This review therefore grades creatine's memory effect in older adults as emerging, not established, and its effect in healthy young adults as not supported. The existing Creatine-Monohydrate and Alternative-Creatine-Forms articles hold the wider evidence.

10 What acute means

Sections 06 to 09 describe the strongest part of this entire field, and it is worth stating plainly how limited that part is.

The reliable acute effects are: caffeine on speeded attention, in habitual consumers who have abstained; tyrosine under an acute depleting stressor; a modified profile when theanine accompanies caffeine. Creatine's signal is not acute at all — it is a several-week exposure with an effect concentrated in older adults.

None of these is a memory-enhancement finding in healthy young adults. None is a neuroprotection finding. None was measured over the years that a long-term brain-health claim implies. An acute arousal effect and a claim about the trajectory of cognitive ageing are separated by every methodological problem in section 04, and the market routinely presents the first as evidence for the second.

11 Bacopa monnieri

Bacopa monnieri has the most coherent chronic-dosing evidence structure of any botanical here, and it is instructive precisely because the structure is coherent while the effect is narrow.

Kongkeaw, Dilokthornsakul, Thanarangsarit, and colleagues meta-analysed randomised, placebo-controlled trials of standardised Bacopa extracts, restricting inclusion to chronic dosing of twelve weeks or more without co-medication. Nine studies met criteria, covering 518 participants; the meta-analysis of 437 eligible participants reported improved cognition on two measures — a shortened Trail-Making B time (−17.9, 95% CI −24.6 to −11.2; p < 0.001) and a decreased choice reaction time (reported as 10.6, 95% CI −12.1 to −9.2; p < 0.001) — and the authors concluded that Bacopa has the potential to improve cognition, "particularly speed of attention," while stating that only a large head-to-head trial against an existing medication would provide definitive data (Kongkeaw et al., 2014).

Two features of that abstract deserve recording rather than smoothing over. First, the point estimate for choice reaction time is reported with a sign inconsistent with its own confidence interval, and the units given for Trail-Making B are implausible for that test; a reader cannot reconstruct the effect magnitude from the abstract alone. This is a citation-integrity problem, not a reason to dismiss the analysis, and it is the reason this article reports the finding as a direction rather than a magnitude. Second, and more importantly, the outcomes that moved are speed-of-attention measures. That is not the memory-enhancement claim under which Bacopa is sold.

The individual trials fit the same shape. Calabrese, Gregory, Leo, and colleagues tested a standardised extract on cognitive performance, anxiety, and depression in elderly participants in a randomised trial (Calabrese et al., 2008). Kean, Kaufman, Lomas, and colleagues tested the CDRI 08 extract on hyperactivity and inattention in a paediatric randomised controlled trial (Kean et al., 2015) — a different population and a different question. Two recent network meta-analyses place Bacopa against comparators: Tiemtad, Ingkaninan, Temkitthawon, and colleagues compared Bacopa and ginkgo directly (Tiemtad et al., 2026), and Feng, Fan, and Wei examined plant active substances on cognitive function in healthy older adults (Feng, Fan, and Wei, 2025). A network meta-analysis cannot be stronger than the trials it pools, and both inherit the small-sample problem described in section 04.

The grade is emerging for an effect on speed-of-attention measures after twelve weeks or more of a standardised extract, and not supported for the memory claim that dominates its marketing. The characteristic delay — nothing at four weeks, something at twelve — is a real and unusual feature of this botanical's file, and it also means that any Bacopa-containing product tested acutely was tested wrongly.

12 Ginkgo biloba, and the largest prevention file in the field

Ginkgo is the only compound here that has been subjected to the study design its claims require: a multi-thousand-participant, multi-year, randomised, placebo-controlled trial with incident dementia as the primary endpoint. It happened twice, on two continents, with the same standardised extract. Both were null.

The Ginkgo Evaluation of Memory study enrolled 3,069 community volunteers aged 75 or older across United States academic medical centres, 2,587 with normal cognition and 482 with mild cognitive impairment at entry, randomised to Ginkgo biloba extract 120 mg twice daily (n = 1,545) or placebo (n = 1,524), with assessment every six months and a median follow-up of 6.1 years (DeKosky, Fitzpatrick, Ives, and colleagues, 2006; DeKosky, Williamson, Fitzpatrick, and colleagues, 2008). Five hundred and twenty-three participants developed dementia. The overall dementia rate was 3.3 per 100 person-years on ginkgo against 2.9 per 100 person-years on placebo; the hazard ratio for all-cause dementia was 1.12 (95% CI 0.94 to 1.33; p = 0.21) and for Alzheimer's disease 1.16 (95% CI 0.97 to 1.39; p = 0.11). Ginkgo had no effect on progression to dementia among participants with mild cognitive impairment (HR 1.13; 95% CI 0.85 to 1.50; p = 0.39). Dropout and loss to follow-up were low at 6.3%, and adverse-effect profiles were similar between groups (DeKosky et al., 2008).

Snitz, O'Meara, Carlson, and colleagues then reported the same cohort's cognitive trajectories. Annual rates of decline in z scores did not differ between groups in any domain — memory, attention, visuospatial ability, language, or executive function — and rates of change on the Modified Mini-Mental State Examination and the ADAS-Cog did not differ by treatment group (p = 0.71 and p = 0.97 respectively). There was no effect modification by age, sex, race, education, APOE ε4 allele, or baseline mild cognitive impairment (Snitz et al., 2009).

GuidAge was the European counterpart. Vellas, Coley, Ousset, and colleagues enrolled 2,854 adults aged 70 or older who had spontaneously reported memory complaints to a primary-care physician in France, randomised 1:1 to standardised extract EGb 761 at 120 mg twice daily or matched placebo, followed for five years, with conversion to probable Alzheimer's disease as the primary outcome. By five years, 61 participants in the ginkgo group had been diagnosed with probable Alzheimer's disease (1.2 cases per 100 person-years) against 73 in the placebo group (1.4 per 100 person-years; HR 0.84, 95% CI 0.60 to 1.18; p = 0.306) — with the authors noting that the risk was not proportional over time. Adverse events were much the same between groups: 76 deaths on ginkgo against 82 on placebo (HR 0.94, 95% CI 0.69 to 1.28; p = 0.68), and 65 strokes against 60 (risk ratio 1.12, 95% CI 0.77 to 1.63; p = 0.57). The trial was funded by Ipsen (Vellas et al., 2012).

Williamson, Vellas, Furberg, and colleagues had already published a comparison of the design differences between the two trials, which is the correct place to look before treating them as a replication (Williamson et al., 2008).

Note the direction of the two point estimates: GEM's hazard ratios were slightly above 1, GuidAge's slightly below 1, and neither excluded 1. That is what a null looks like when it is measured twice properly. The National Center for Complementary and Integrative Health summarises the position for a general readership (NCCIH, 2026a).

This is the most important entry in the article, and its grade is unambiguous. That standardised Ginkgo biloba extract at 120 mg twice daily does not reduce incident dementia or Alzheimer's disease in older adults with normal cognition or mild cognitive impairment is established. That it does not slow the rate of domain-specific cognitive decline in that population is established. The Ginkgo-Biloba article holds the compound-level pharmacology, the earlier symptomatic-treatment literature, and the bleeding-interaction question.

What follows from this generalises beyond ginkgo. Ginkgo entered these trials with a plausible mechanism, a large earlier positive literature, decades of use, and a pharmaceutical-grade standardised extract. It still failed. Every other botanical here has less evidence than ginkgo had in 1999, and none has been tested at this scale.

13 Rhodiola rosea

Rhodiola rosea is sold for mental fatigue rather than memory, and its evidence base matches that framing while remaining thin.

Darbinyan, Kteyan, Panossian, and colleagues studied the standardised extract SHR-5 in stress-induced fatigue using a double-blind crossover design with repeated low-dose administration (Darbinyan et al., 2000). Olsson, von Schéele, and Panossian conducted a randomised, double-blind, placebo-controlled, parallel-group study of the same SHR-5 extract (Olsson et al., 2009). Both are small, both used a single manufacturer's standardised extract, and neither is a memory trial. NCCIH summarises the general position (NCCIH, 2026b).

Two structural cautions apply. The evidence concentrates on one extract from one producer, so it is evidence about SHR-5 rather than about Rhodiola as a genus or about arbitrary commercial products. And fatigue outcomes are subjective, placing this evidence low on the ladder in Figure 1.

The grade is plausible for reduced self-rated mental fatigue under stress with that specific standardised extract, and not supported for memory, attention in unstressed people, or any long-term cognitive claim. Rhodiola also appears in the forthcoming Stress, Anxiety and Adaptogenic Nutrition and Mitochondrial Energy, Fatigue and Vitality titles in this series, where the fatigue literature belongs; it is cross-referenced rather than repeated.

14 Lion's mane

Hericium erinaceus has a single frequently cited human trial, and reading it closely is more useful than citing it.

Mori, Inatomi, Ouchi, and colleagues conducted a double-blind, parallel-group, placebo-controlled trial in Japanese men and women aged 50 to 80 diagnosed with mild cognitive impairment. After a two-week preliminary examination, 30 participants were randomised into two groups of 15. The active group took four 250 mg tablets of 96% Hericium dry powder three times daily for sixteen weeks, with four weeks of observation afterwards. Scores on a cognitive scale based on the Revised Hasegawa Dementia Scale were significantly higher than placebo at weeks 8, 12, and 16, increased with duration of intake, and decreased significantly four weeks after intake stopped. Laboratory tests showed no adverse effect (Mori et al., 2009).

The washout reversal is a genuinely valuable design feature — it is harder to explain by practice effects than an on-treatment difference alone. But fifteen participants per arm, one site, one instrument, and one trial is the entire human efficacy file for a product category now sold at scale. Muhanna, Lund, Bromberg, and colleagues reviewed lion's mane for amyotrophic lateral sclerosis in the ALSUntangled series and did not find grounds to recommend it, which is a different indication but a useful check on how the mushroom's reputation outruns its data (Muhanna et al., 2024).

The grade is emerging, in mild cognitive impairment, from one trial of thirty people, on one scale. It is not supported as evidence for cognitive enhancement in healthy adults — the population that buys it — because that population was not studied. The Lions-Mane article carries the erinacine and hericenone chemistry and the nerve-growth-factor hypothesis, which is preclinical.

15 Ashwagandha

Withania somnifera belongs primarily to the stress and anxiety literature, but it carries cognitive claims and two randomised trials that touch them.

Choudhary, Bhattacharyya, and Bose tested a root extract for memory and cognitive function in a randomised trial (Choudhary et al., 2017). Gopukumar, Thanawala, Somepalli, and colleagues tested a root extract on cognitive function in healthy, stressed adults in a randomised, double-blind, placebo-controlled design (Gopukumar et al., 2021). NCCIH summarises usefulness and safety, including the hepatotoxicity reports that have accumulated for this botanical (NCCIH, 2026c).

Two things matter here. The population in the Gopukumar trial is stressed adults, which makes the plausible pathway indirect — reduced stress improving performance is a different mechanism from a direct cognitive effect, and the trials are not designed to separate them. And ashwagandha's safety file is not empty in the way the botanicals above are; the liver signal is discussed in section 24.

The grade is plausible for cognitive measures in stressed adults, most likely mediated by stress reduction rather than by a direct nootropic action, and not supported as a memory-enhancement claim. The full treatment belongs to the forthcoming Ashwagandha standalone title and the Stress, Anxiety and Adaptogenic Nutrition system title.

16 Mucuna pruriens, the botanical that is a drug

Mucuna pruriens is here for a reason that has nothing to do with cognitive enhancement: it is the clearest available demonstration that "botanical" and "gentle" are unrelated properties.

Mucuna seeds contain L-DOPA. Cilia, Laguna, Cassani, and colleagues investigated whether it could serve as an alternative levodopa source for people with Parkinson's disease who cannot afford marketed preparations, in a double-blind, randomised, controlled, crossover study of eighteen patients with advanced disease. Participants received, in randomised sequence, dispersible levodopa at 3.5 mg/kg with benserazide as the reference treatment; high-dose Mucuna powder at 17.5 mg/kg; low-dose Mucuna at 12.5 mg/kg; levodopa without a decarboxylase inhibitor at 17.5 mg/kg; Mucuna plus benserazide at 3.5 mg/kg; and placebo. Compared with levodopa plus benserazide, low-dose Mucuna produced similar motor response with fewer dyskinesias and adverse events, while high-dose Mucuna produced greater motor improvement at 90 and 180 minutes, longer ON duration, and fewer dyskinesias. Single-dose Mucuna met all non-inferiority efficacy and safety outcomes against dispersible levodopa/benserazide (Cilia et al., 2017).

The same group then ran a sixteen-week non-inferiority randomised crossover pilot of daily intake in advanced Parkinson's disease (Cilia et al., 2018), and Boonmongkol, Phokaewvarangkul, Vimolmangkang, and colleagues subsequently compared Mucuna against conventional levodopa in a randomised controlled trial with pharmacokinetic measurement (Boonmongkol et al., 2025).

That a roasted-seed powder is pharmacologically non-inferior to a dispersible levodopa preparation is strongly supported for single-dose motor response in advanced Parkinson's disease. It is also the reason Mucuna should not be treated as a wellness ingredient. A preparation with measurable dopaminergic activity carries dopaminergic risks — dyskinesia, cardiovascular response, psychiatric effects, and interaction with levodopa therapy — and its L-DOPA content varies with cultivar and processing. The evidence for Mucuna as a cognitive or mood enhancer in people without Parkinson's disease is speculative; the evidence that it is a genuine dopaminergic exposure is not.

Botanical	Best available design	Population	What moved	Grade	What was not shown
Ginkgo biloba	two multi-year RCTs, n = 3,069 and 2,854	older adults, normal cognition or MCI	nothing	established null	no prevention, no slowed decline
Bacopa monnieri	meta-analysis, 9 trials, 518 participants	mixed, ≥12 weeks	speed-of-attention measures	emerging	memory; effect magnitude unreconstructable
Hericium erinaceus	one RCT, 15 per arm, 16 weeks	mild cognitive impairment	HDS-R-based scale, reversed on washout	emerging	healthy adults; replication
Rhodiola rosea	two small RCTs, one extract (SHR-5)	stressed or fatigued adults	self-rated mental fatigue	plausible	memory; other extracts; long term
Withania somnifera	two RCTs	stressed adults, memory complaints	cognitive measures, plausibly via stress	plausible	direct cognitive mechanism
Mucuna pruriens	crossover RCT vs dispersible levodopa	advanced Parkinson's disease	motor response, non-inferior	strongly supported	any cognitive claim in healthy people

Table 1. Chronic-dosing botanicals. The column that matters is the fourth. Where a compound has been tested at the scale its claim requires, it failed; where it has produced a signal, the signal is narrower than the claim.

17 Choline is a nutrient, not a nootropic

Choline (C5H14NO⁺; PubChem, 2026a) is the one substance here that is unambiguously required. It is a precursor of acetylcholine and of the membrane phospholipids phosphatidylcholine and sphingomyelin, and it participates in one-carbon metabolism as a methyl donor. Zeisel and da Costa set out the case for treating it as an essential nutrient of public-health significance (Zeisel and da Costa, 2009); the Office of Dietary Supplements records the adequate intake values and the deficiency consequences (NIH ODS, 2026).

Two features of choline's biology are more interesting than any supplement claim.

Requirements vary genetically. Ganz, Klatt, and Caudill showed that common genetic variants alter choline metabolism and influence dietary requirements (Ganz et al., 2017). That is strongly supported and it means population-level intake recommendations understate individual variation in a way that is unusual among nutrients.

Requirements rise in pregnancy and lactation, when choline supports fetal neurodevelopment. Wallace, Blusztajn, Caudill, and colleagues reviewed the position for obstetric and gynaecological practice (Wallace et al., 2020). This is the strongest neurocognitive argument for choline anywhere in the literature, and it is an argument about adequacy during development, not about enhancement in adults.

The adult cognitive evidence is observational. Poly, Massaro, Seshadri, and colleagues related dietary choline to cognitive performance and white-matter hyperintensity in the Framingham Offspring Cohort (Poly et al., 2011). Yuan, Liu, Liu, and colleagues examined whether dietary choline intake relates to dementia and Alzheimer's disease risk in the Framingham Heart Study (Yuan et al., 2022). Both are cohort analyses of dietary intake, which means they carry the standard confounding of nutritional epidemiology: choline intake tracks with eggs, fish, meat, and dairy, and therefore with overall diet quality, education, and income. They are emerging as hypothesis-generating associations and cannot support a supplementation claim.

Choline intake also sits inside a separate cardiovascular literature concerning its microbial metabolism, which this article does not adjudicate; that question belongs to the Cardiovascular Nutraceuticals and Trimethylglycine titles.

The honest position: choline deficiency impairs function, adequacy matters most in pregnancy, requirements are genetically variable, and there is no randomised evidence that supplementing a replete adult improves cognition.

18 Alpha-GPC

L-α-glycerylphosphorylcholine, or choline alfoscerate (C8H20NO6P; PubChem, 2026b), is a choline-containing phospholipid. Its status is unusual: in several countries it has been a prescription medicine since 1985, while in the United States it is sold as a dietary supplement. That difference explains most of the confusion around it.

Gatti, Barzaghi, Acuto, and colleagues compared free plasma choline levels after intramuscular administration of L-α-glycerylphosphorylcholine against a comparator, establishing that it does raise circulating choline (Gatti et al., 1992). That is established pharmacokinetics and the basis of the mechanistic story.

The efficacy evidence is where care is required. Sagaro, Traini, and Amenta screened 1,326 studies and 300 full-text articles and included seven randomised controlled trials and one prospective cohort study. They reported significant effects for α-GPC in combination with donepezil on cognition (4 RCTs; mean difference 1.72, 95% CI 0.20 to 3.25), functional outcomes (3 RCTs; MD 0.79, 95% CI 0.34 to 1.23), and behavioural outcomes (4 RCTs; MD −7.61, 95% CI −10.31 to −4.91), and that patients receiving α-GPC had better cognition than those receiving placebo or other medications (MD 3.50, 95% CI 0.36 to 6.63). Their conclusion named the population precisely: patients with neurological conditions associated with cerebrovascular injury (Sagaro et al., 2023).

Read that population and comparator honestly. The strongest results are for α-GPC added to a cholinesterase inhibitor in people with cerebrovascular injury. That is an adjunctive pharmacotherapy finding. It is not evidence that a healthy adult taking α-GPC before a meeting will think more clearly, and no trial in this file examined that question.

Kim, Park, Kim, and colleagues conducted a nationwide cohort study using South Korea's National Health Insurance Service data on 508,107 patients newly diagnosed with mild cognitive impairment between 2013 and 2016, classifying them as users or non-users by prescription record and using time-dependent Cox regression. Users had a lower risk of progression to Alzheimer's disease dementia (HR 0.899, 95% CI 0.882 to 0.918) and to vascular dementia (HR 0.832, 95% CI 0.801 to 0.865); stroke risk decreased significantly in patients who did not progress to dementia but not in those who did (Kim et al., 2025).

Half a million people is an impressive denominator and it does not fix the design. This is a prescription-record cohort, so users differ systematically from non-users in ways a claims database cannot fully capture — the classic confounding by indication, plus healthy-adherer effects. The authors themselves called for large-scale randomised studies. The finding is emerging as a real-world association and is not randomised evidence of prevention.

Grade: strongly supported as an adjunct in cognitive impairment associated with cerebrovascular injury; emerging as a cohort association with delayed dementia conversion; not supported as a cognitive enhancer in healthy adults.

19 Phosphatidylserine, and the substitution that broke its evidence

Phosphatidylserine is the most instructive entry here for a reader trying to judge a supplement claim, because its two central trials disagree and the reason they disagree is in the product, not the statistics.

Crook, Tinklenberg, Yesavage, and colleagues treated 149 patients meeting criteria for age-associated memory impairment for twelve weeks with 100 mg BC-PS three times daily or placebo. Treated patients improved relative to placebo on performance tests related to learning and memory tasks of daily life, and analysis of clinical subgroups suggested that those performing at a relatively low level before treatment were most likely to respond; within that subgroup there was improvement on computerised and standard neuropsychological tests and on clinical global ratings (Crook et al., 1991).

"BC-PS" means bovine-cortex-derived phosphatidylserine. That source was subsequently abandoned for products intended for human consumption, for reasons of transmissible spongiform encephalopathy risk, and the commercial supply moved to soy-derived phosphatidylserine, which has a different fatty-acid composition.

Jorissen, Brouns, Van Boxtel, and colleagues then tested the replacement. One hundred and twenty participants over 57 years of age meeting stringent age-associated memory impairment criteria were randomised to placebo, 300 mg soy-derived phosphatidylserine daily, or 600 mg daily, assessed at baseline, six weeks, and twelve weeks, with a three-week washout, using delayed recall and recognition of a learned word list as primary outcomes. No significant differences were found in any outcome variable between treatment groups, and there were no significant interactions between treatment and severity of memory complaints. The authors concluded that a daily supplement of soy phosphatidylserine does not affect memory or other cognitive functions in older individuals with memory complaints (Jorissen et al., 2001).

