Gym ProductReview

Home / How we rate

Method

How we rate

Published in full, because a rating you cannot audit is just an opinion with a number attached.

A weighting per shelf, not one formula

A protein bar, a pre-workout and an electrolyte mix are not comparable on the same axes, so they are not scored on the same axes. Every ranking carries a weighting chosen for its category (the protein shelves and energy drinks share one), all of them below, and each produces a score out of ten. An electrolyte powder has no dose-integrity component at all, because auditing a hydration mix against reference doses written for 15-active formulas would measure nothing. A weighting does not make an axis scoreable, though: the saw palmetto weighting includes dose integrity, but none of the 10 saw palmetto products on record has a dose-integrity score, because none of the actives on their labels has a reference dose, and no saw palmetto ranking is published.

Protein PowdersWeightBasis
Value40%Cost per gram of protein, ranked against the category
Formula35%Protein per calorie, sugar alcohol load, and 1.5 points off for each amino-spiking flag
Consensus25%Buyer rating shown on the brand’s own product page (read from the page, or from the review service that draws its stars; source-linked), minimum 25 reviews, re-read within the last 30 days, count, source and capture date printed on every product page
Protein BarsWeightBasis
Value40%Cost per gram of protein, ranked against the category
Formula35%Protein per calorie, sugar alcohol load, and 1.5 points off for each amino-spiking flag
Consensus25%Buyer rating shown on the brand’s own product page (read from the page, or from the review service that draws its stars; source-linked), minimum 25 reviews, re-read within the last 30 days, count, source and capture date printed on every product page
Protein SnacksWeightBasis
Value40%Cost per gram of protein, ranked against the category
Formula35%Protein per calorie, sugar alcohol load, and 1.5 points off for each amino-spiking flag
Consensus25%Buyer rating shown on the brand’s own product page (read from the page, or from the review service that draws its stars; source-linked), minimum 25 reviews, re-read within the last 30 days, count, source and capture date printed on every product page
Pre-WorkoutsWeightBasis
Dose integrity45%Each audited active’s disclosed dose as a share of its reference dose (capped at the reference dose), averaged and put on a 10-point scale, minus 2 points for a proprietary blend
Value30%Cost per fully dosed serving, ranked against the category
Consensus25%Buyer rating shown on the brand’s own product page (read from the page, or from the review service that draws its stars; source-linked), minimum 25 reviews, re-read within the last 30 days, count, source and capture date printed on every product page
Energy DrinksWeightBasis
Value40%Cost per can (per serving for powder tubs and stick packs), ranked against the category
Formula35%Sugar load, calorie load, and whether the panel discloses its caffeine content
Consensus25%Buyer rating shown on the brand’s own product page (read from the page, or from the review service that draws its stars; source-linked), minimum 25 reviews, re-read within the last 30 days, count, source and capture date printed on every product page
CreatineWeightBasis
Value45%Cost per gram of creatine, ranked against the category — not per gram of everything else in the tub, so a creatine sold with HMB or betaine answers on the creatine you are actually buying
Dose integrity20%Each audited active’s disclosed dose as a share of its reference dose (capped at the reference dose), averaged and put on a 10-point scale, minus 2 points for a proprietary blend
Label disclosure20%Share of the actives the label puts a number against. An amount hidden in a proprietary blend, or simply not stated, scores zero here rather than counting as missing data — non-disclosure is a fact about a label, not an absence of one
Consensus15%Buyer rating shown on the brand’s own product page (read from the page, or from the review service that draws its stars; source-linked), minimum 25 reviews, re-read within the last 30 days, count, source and capture date printed on every product page
CollagenWeightBasis
Label disclosure35%Share of the actives the label puts a number against. An amount hidden in a proprietary blend, or simply not stated, scores zero here rather than counting as missing data — non-disclosure is a fact about a label, not an absence of one
Value30%Cost per gram of the labelled active, ranked against the category
Dose integrity20%Each audited active’s disclosed dose as a share of its reference dose (capped at the reference dose), averaged and put on a 10-point scale, minus 2 points for a proprietary blend
Consensus15%Buyer rating shown on the brand’s own product page (read from the page, or from the review service that draws its stars; source-linked), minimum 25 reviews, re-read within the last 30 days, count, source and capture date printed on every product page
Caffeine, Green Tea, Carnitine & FiberWeightBasis
Dose integrity40%Each audited active’s disclosed dose as a share of its reference dose (capped at the reference dose), averaged and put on a 10-point scale, minus 2 points for a proprietary blend
Value30%Cost per fully dosed serving, ranked against the category
Label disclosure15%Share of the actives the label puts a number against. An amount hidden in a proprietary blend, or simply not stated, scores zero here rather than counting as missing data — non-disclosure is a fact about a label, not an absence of one
Consensus15%Buyer rating shown on the brand’s own product page (read from the page, or from the review service that draws its stars; source-linked), minimum 25 reviews, re-read within the last 30 days, count, source and capture date printed on every product page
Vitamin D, Omega-3, NAC & Milk ThistleWeightBasis
Dose integrity45%Each audited active’s disclosed dose as a share of its reference dose (capped at the reference dose), averaged and put on a 10-point scale, minus 2 points for a proprietary blend
Value30%Cost per fully dosed serving, ranked against the category
Label disclosure10%Share of the actives the label puts a number against. An amount hidden in a proprietary blend, or simply not stated, scores zero here rather than counting as missing data — non-disclosure is a fact about a label, not an absence of one
Consensus15%Buyer rating shown on the brand’s own product page (read from the page, or from the review service that draws its stars; source-linked), minimum 25 reviews, re-read within the last 30 days, count, source and capture date printed on every product page
ElectrolytesWeightBasis
Value40%Cost per gram of sodium, ranked against the category — sodium is the electrolyte every label on this shelf states (49 of 49), so it is the one like-for-like denominator
Label disclosure25%Share of the actives the label puts a number against. An amount hidden in a proprietary blend, or simply not stated, scores zero here rather than counting as missing data — non-disclosure is a fact about a label, not an absence of one
Formula20%Sugar load and calorie load per serving, read from the panel
Consensus15%Buyer rating shown on the brand’s own product page (read from the page, or from the review service that draws its stars; source-linked), minimum 25 reviews, re-read within the last 30 days, count, source and capture date printed on every product page
BCAA & EAAWeightBasis
Value40%Cost per gram of essential amino acids, ranked against the category — counting the nine essential aminos and nothing else, so a tub padded with glutamine, taurine or coconut water answers on the aminos you are actually buying. The BCAAs are three of the nine, so a BCAA and a full EAA formula are priced on the same denominator
Dose integrity25%Each audited active’s disclosed dose as a share of its reference dose (capped at the reference dose), averaged and put on a 10-point scale, minus 2 points for a proprietary blend
Label disclosure20%Share of the actives the label puts a number against. An amount hidden in a proprietary blend, or simply not stated, scores zero here rather than counting as missing data — non-disclosure is a fact about a label, not an absence of one
Consensus15%Buyer rating shown on the brand’s own product page (read from the page, or from the review service that draws its stars; source-linked), minimum 25 reviews, re-read within the last 30 days, count, source and capture date printed on every product page

How the formula and value scores are worked out

Formula, on protein powders, protein bars and protein snacks. Protein per calorie: the label’s grams of protein times 4 kcal, over its calories, times 10 and capped at 10, so a product whose every calorie is protein scores 10. Each gram of sugar alcohol takes off 0.25 points, at most 3, and each amino-spiking flag takes off 1.5 (see below). A product whose record lists any third-party certification gains 1 point; the list is the record’s own certifications field, and a certification counts only once it is in that list. No step of our pipeline collects certifications yet, so none of the 679 products on record lists one and the term changes no score today. The score stays between 0 and 10. A label with no protein or calorie figure gets no formula score. On protein bars and protein snacks, a label whose ingredient list names a sugar alcohol (maltitol, erythritol, sorbitol, xylitol, isomalt, lactitol or mannitol; glycerol is not counted) while its panel figure for sugar alcohols was not recorded gets no formula score, and its page says which sugar alcohol the label lists: that term cannot be measured, and reading it as none would score a guess. A label that names none of them and has no sugar-alcohol figure is scored as having none.

Formula, on energy drinks. It starts at 10: each gram of sugar takes off 0.35 points, at most 7, and each kcal above 20 kcal takes off 0.02 points, at most 2; a panel that prints no numeric caffeine content loses 2 points (a caffeine salt with no stated caffeine yield does not count as one). The score does not go below 0. A panel with no sugar row gets no formula score.

Formula, on electrolytes. It starts at 10: each gram of sugar takes off 0.35 points, at most 7, and each kcal above 20 kcal takes off 0.02 points, at most 2. There is no caffeine term. A panel with no sugar row gets no formula score.

Value. A product’s cost, in the unit its weight table names, is ranked against every product on record in its category that has a cost figure, including those whose score is not published. Its value score is 10 times one minus the share of the other products in that group that cost strictly less: the cheapest scores 10, and a product every other one undercuts scores 0. Where fewer than 5 products in a category have a cost figure, or all of them cost the same, no product there gets a value score, and its page says why: a comparison that cannot separate products is not a score. Where 5 to 4 have one, each scores a neutral 5, because a ranking that small carries no information.

Price age. A value score is worked out only from a brand-store price read within 7 days of the scoring run, as a buyer rating counts only within its own age limit. A product whose newest brand-store price is older still shows that price with the date it was read, but gets no value score and is left out of its shelf’s value comparison, and its page gives the date.

A missing component is never scored as zero. Where a component cannot be computed — no verified price, no machine-readable buyer rating, a formula too sparse to audit — its weight is redistributed across the components that exist rather than counted against the product. A product we know little about stays distinguishable from a product we know is bad, and the page says which.

Products scored on everything a label and a price can answer for rank first. Redistributing a missing component’s weight keeps a score honest on its own page, but it means the product was never measured on that axis — a product with no verified price has not been compared on price at all. So every ranking lists first the products scored on every component except buyer rating, then those scored on at least 60% of that weighting, then the rest. A missing buyer rating does not move a product down a group: whether a brand’s page shows one splits by brand rather than by product, so it says nothing about how well the product itself was measured. Where there is one, it keeps its full weight in the composite. