Supplement Studies  /  evidence, scored
Creatine8 claimsMagnesium4 claimsVitamin D4 claimsVitamin E4 claimsOmega-34 claimsMelatonin4 claimsAshwagandha4 claimsCurcumin4 claimsCoQ104 claimsBerberine4 claimsGlucosamine & Chondroitin4 claimsMultivitamins4 claimsAlpha-Lipoic Acid4 claimsAcetyl-L-Carnitine4 claimsNMN4 claimsNicotinamide Riboside4 claimsBiotin4 claimsSt John’s Wort4 claimsRed Yeast Rice4 claimsPsyllium Fibre4 claimsBovine Colostrum4 claimsZinc4 claimsVitamin B124 claimsResveratrol4 claimsTaurine4 claims

25 supplements · 104 claims scored · 18 more in review

Evidence File 014 · Creatine monohydrate · Method v0.18 of 8 claims researched

CreatineMonohydrate · 8 claims

Every claim below was researched against retrieved sources. Each carries a note comparing the researched score to the one generated before any of the reading was done.

8Claims scored
18Sources
3–5 gStudied dose
Jul 26Reviewed

Claim 01

Increases muscle strength
7.3/10
Moderate·69 RCTs · 1,937 participants · band 6.5–8.3

Real, smaller than commonly claimed, and highly uneven. Squat +5.64 kg (95% CI 3.87–7.40, p<0.001) is the most trustworthy figure on this page — low heterogeneity, Egger's p=0.92, no publication bias. Bench press +1.43 kg (95% CI 0.53–2.34, p=0.002) is significant but small, and Egger's flagged publication bias at p=0.0003. Leg press (p=0.11) and handgrip (p=0.10) were not significant at all.

The subgroups break the claim. In over-50s: no significant gain on bench (p=0.26), leg press (p=0.71) or squat (p=0.73). In women: bench press reached significance at +0.15 kg — 150 grams — with leg press, squat, jump and power all null. In untrained participants: bench (p=0.19) and squat (p=0.70) not significant. Without a training programme the bench effect was negative.

Study quality was high (PEDro 7–10 of 11; 51 of 69 studies scored 9+), PRISMA-compliant, pre-registered on PROSPERO. Trim-and-fill imputed no missing studies and estimates held under leave-one-out.

The working

9Quantity
7Quality
5Consistency
9Directness

(9 × 0.25) + (7 × 0.35) + (5 × 0.25) + (9 × 0.15) = 7.3
Quality docked for publication bias on the primary upper-body outcome. Consistency docked to 5: direction is not maintained across sex, age or training status.

What the research changed

Invented: 9.2 / Strong, “+5–15% on 1RM”, “both sexes, all adult age groups.” All three wrong, all three in the direction that flatters the supplement.

Not scorable as one claim. 8.3 for trained young men, 6.6 for everyone else. Needs two rows.

Claim 02

Increases lean body mass, with resistance training
8.5/10
Strong·143 studies · ~3,600 participants · GRADE high

The largest synthesis available: 143 studies, roughly 3,600 participants, over 160 effect sizes. Creatine added +0.82 kg of fat-free mass versus placebo, at GRADE-high certainty with zero statistical heterogeneity. That combination is rare in supplement research.

The gain was significant only when paired with resistance or combined training. Supplementation alone produced no measurable change. Body fat percentage fell 0.28% — but fat mass itself did not move; the percentage shifted because lean tissue grew. Loading at 20 g/day then dropping to 3–5 g produced the same end result as 3–5 g from day one.

The water question is not resolved. Trials used DXA, bioimpedance, underwater weighing and calipers interchangeably, and none of these distinguish contractile tissue from fluid drawn into the cell. The authors also note most trials treated body composition as a secondary outcome and few measured baseline intramuscular creatine or dietary intake.

The working

10Quantity
8Quality
9Consistency
6Directness

(10 × 0.25) + (8 × 0.35) + (9 × 0.25) + (6 × 0.15) = 8.5
Directness is the limiter at 6. Fat-free mass was estimated by DXA, bioimpedance and skinfolds across trials — none of which separates muscle tissue from intracellular water, which is exactly what the claim turns on.

What the research changed

Invented: 8.6, “+1–2 kg over 4–12 weeks.” The real figure is +0.82 kg — roughly half the bottom of my range.

I got the band right by accident. The reasoning underneath it was wrong.

Claim 03

Improves repeated high-intensity effort and power
7.7/10
Moderate·Wingate: 12 studies · jump: 23 studies

Wingate peak power rose +47.8 W (95% CI 15.6–80.1, p=0.004) with no significant publication bias (Egger's p=0.07). Vertical jump rose +1.48 cm (95% CI 0.30–2.66, p=0.01), though heterogeneity was high at I²=70.8%.

