The Problem With Generic Supplement Advice
Population averages hide the variation that determines whether a compound helps you, harms you or does nothing, and four sources of that variation are large enough to invert a recommendation.
The Short Answer
A trial reports an average. That average is the sum of people who improved, people who did not change and people who got worse, and the size of that spread determines how much a population recommendation tells any individual. For several common supplements the spread is wide enough that the same compound at the same dose is genuinely useful for one person and pointless or harmful for another. Four sources of variation account for most of it.
Variation One: Baseline Status
This is the largest and most consistent source, and it explains a striking pattern across the supplement literature: compounds that work in people with low status repeatedly fail in people who are already replete.
Vitamin D. Correction in people with low status shows benefits across several endpoints. Supplementation in replete populations has repeatedly failed to reproduce them, which is why large trials in generally sufficient populations returned neutral results.
Iron. Correcting low stores improves fatigue and exercise capacity. Supplementing an adequate person achieves nothing and risks harm, since iron absorption is regulated and excess is damaging.
Zinc. Correction where intake is low raises testosterone; supplementation in adequate men does not.
Omega-3. Trials enrolling populations with already adequate intake have less room to show benefit, which is one proposed explanation for the heterogeneity in cardiovascular outcome trials.
The practical implication is that the question "does vitamin D work" is not answerable without knowing whose vitamin D. A recommendation that does not account for status is averaging across two groups for whom the answer differs.
Variation Two: Genetics and Metabolism
| Variant or trait | Effect on response |
|---|---|
| CYP1A2 variants | Caffeine metabolism rate; slow metabolisers experience longer effects and greater sleep disruption |
| MTHFR variants | Modest effect on folate handling; common enough that it is not a strong individual determinant |
| Lactase persistence | Determines dairy tolerance entirely |
| HFE variants | Haemochromatosis risk; makes iron supplementation genuinely hazardous |
| APOE status | Affects lipid response to dietary fat in some analyses; associated with cognitive risk |
| Equol-producing capacity | Microbial rather than genetic; determines response to soy isoflavones |
| Urolithin A conversion | Microbial; only some people convert dietary ellagitannins |
Two of these are worth acting on and most are not. HFE variants matter because iron supplementation in an undiagnosed carrier is genuinely harmful, and CYP1A2 variation matters practically for caffeine timing, though most people can work it out from experience more reliably than from a test.
The microbial entries are the more interesting category, because they mean response to a food-derived compound depends on which bacteria you carry rather than on your genome. Equol and urolithin A are the clearest examples, and they explain why some people respond to soy or pomegranate compounds and others do not.
What this does not support is the elaborate nutrigenomic testing sold on the strength of these variants. Most have small effects, most are common in healthy people, and most do not change what a person should do.
Variation Three: Medications and Conditions
This is where generic advice moves from unhelpful to unsafe, and it is the variation that a population recommendation cannot accommodate at all.
Anticoagulants. High-dose omega-3, ginkgo, garlic, vitamin E and several others affect bleeding. A generic recommendation to a population containing people on warfarin or direct oral anticoagulants is a recommendation to a group for whom it is contraindicated.
CYP3A4 and P-glycoprotein. Berberine, quercetin, St John's wort and grapefruit constituents affect the metabolism of a long list of medications, including statins, immunosuppressants, some anticoagulants and many others.
Levothyroxine. Calcium, iron and magnesium reduce absorption, which is a frequent cause of unexplained dose requirements.
Kidney impairment. Changes the handling of magnesium, potassium and several other compounds, and potassium supplementation can be dangerous.
Immunosuppression. Live probiotics carry documented risks, and immunostimulant products are inappropriate.
Hormone-sensitive conditions. Phytoestrogenic products and precursor hormones require clinical input.
Pregnancy. Several common supplements are contraindicated, and vitamin A in excess is teratogenic.
No population recommendation can check these, which is why an interaction check is the highest-value function any personalisation system performs.
Variation Four: Goal, Age and Context
The same intervention has different value depending on who is receiving it, and in some cases the sign of the effect reverses.
Protein and IGF-1. Lower protein intake may be favourable in midlife from an IGF-1 perspective, and higher intake is clearly favourable after 65 where sarcopenia risk dominates. The same recommendation is wrong for one of those groups.
Caloric restriction. Improves metabolic markers in midlife and threatens bone and lean mass in later decades, where the net effect can invert.
Hormone recommendations. Differ entirely by sex and life stage, and a general hormone protocol would be wrong for most people receiving it.
Training emphasis. Power and balance become primary rather than supplementary after 60, and a general fitness recommendation misses that.
Antioxidant timing. Irrelevant to someone who does not train, and consequential for someone training for adaptation.
Fasting protocols. Reasonable for some people, contraindicated for anyone with a history of disordered eating, on glucose-lowering medication, pregnant, or at risk of sarcopenia.
These are not marginal adjustments. They are cases where a recommendation appropriate for one group is actively counterproductive for another, which is the strongest argument against a single protocol.
What Generic Advice Gets Right
Being fair about this matters, because the case for personalisation is weakened by overstating it.
A great deal of good health advice is genuinely universal. Do not smoke. Sleep regularly and adequately. Train for strength, aerobic capacity, power and balance. Eat mostly unprocessed food with adequate protein and fibre. Avoid chronic energy surplus. Moderate alcohol. Maintain social connection. Address blood pressure. None of that requires personalisation, and all of it outperforms any personalised supplement stack.
The universal advice also covers most of the available benefit. A person following it well is capturing the large majority of what is achievable, and personalisation operates at the margin.
Where personalisation genuinely earns its place is narrower and specific: correcting a measured shortfall rather than a presumed one, checking interactions against actual medications, matching the recommendation to age and life stage, and detecting whether a change worked for this person rather than for a trial average.
That is a modest claim and a defensible one. Anything larger, a genetic test producing a bespoke stack, an algorithm optimising a proprietary score, tends to be personalising the small variable while the universal advice goes unfollowed.
The Honest Version of Personalisation
Given all of the above, a personalisation system that is not overclaiming does four things.
Measures before recommending. Since baseline status is the largest source of variation, a recommendation without measurement is guessing which of two groups you are in.
Checks interactions exhaustively. The highest-value and least visible function, because the pairs run to thousands and no practitioner holds them in memory.
Accounts for age, sex and life stage, since these change the direction of some recommendations rather than only their magnitude.
Observes response over time, changing one thing and re-measuring, because individual response is the one thing no population study can supply.
And it does one negative thing: it declines to recommend where the evidence does not support a recommendation either way. A system that always finds something to suggest is not evaluating, and the willingness to return nothing is the clearest signal that it is.
Everything else in this cluster is a description of how those four functions are built, and this article is the reason they are worth building.
The AEONNN Perspective
This article states the problem AEONNN exists to address, and the honest version of it is narrower than the market's. Population averages hide four sources of variation large enough to change a recommendation: baseline status, genetics and microbial capacity, medications and conditions, and age with life stage.
Baseline status is the largest, which is why the platform measures before recommending. The pattern recurs across the Evidence layer: compounds that work in people with low status repeatedly fail in the replete, so a recommendation made without measurement is guessing which of two groups a member is in.
The Safety layer's interaction checking is the highest-value function and the least visible, because the documented pairs run to thousands and no practitioner holds them in memory. And the Population layer carries the cases where the sign of an effect reverses with age, protein and IGF-1, restriction against lean mass, which is why AEONNN will not run a single protocol across life stages. The platform is also explicit that the universal advice captures most of the available benefit, and that personalisation operates at the margin.
Pillar Matrix mapping
Database Matrix layers
- Population Layer (UK Biobank, NHANES)
- Evidence Layer (PubMed, Cochrane, ClinicalTrials.gov)
- Safety Layer (DrugBank, FAERS)
- Meta / Consensus Layer (JAMA, BMJ, specialty society positions)
Frequently Asked
Why does generic supplement advice fail?
Because a trial reports an average that sums people who improved, did not change and got worse. Four sources of variation, baseline status, genetics, medications and age, are large enough to change the recommendation.
What is the largest source of variation?
Baseline status. Compounds that work in people with low vitamin D, iron or zinc status repeatedly fail in people who are already replete, which is why many large trials returned neutral results.
Does genetic testing help personalise supplements?
Rarely. Most variants have small effects, are common in healthy people and do not change what someone should do. HFE variants matter because iron supplementation becomes hazardous.
What role does the microbiome play?
It determines response to some food-derived compounds. Equol-producing capacity governs soy isoflavone response, and only some people convert dietary ellagitannins to urolithin A.
Why do medications matter so much?
Because generic advice cannot check them. High-dose omega-3 and several botanicals affect bleeding, and berberine, quercetin and St John’s wort affect the metabolism of many medications.
Can one recommendation be wrong for some people?
Yes, and sometimes the sign reverses. Lower protein may favour midlife IGF-1 goals while higher protein clearly favours function after 65, so the same advice is wrong for one group.
What is generic advice good for?
A great deal. Not smoking, sleeping regularly, training across four qualities, unprocessed food with adequate protein and fibre, avoiding surplus, moderating alcohol and addressing blood pressure are universal and capture most of the benefit.
Evidence and review
Any dosage ranges cited here reflect the ranges used in published human trials, not personal recommendations. Evidence in this field moves, so this article is reviewed quarterly and carries its last-updated date above. Nothing here is intended as medical advice, and supplementation should be discussed with a qualified clinician, particularly alongside prescribed medication or an existing condition.