McDaniel, Maier, and Einstein reviewed the wider category of "brain-specific" nutrients under the question of whether they constitute a memory cure, which is the appropriate frame (McDaniel et al., 2003).

Phosphatidylserine also carries a United States qualified health claim relating to cognitive dysfunction, exercised under enforcement discretion (FDA, 2026). A qualified health claim is a regulatory permission to make a carefully hedged statement given limited and inconsistent evidence. It is not a finding of efficacy, and the distinction is routinely elided in marketing.

Grade: the bovine-cortex result is strongly supported as a finding about a product that is no longer sold; the soy replacement's null result is strongly supported; the claim that currently marketed soy phosphatidylserine improves memory is not supported. Any citation of Crook (1991) in support of a soy product is a citation of the wrong molecule.

20 Acetyl-L-carnitine

Acetyl-L-carnitine (C9H17NO4; PubChem, 2026c) has a mitochondrial rationale — it participates in fatty-acid transport and acetyl-group donation — and a meta-analysis that is frequently cited without its effect size.

Montgomery, Thal, and Amrein meta-analysed double-blind, placebo-controlled, prospective, parallel-group trials of at least three months' duration in mild cognitive impairment and mild Alzheimer's disease. Study durations were three, six, or twelve months and daily amounts ranged from 1.5 to 3.0 g. The integrated summary effect size was 0.201 (95% CI 0.107 to 0.295), and the effect size for Clinical Global Impression of Change was 0.32 (95% CI 0.18 to 0.47). Benefit was seen on both clinical scales and psychometric tests, appeared by the first assessment at three months, and increased over time; the compound was well tolerated in all studies (Montgomery et al., 2003).

An effect size of 0.2 is small by convention. It was obtained in a population with cognitive impairment, in trials conducted largely in the 1980s and 1990s, and it has not been followed by a large modern confirmatory trial. Pennisi, Lanza, Cantone, and colleagues published a critical update on acetyl-L-carnitine in dementia and other cognitive disorders that is the appropriate place to see how the field's view has aged (Pennisi et al., 2020). Ames and Liu's account of delaying mitochondrial decay of ageing with acetylcarnitine is the mechanistic and animal-model line of argument (Ames and Liu, 2004), and should not be read as human outcome evidence.

Grade: strongly supported as a small effect in mild cognitive impairment and mild Alzheimer's disease as of the 2003 synthesis; not supported in healthy adults. The Acetyl-L-Carnitine article in this series holds the pharmacology and the separate literatures on diabetic neuropathy and male fertility.

21 Pyrroloquinoline quinone

PQQ (C14H6N2O8; PubChem, 2026f) is a redox cofactor with a mitochondrial-biogenesis hypothesis and a human evidence base consisting of several small trials, most of them conducted around a single ingredient brand.

Itoh, Hine, Miura, and colleagues reported effects of pyrroloquinoline quinone disodium salt, marketed as BioPQQ, on cognitive functions (Itoh et al., 2016). Shiojima, Takahashi, Takahashi, and colleagues conducted a randomised, double-blind trial of the disodium salt on cognitive function in healthy volunteers (Shiojima et al., 2022). Tamakoshi, Suzuki, Nishihara, and colleagues reported improved brain function in both younger and older adults (Tamakoshi et al., 2023). Baltic, Nedeljkovic, Todorovic, and colleagues examined six weeks of dihydrogen-pyrroloquinoline quinone on mitochondrial biomarkers, brain metabolism, and cognition (Baltic et al., 2024).

The structural weaknesses are consistent across the set: small samples, a narrow set of research groups, commercial ingredient sponsorship, and outcome batteries of the kind section 04 warns about. The mitochondrial-biogenesis mechanism, whatever its merits in cell systems, has not been demonstrated to be the cause of any human cognitive result here.

Grade: emerging at best for cognitive measures in small trials; the mitochondrial-biogenesis explanation is speculative in humans. The existing Pyrroloquinoline-Quinone article carries the compound detail.

Compound	Mechanism claimed	Mechanism status	Human outcome status	Population actually studied
Choline	acetylcholine and membrane precursor; methyl donor	established nutrient biology	deficiency and pregnancy adequacy matter; supplementation in replete adults untested	deficiency states; pregnancy; cohorts
Alpha-GPC	raises circulating choline	established PK	adjunct effect with donepezil in cerebrovascular injury	cognitive impairment, not healthy adults
Phosphatidylserine	membrane phospholipid support	plausible	bovine-derived positive; soy-derived null	age-associated memory impairment
Acetyl-L-carnitine	mitochondrial fatty-acid transport, acetyl donation	plausible; animal support	small effect (ES 0.20) in MCI and mild AD	cognitive impairment
PQQ	mitochondrial biogenesis	speculative in humans	small trials, narrow sponsorship	mostly healthy volunteers, small n
Creatine	brain energy buffering	strongly supported	small memory effect concentrated in ages 66-76	healthy adults, both ages

Table 2. Mechanism against outcome. A well-established mechanism does not produce a human outcome, and the two columns are independent. Alpha-GPC has the best-established pharmacokinetics and the least evidence in the population that buys it.