Same population pattern as strength. Males gained +55.3 W on Wingate; females came back at −10.9 W with a confidence interval spanning more than 400 watts — three trials, essentially uninformative. Trained participants gained +74.8 W; untrained participants +21.4 W, not significant.

Higher maintenance doses (>8 g/day) outperformed lower ones on power outcomes specifically (+93.2 W vs +27.3 W non-significant), which is the one place in this dataset where dose clearly mattered.

The working

8Quantity
8Quality
6Consistency
9Directness

(8 × 0.25) + (8 × 0.35) + (6 × 0.25) + (9 × 0.15) = 7.7
Consistency docked to 6: no publication bias on either outcome, but vertical jump heterogeneity was high (I²=70.8%) and both outcomes were null in women and in untrained participants.

What the research changed

Invented: 8.8, “Strong evidence, athletes.” Closer than my other guesses, but I asserted it without the female and untrained nulls, which are the part a reader actually needs.

Claim 04

Supports cognition under sleep deprivation
4.1/10
Emerging·Very small trials · wide band · EFSA rejected the health claim

The European Food Safety Authority assessed this claim formally in 2024 and rejected it. Of the trials submitted, the sleep-deprivation study used 20 g/day for seven days in sports science students — ten per group, with the basis of allocation unclear and blinding methods not reported. EFSA accepted that it showed an effect on working memory after 36 hours of sleep deprivation combined with moderate-intensity exercise, and no effect elsewhere.

A 2024 trial using a single high dose of 0.35 g/kg during 21 hours of sleep deprivation reported improved cognitive performance alongside changes in brain energy availability. That is a genuinely interesting result and it is one study.

This is the honest shape of the claim: a plausible mechanism, two or three small suggestive trials, and a regulator that looked at the file and declined to endorse it.

The working

3Quantity
3Quality
5Consistency
7Directness

(3 × 0.25) + (3 × 0.35) + (5 × 0.25) + (7 × 0.15) = 4.1
Quantity and quality both score 3. The founding trial had ten participants per arm, unclear allocation and unreported blinding methods.

What the research changed

Invented: 5.4, sourced to a 2018 review and a 2003 vegetarian trial. Roughly the right zone, entirely the wrong evidence. I did not know the EFSA opinion existed, and it is the most consequential document on this claim.

Claim 05

Improves cognition in healthy, rested adults
3.8/10
Weak·Two flagship meta-analyses, both with documented statistical errors

This is the most instructive claim on the page, because the decisive evidence is not in the trial literature at all.

Two meta-analyses reported positive memory effects — one covering 16 RCTs and 492 participants, another finding benefit concentrated in older adults. A letter to the editor then demonstrated that both had pooled multiple non-independent cognitive tests from the same participants into single effect estimates, inflating sample sizes and manufacturing false positives. When the authors re-analysed their own data correctly, the overall memory effect disappeared except in older adults.

EFSA raised the identical objection about the other meta-analysis and concluded that no conclusions could be drawn from it. One of the researchers behind the original sleep-deprivation work has since published a systematic review arguing the literature fails to support the theoretical basis for a cognitive effect at all.

In older adults specifically the signal survives correction and would score around 5.2 / Emerging — another claim that needs splitting by population.

The working

6Quantity
2Quality
2Consistency
7Directness

(6 × 0.25) + (2 × 0.35) + (2 × 0.25) + (7 × 0.15) = 3.8
Quality scores 2 and consistency 2. Both pooled estimates supporting this claim have been shown to double-count participants; the corrected re-analysis is null.

What the research changed

Invented: 3.1 / Weak. The band was right and the reasoning was hollow — I would have cited exactly the meta-analyses that have since been discredited, because they are what the ambient literature points at.

Note also that one of these papers carries a corrigendum for a Chinese-to-English translation error. Nothing in the citation string tells you that.

Claim 06

Useful as an adjunct in depression
3.7/10
Weak·11 trials · 1,093 participants · GRADE very low

The 2025 synthesis pooled 11 trials and 1,093 participants: SMD −0.34 (95% CI −0.68 to −0.00), GRADE very low certainty. Translated onto the 17-item Hamilton scale that is 2.2 points — below the 3.0-point minimal important difference. The confidence interval touches zero.

Heterogeneity was substantial (I²=71.3%), and subgroup and trim-and-fill analyses indicated substantial bias favouring creatine. The secondary endpoints contradict each other: remission was significant across three trials (OR 3.60, 95% CI 1.76–7.56) while treatment response across two trials was not (OR 0.72, 95% CI 0.28–1.88).

