Andrew Huberman's Diary of a CEO Supplement Stack, Graded Against the Evidence

On Diary of a CEO, Steven Bartlett asked Andrew Huberman the podcast-guest version of a desert-island question: five supplements for the rest of your life, go. Huberman named magnesium, omega-3, vitamin D, alpha-GPC and creatine, then spent the next few minutes on something that was not a supplement at all: walking after you eat.

Huberman is a working neuroscientist and he mostly cites real papers. That is exactly why the claims are worth checking rather than repeating. A correct citation delivered with total confidence can still be doing more work than the underlying paper supports, and one segment here is a clean example of that. Here is each claim against the primary research, plus the one habit that turned out to be the best-supported thing in the whole episode.

The five supplements, graded

Magnesium: real, but not quite "interchangeable"

Huberman's line: magnesium threonate and magnesium bisglycinate are interchangeable for sleep, and only worth taking if you are not already sleeping well. See our magnesium forms comparison for the full breakdown of magnesium forms.

• What holds up: both forms now have their own randomised, placebo-controlled trial in adults with self-reported sleep problems. Magnesium L-threonate at 1g/day for 21 days (80 adults, 35 to 55) improved sleep quality measures including mental alertness and mood on waking. Magnesium bisglycinate at 250mg elemental magnesium/day for 4 weeks (155 adults) significantly improved Insomnia Severity Index scores over placebo.
• The caveat: "interchangeable" overstates it. The bisglycinate trial's effect size was small (Cohen's d = 0.2), which is a real but modest benefit, not a knockout. The two forms are also not mechanistically equivalent: threonate is marketed on crossing into brain tissue more readily, bisglycinate on being gentle on the gut and well absorbed generally. Both have genuine trial support for sleep specifically. Neither has been shown to beat the other head to head.

Grade: directionally right. The caveat Huberman states himself, that this is only worth doing if your sleep is not already good, is the correct frame and matches the evidence better than "interchangeable" does.

Omega-3: the dose advice is sound, the flagship trial is shakier than it sounds

Huberman's line: at least 1g/day of EPA, and prescription-strength icosapent ethyl (branded Vazkepa in the UK and Europe, Vascepa in the US, which is likely what he means by "Lavazza") is the highest-concentration, most heavy-metal-free option. See our EPA vs DHA breakdown for how omega-3 dosing changes by goal.

• What holds up: the REDUCE-IT trial found a 25% relative risk reduction in major cardiovascular events with 4g/day of icosapent ethyl, and prescription EPA genuinely is higher-concentration and more consistently tested for purity than most over-the-counter fish oil. 1g/day combined EPA+DHA as a general target is in line with what most health bodies recommend.
• The caveat, and it is a real one: REDUCE-IT used mineral oil as the placebo, and mineral oil is not metabolically inert. At 12 months, the placebo arm showed a 10.9% rise in oxidised LDL, a 16.2% rise in interleukin-6 and a 21.9% rise in hs-CRP, three markers moving in the wrong direction with no active drug in them. Several cardiology researchers have since argued that some of icosapent ethyl's apparent benefit may be the placebo arm getting worse rather than the drug arm getting better, and have called for a trial against a genuinely neutral comparator like corn oil. That has not happened yet.

Grade: the dosing guidance is fine and worth following. The specific trial behind the strongest version of the claim is under real, unresolved scientific dispute, which "very very inexpensive" and a doctor-threatening joke glosses over entirely.

Vitamin D: right that most people under-dose, wrong to skip the blood test

Huberman's line: 5,000 to 10,000 IU/day, more than the RDA but not so much you "blow a gasket", justified by self-monitored bloodwork. See our vitamin D dosing guide for why the label IU and the RDA rarely match. Vitamin D3 ingredient page has the forms breakdown.

• What holds up: he is right that most people are under-dosing relative to what keeps 25-hydroxyvitamin D in a healthy range, especially at higher latitudes or with more body fat. The official upper limit (4,000 IU/day, set by the US Institute of Medicine and the European Food Safety Authority) is genuinely conservative by some researchers' reading, and case reports of hypercalcaemia, the actual harm from too much vitamin D, are rare below 10,000 IU/day.
• The caveat: the safe way to land at 5,000 to 10,000 IU/day is the one step Huberman's own account skips for the listener: test your 25(OH)D level first, dose toward a target (most guidance aims for roughly 75 to 125 nmol/L, or 30 to 50 ng/mL), and retest. "My blood's getting better" on self-directed 5,000 IU is an n-of-1 anecdote from someone who happens to also get regular bloodwork. Doing the dose without the test is doing half of his own protocol.

Grade: correct on the direction, incomplete on the method. Take the headline (most people are under-dosed) and leave the number (5,000 to 10,000 IU) until you have a test that says you need it.

Alpha-GPC: real mechanism, unsettled safety story, and an unproven fix

Huberman's line: 600 to 900mg of alpha-GPC as a stimulant-sparing focus aid, which also increases something called TMAO and raises cardiovascular risk, mitigated by taking a 600mg garlic capsule alongside it.