22 Stacks: evidence versus co-occurrence

The workbook underlying this article contained nineteen multi-ingredient cognitive blends. Not one of them was the subject of a trial of itself. This is the normal state of the category, and it has a specific consequence: a blend inherits neither the evidence nor the safety data of its constituents.

Three genuine combination findings exist in this file, and each is narrower than the practice it is used to justify.

Caffeine plus L-theanine. Tested directly, acutely, single dose, in one crossover study, with a caffeine-only arm — which is what makes it interpretable (Haskell et al., 2008). Note again that theanine alone worsened one measure in that study.

Alpha-GPC plus donepezil. The strongest results in the α-GPC meta-analysis were for the combination with a cholinesterase inhibitor in people with cerebrovascular injury (Sagaro et al., 2023). This is a prescription-drug combination in a clinical population, not a supplement stack.

A multivitamin-mineral. A fixed-composition micronutrient product tested as a unit in a large randomised trial, discussed in section 25 (Baker et al., 2023).

Everything else is co-occurrence. When a product contains α-GPC, Bacopa, ginkgo, lion's mane, PQQ, theanine, tyrosine, and caffeine, the following are all unknown: whether the amounts match those used in any trial; whether the constituents interact pharmacokinetically; which constituent produces any effect experienced; and whether the total exceeds a tolerable intake for any component. What is knowable is that the product's acute effect is most plausibly attributable to its caffeine, and that the blend's own evidence base is empty.

The clean question to ask of any blend is the one from section 06: was there a caffeine-matched comparator? If there was not, an acute subjective effect is uninformative about every other ingredient on the label.

23 Adulteration and undeclared drugs

This is not a hypothetical risk, and it is the single most concrete safety finding in the article.

Cohen, Avula, and Khan analysed dietary supplements sold over the counter in the United States that declared centrophenoxine (meclofenoxate) — a drug prescribed in China and elsewhere for conditions including dementia, and not approved by the Food and Drug Administration for any indication. Products were included if the label carried both the term "dietary supplement" and centrophenoxine as a declared ingredient, then analysed by ultra-high-performance liquid chromatography. Centrophenoxine was present in all seven products analysed, at 79 to 251 mg per serving. A consumer following the maximum recommended daily intake on the label would be exposed to 237 to 752 mg per day. Only one of seven products, 14%, listed a quantity within ±10% of the actual amount (Cohen et al., 2022).

Three separate findings sit inside that paragraph. An unapproved drug was being sold as a dietary supplement. The doses were pharmacologically substantial. And label quantities were wrong in six of seven products, meaning even a consumer who researched the declared ingredient could not know the exposure.

This is established as an analytical finding in that product sample. Its implication for the whole category is not that every product is adulterated, but that the product class has a demonstrated failure mode in which the label does not describe the contents — and that a cognitive effect experienced from an unregulated "nootropic" blend cannot be safely attributed to the botanical on the front of the bottle.

24 Safety, interactions and populations

The botanicals here are not uniformly benign, and the pattern of their risks does not track the strength of their efficacy evidence.

Ginkgo. The two large trials are the best safety data in the field, precisely because they were large and long. In GEM, adverse-effect profiles were similar between groups (DeKosky et al., 2008). In GuidAge, deaths were 76 on ginkgo against 82 on placebo (HR 0.94, 95% CI 0.69 to 1.28) and strokes 65 against 60 (risk ratio 1.12, 95% CI 0.77 to 1.63; p = 0.57), with other haemorrhagic and cardiovascular events not differing (Vellas et al., 2012). The long-standing bleeding concern is therefore not confirmed at this scale, while also not excluded — the confidence interval for stroke spans a 63% relative increase. NCCIH summarises the practical position (NCCIH, 2026a).

Ashwagandha. NCCIH's assessment includes the liver-injury reports that have accumulated for this botanical (NCCIH, 2026c). Of the compounds here, it has the most substantive organ-toxicity signal relative to its efficacy evidence.

Mucuna pruriens. A genuine dopaminergic exposure (Cilia et al., 2017; Cilia et al., 2018; Boonmongkol et al., 2025). Dyskinesia, cardiovascular response, psychiatric effects, and interaction with levodopa therapy are foreseeable consequences of dopaminergic activity, and L-DOPA content varies with cultivar and preparation.

Caffeine-bearing products. Dependence and withdrawal are the documented phenomena (Rogers et al., 2005; Heatherley, 2011), and a withdrawal state is what makes the next dose feel effective.

L-theanine. Increased headache ratings in the one careful crossover study of theanine alone (Haskell et al., 2008).

Undeclared drugs. Section 23.

Populations not studied. Pregnancy, lactation, childhood, and adolescence are essentially absent from this evidence base, with one exception in a paediatric attention-deficit population (Kean et al., 2015). Choline is the reverse case: its pregnancy literature is the strongest part of its file (Wallace et al., 2020).

Concern	Evidence	Grade	What is not known
Undeclared unapproved drugs	7/7 products contained centrophenoxine; 6/7 mislabelled quantity	established in that sample	prevalence across the category
Ginkgo bleeding / stroke	GuidAge stroke RR 1.12 (0.77-1.63), ns; GEM profiles similar	not confirmed at trial scale	rare events; anticoagulant co-use
Ashwagandha hepatotoxicity	case reports collated by NCCIH	plausible, rare	incidence; causal mechanism; extract dependence
Mucuna dopaminergic effects	non-inferior to dispersible levodopa	strongly supported	consequences of unsupervised use; content variability
Caffeine dependence / withdrawal	experimental withdrawal studies	strongly supported	contribution to perceived blend efficacy
Theanine alone	worsened serial sevens, increased headache in one crossover	emerging	replication; dose-response
Blend interactions	no trials of blends as products	no evidence	everything
Pregnancy, lactation, children	essentially unstudied except choline	insufficient	risk estimates

Table 3. Safety ledger. The row with the strongest evidence is the one about labelling, not about pharmacology.