The authors' own conclusion is the fair summary: a possible small-to-moderate benefit, average effects not clinically important, and a true effect that may be trivial or null. Trial-level detail is thinner still — a separate 2025 review of creatine across mental disorders found five RCTs totalling 238 participants, of which only two were low risk of bias.

The working

6Quantity
2Quality
3Consistency
5Directness

(6 × 0.25) + (2 × 0.35) + (3 × 0.25) + (5 × 0.15) = 3.7
Directness scores 5: the outcome measure is correct, but the pooled effect sits below the threshold at which a patient would notice a difference.

What the research changed

Invented: 3.8, “promising, unreplicated.” The number happens to land within 0.1 of the researched score. That is luck, not method — and it is the most dangerous outcome of the whole exercise, because a reader comparing the two pages would conclude the invented one was reliable.

Claim 07

Causes hair loss
1.9/10
Evidence against·Narrow band — now tested directly

The claim traces to one 2009 trial of 20 college-aged rugby players taking 25 g/day for a week then 5 g/day for two. DHT rose 56% during loading and stayed 40% above baseline afterwards. The study measured hormones. It never measured hair.

A 2021 systematic review pooling 12 studies and 276 participants found no significant change in total testosterone, free testosterone or DHT. The 2009 result stands alone.

The claim has now been tested directly. A 12-week randomised controlled trial published in 2025 measured hair density, follicular count and cumulative hair thickness — the first study to assess hair follicle health under creatine rather than infer it from a hormone marker. No difference on any hair outcome, and none on testosterone, free testosterone, DHT or the DHT:testosterone ratio.

The working

3Quantity
2Quality
1Consistency
1Directness

(3 × 0.25) + (2 × 0.35) + (1 × 0.25) + (1 × 0.15) = 1.9
Scored as support for the claim. Directness is 1: the founding study measured a hormone, not hair. The band is narrow because a direct trial now exists.

What the research changed

Invented: “no trial has since measured actual hair loss and found an effect.” True when I was trained, false now. The direct trial is the most important source on this claim and I did not know it existed.

The “20 rugby players” detail I stated confidently was correct. From the inside, the accurate details and the stale ones were indistinguishable.

The 2009 paper's DOI appears in two different forms across secondary sources. Resolve against the journal record before publishing.

Claim 08

Damages the kidneys in healthy adults
1.6/10
Evidence against·685 trials reviewed · convergent independent methods

A 2025 analysis reviewed side effects reported across 685 human clinical trials plus worldwide adverse event report databases and found no consistent evidence of renal harm. Separate meta-analyses of renal function in 2019 and 2025 reached the same conclusion, as did a 2024 Mendelian randomisation study — which matters because it uses genetic instruments rather than trial data and therefore carries a completely different bias structure. Convergence across independent methods is the strongest form of this evidence.

The mechanism of the myth is confirmed. Creatine raises serum creatinine, which is the marker used to estimate kidney function. A depression trial recorded exactly this: creatinine rose in the creatine group only (0.79 vs 0.67 mg/dL, p=0.03) with no renal pathology. The blood test looks worse; the kidney is fine. Tell your GP you take it.

Funding flag. The 685-trial safety analysis had its publication charges paid by Alzchem Group AG, a creatine manufacturer, as part of a conference special issue. This does not invalidate it and it is disclosed in the paper. It is recorded here because the method requires it, and because the independent Mendelian randomisation result is what makes the conclusion safe without it.

The working

2Quantity
2Quality
1Consistency
1Directness

(2 × 0.25) + (2 × 0.35) + (1 × 0.25) + (1 × 0.15) = 1.6
Scored as support for the claim. The band is narrow because three methodologically independent lines of evidence converge.

What the research changed

Invented: 1.0, citing a 2003 trial and a 2021 review. Directionally right, and I missed both the 685-trial analysis and the Mendelian randomisation work — the two things that actually make this conclusion robust.

I also would not have flagged the industry funding, because I did not know it was there.

Dosage as studied

Most-studied regimen
3–5 g per day

Every day, including rest days.

Loading
Optional. The dose-response meta-analysis found loading at 20 g/day then dropping to maintenance produced the same body composition outcome as starting at 3–5 g.
Form
Monohydrate. Of 95 trials measuring fat-free mass, 89 used monohydrate; alternative forms have too few studies to analyse, and cost more.
Higher doses
Above 8 g/day showed larger effects on power outcomes specifically (+93 W vs +27 W). No corresponding advantage on strength or lean mass.
Training
Required. Without resistance training the lean mass effect was null and the bench press effect was negative.