• What holds up: alpha-GPC is a choline source and choline is a genuine precursor to acetylcholine, the neurotransmitter tied to focus and to REM sleep. Some trial data does show increased REM density at higher doses. The TMAO mechanism he names is also real biochemistry: gut bacteria convert dietary choline into trimethylamine, the liver oxidises it into TMAO, and TMAO is an established marker linked to atherosclerosis.
• The caveat that matters most: a South Korean nationwide cohort study found that people prescribed alpha-GPC had a significantly higher risk of total, ischaemic and haemorrhagic stroke than non-users, with risk increasing the longer someone had been taking it and highest past 12 months. That population was mostly older adults prescribed alpha-GPC for cognitive symptoms, not healthy adults taking it for focus, so it may not transfer directly. But it is a real enough signal that researchers are actively debating it, not a settled non-issue.
• The garlic fix specifically has no trial behind it: garlic and allicin do have general anti-atherosclerotic research, but nothing has tested whether a garlic capsule offsets alpha-GPC's TMAO or stroke signal in humans. That is an extrapolation dressed as a solution.

Grade: the weakest of the five. The focus mechanism is plausible and the stroke data is not proof of harm in a young, healthy user. But presenting a genuinely unresolved safety question as solved by a garlic capsule is the one place in the episode where confidence outran the evidence.

Creatine: solid, at half the dose he mentioned

Huberman's line: 5 to 10g/day, useful for cognition especially under sleep deprivation or stress. See our full creatine beyond the gym piece for the cognition, ageing and mood evidence.

• What holds up: creatine is the most-studied supplement that exists, and the cognitive benefit under sleep deprivation and mental fatigue is genuinely one of its better-supported non-gym uses. Huberman's framing here is accurate.
• The caveat: the trial evidence, including a 2024 meta-analysis of 23 trials, supports 3 to 5g/day. Once muscle and brain creatine stores are saturated, more does not do more. 10g/day mainly buys extra GI side effects, not extra benefit.

Grade: directionally correct, dosed high without a stated reason.

The walk after dinner: the best-supported claim in the episode

The soleus push-up segment is the one part of this episode that undersells itself. Huberman calls it "not practical" and moves straight to the version you can actually do: walk for 5 to 10 minutes after you eat.

• The mechanism is real and well replicated: a 2022 systematic review and meta-analysis in Sports Medicine (7 randomised trials) found that breaking up sitting with short light-intensity walking reduced post-meal blood glucose by roughly 17% compared with staying seated. Standing alone helped too, by about 9.5%, but walking beat standing.
• The soleus push-up study is real, and the numbers are striking for how small the study was: Hamilton, Hamilton and Zderic (iScience, 2022) had 15 people perform slow, non-weighted seated calf raises for 3 hours during an oral glucose tolerance test. Blood glucose excursion dropped by 52% and insulin requirement dropped by 60% compared to sitting still. Between meals, the same movement roughly doubled fat oxidation and reduced circulating VLDL triglycerides. This is a small, single-lab pilot study, not yet a large replicated trial, which is exactly the caveat Huberman himself gives it ("not practical") before recommending you just go for a walk instead.

The metabolism point is also correctly cited: the paper he references is Pontzer et al., published in Science in 2021, which measured total daily energy expenditure in over 6,400 people aged 8 days to 95 years across 29 countries using doubly-labelled water, the gold-standard method. Adjusted for lean mass, energy expenditure is essentially flat from roughly age 20 to 60, then declines gradually after 60. The popular idea that metabolism quietly slows through your 30s and 40s is not what this dataset shows. What actually differs between people is how much they move.

Grade: strongest claim in the episode, correctly cited, and free. If you take one thing from this whole segment, this is the one with the least daylight between what was said and what the research shows.

One more thing: the CO2 aside

Near the end, Huberman and Bartlett note their studio's CO2 monitor climbing past 1,700 parts per million and link it to feeling foggy, citing research that cognitive performance drops above 1,000 ppm.

That research exists and is real (Satish et al., 2012, found large drops in decision-making scores at 2,500 ppm), and a 2016 Harvard follow-up partially supported it. But it has not settled: a Danish research group found no cognitive decline even at 5,000 ppm, and other labs have failed to reproduce the low-concentration effect, with some proposing that regular exposure builds physiological adaptation. Worth cracking a window. Not yet a number you should treat as fixed.

The short version

Two of the five supplements hold up close to as stated: magnesium (with a smaller effect than "interchangeable" implies) and creatine (at half his stated dose). Two need a real caveat before you act on them: omega-3's flagship trial has a genuine placebo problem, and vitamin D's right number depends on a blood test he does not mention taking himself. One is presented with more confidence than the underlying safety data currently supports: alpha-GPC's stroke signal is real enough to be actively debated, and the garlic fix is not backed by any human trial.

The single best-supported claim in the entire 12 minutes was not a supplement. It was walking after you eat.

None of this is medical advice. If you take a blood thinner, are pregnant, or are titrating vitamin D or omega-3 for a diagnosed condition, that is a conversation with your doctor, not a podcast clip. This piece has no relationship with Andrew Huberman, Steven Bartlett or Diary of a CEO; it is an independent read of claims made in a publicly released episode.

How Stack Almanac helps

If you are going to run any of this yourself, the two variables that actually distinguish a real effect from a good week are dose and consistency, which is exactly what a podcast clip cannot track for you. Log magnesium, omega-3 or vitamin D3 at the dose you actually took, and the Almanac Advisor flags when a dose sits well above what the cited trials used, rather than telling you it is fine because a podcast said so.

If you pair alpha-GPC with a wearable, the Advisor can hold your HRV, sleep and resting heart rate data next to your logged dose and tell you whether anything is actually moving, rather than leaving you to guess from how focused you felt on a Tuesday. And if you log a 25(OH)D result, vitamin D dosing gets tied to your actual number instead of a round figure someone else takes.