25 What actually moved cognitive outcomes

Two randomised programmes here produced statistically significant improvements in cognitive outcomes in older adults. Neither was a botanical.

A multidomain lifestyle intervention. Ngandu, Lehtisalo, Solomon, and colleagues enrolled individuals aged 60 to 77 from previous national surveys with a CAIDE dementia risk score of at least 6 and cognition at or slightly below the age-expected level, randomising 1,260 participants to a two-year intervention of diet, exercise, cognitive training, and vascular risk monitoring (n = 631) or to general health advice (n = 629). Estimated mean change in the neuropsychological test battery total z score at two years was 0.20 in the intervention group and 0.16 in the control group; the between-group difference in change per year was 0.022 (95% CI 0.002 to 0.042; p = 0.030). Adverse events occurred in 46 (7%) intervention participants against 6 (1%) controls, most commonly musculoskeletal pain (Ngandu et al., 2015). Rosenberg, Ngandu, Rusanen, and colleagues later reported that the benefit extended across baseline characteristics in a large at-risk elderly population (Rosenberg et al., 2018).

A multivitamin-mineral. Baker, Manson, Rapp, and colleagues ran COSMOS-Mind, a two-by-two factorial three-year randomised trial assessing cognition by telephone annually in 2,262 participants (mean age 73; 60% women; 89% non-Hispanic White), with 92% completing baseline and at least one annual assessment. Cocoa extract containing 500 mg/day flavanols had no effect on global cognition (mean z 0.03, 95% CI −0.02 to 0.08; p = 0.28). Daily multivitamin-mineral supplementation produced a statistically significant benefit on global cognition (mean z 0.07, 95% CI 0.02 to 0.12; p = 0.007), most pronounced in participants with a history of cardiovascular disease, with benefits also seen for memory and executive function; the cocoa-by-multivitamin interaction was not significant for any composite (Baker et al., 2023). Sachs, Williams, Gaussoin, and colleagues reported the effect on incidence of mild cognitive impairment and dementia (Sachs et al., 2023), and Vyas, Manson, Sesso, and colleagues reported the clinic subcohort with in-person testing (Vyas et al., 2024).

Now read both effect sizes honestly. FINGER's between-group difference was 0.022 z per year, from a two-year intensive multidomain programme requiring supervised exercise, dietary change, and cognitive training. COSMOS-Mind's multivitamin effect was 0.07 z on a global composite over three years. These are small. They are also, in this entire article, the only randomised improvements in cognitive outcomes obtained in adequately powered trials.

Three inferences follow.

First, the effect sizes available in this field are small even when real. A product promising noticeable mental sharpening is promising something larger than the best randomised evidence has produced by any means.

Second, the multivitamin result is a micronutrient-sufficiency finding, not a nootropic finding, and it is consistent with section 05: the interventions that work are the ones correcting something. Its greater prominence in participants with cardiovascular disease points the same way. It requires confirmation in a more diverse cohort, as the authors state.

Third, the intervention with the largest apparatus — FINGER — is behavioural. The existing Exercise-Intervention, Strength-Training, Endurance-Training, Meditation, Insomnia, and Circadian-Rhythms articles in this series carry those literatures, and the Healthspan and Inflammaging titles carry the ageing frame. The Global Council on Brain Health's public-facing synthesis reaches a comparable conclusion about where the evidence sits (GCBH, 2026).

26 The "brain health" sentence

The market sentence says: these compounds enhance memory and focus; they protect the ageing brain; mechanisms in cells and animals show how; stacking them multiplies the benefit; being natural makes them safe; and if you feel sharper, your cognition improved.

The file says otherwise.

Caffeine acutely improves speeded attention in the abstained state most consumers occupy, and the withdrawal-reversal confound is unresolved (Einöther and Giesbrecht, 2013; Rogers et al., 2005). L-theanine alone worsened a calculation measure and increased headache; combined with caffeine it altered the profile (Haskell et al., 2008). Tyrosine acts under acute depleting stress and not otherwise (Jongkees et al., 2015; Attipoe et al., 2015). Creatine's memory effect is small, heterogeneous, and concentrated in ages 66 to 76 (Prokopidis et al., 2023).

Bacopa moved speed-of-attention measures after twelve weeks or more, not memory (Kongkeaw et al., 2014). Ginkgo, tested twice at the scale its claim requires, did not reduce incident dementia or Alzheimer's disease and did not slow domain-specific decline (DeKosky et al., 2008; Snitz et al., 2009; Vellas et al., 2012). Lion's mane rests on fifteen participants per arm in mild cognitive impairment (Mori et al., 2009). Rhodiola rests on two small trials of one manufacturer's extract, with subjective fatigue endpoints (Darbinyan et al., 2000; Olsson et al., 2009). Ashwagandha's cognitive signal is in stressed adults and is plausibly mediated by stress reduction (Gopukumar et al., 2021).

Alpha-GPC's best results are as an adjunct to donepezil in cerebrovascular injury (Sagaro et al., 2023), with a half-million-person cohort that cannot exclude confounding by indication (Kim et al., 2025). Phosphatidylserine's positive trial used a bovine-cortex product that is no longer sold, and the soy replacement was null (Crook et al., 1991; Jorissen et al., 2001). Acetyl-L-carnitine's meta-analytic effect size was 0.20 in cognitively impaired populations (Montgomery et al., 2003). PQQ rests on small trials with narrow sponsorship (Itoh et al., 2016; Shiojima et al., 2022).

Blends have no evidence of themselves. Seven of seven products declaring one unapproved drug contained it, six of seven at a quantity the label got wrong (Cohen et al., 2022). And the only adequately powered randomised improvements in cognitive outcomes anywhere here came from a two-year multidomain lifestyle programme (0.022 z per year) and a multivitamin-mineral (0.07 z over three years) (Ngandu et al., 2015; Baker et al., 2023).

Figure 3 sets the seven sentences the category sells beside what was measured.

Strip the category down to what the adequately powered trials actually show and the sentence is narrow: the cognitive-supplement category has one reliable acute effect that belongs to caffeine, a handful of narrow chronic signals whose endpoints do not match their marketing, a flagship botanical that failed the only two trials large enough to test its central claim, and a demonstrated labelling failure mode — while the two interventions that did improve cognitive outcomes in adequately powered randomised trials were a lifestyle programme and a multivitamin, both by small margins, and neither sold as a nootropic.

27 Standing constraint

28 References

29 Evidence handling

The numbered list above is generated at build from records retrieved from NCBI and from regulator, agency, and chemical-database sources whose URLs were resolved and checked; nothing in it was typed from memory.

Study types are named in the sentence that uses them. Receptor and cell work, animal models, single-dose crossover studies in volunteers, uncontrolled series, small parallel-group trials, meta-analyses, network meta-analyses, multi-year prevention trials with clinical endpoints, insurance-claims cohorts, analytical chemistry of retail products, and agency assessments are not interchangeable. Where results conflict, both are reported. Where a synthesis reported heterogeneity, the heterogeneity statistic is given. Where a published abstract contains an internal inconsistency, it is recorded rather than silently corrected — the sign and units problem in the Kongkeaw abstract is reported in section 11 for that reason, and the effect is described as a direction rather than a magnitude.

Populations are named every time, because the central failure mode of this literature is transferring a result from an impaired population to a healthy reader. Every reference resolves to a record retrieved from NCBI or to an agency or database page whose URL returned HTTP 200 and the expected document title at build. Project 06 was consulted read-only as a general-science corpus and is not the evidence spine for this title; no write was made to it or to project06_catalog.sqlite. Project 07 was used for trade-claim discovery only. News reporting is not treated as scientific evidence anywhere in this document.

Adversarial interrogation — the required adversarial challenge list — is disposed as follows.

Which claims are overstated? ACCEPTED as the organising question. The memory claim for Bacopa, the enhancement claim for α-GPC, the current-product claim for phosphatidylserine, and the neuroprotection claim for ginkgo are each named and refused in the sections that report their evidence.

What contrary trials exist? ACCEPTED and given priority. GEM and GuidAge are reported at length rather than summarised as mixed evidence (DeKosky et al., 2008; Snitz et al., 2009; Vellas et al., 2012), and the soy phosphatidylserine null is given equal weight to the bovine positive (Jorissen et al., 2001).

Are effect sizes clinically meaningful? ACCEPTED as a standing doubt. Acetyl-L-carnitine's 0.20, creatine's 0.29, FINGER's 0.022 z per year, and COSMOS-Mind's 0.07 z are reported as numbers, in context, and described as small.

Is the evidence population relevant? ACCEPTED as the article's spine. Section 05 states the three-population problem and every subsequent grade names its population.

Is there publication bias? ACCEPTED as a structural expectation given the sample sizes in section 04, and not asserted as a detected funnel-plot finding in any specific synthesis, because none of the syntheses cited here reported one.

Are commercial claims stronger than the evidence? ACCEPTED. Figure 3 states the gap claim by claim.

Are mechanisms being mistaken for outcomes? ACCEPTED. Table 2 separates the two columns explicitly; α-GPC's established pharmacokinetics beside its absent healthy-adult outcome data is the reference case.

Are safety concerns underrepresented? ACCEPTED. The centrophenoxine analysis, the ashwagandha liver signal, and Mucuna's dopaminergic activity are reported in Table 3 rather than in a closing caveat.

Are doses and formulations genuinely comparable? ACCEPTED and treated as decisive twice: bovine-cortex against soy phosphatidylserine, and SHR-5 against Rhodiola generally.

Does any paragraph imply certainty not supported by evidence? ACCEPTED as a drafting constraint. Every efficacy statement carries an in-place grade, and "not supported" is used where the population that buys a product was not studied.

Unresolved science remains substantial: an adequately powered randomised trial of any single "nootropic" botanical in healthy adults with a pre-registered composite primary endpoint; a caffeine-matched comparator arm in any commercial blend; a replication of the lion's mane trial at adequate size; a randomised test of α-GPC in people without cerebrovascular disease; confirmation of the COSMOS-Mind multivitamin finding in a more diverse cohort; and a prevalence estimate for undeclared pharmaceuticals across the retail category rather than within a targeted sample.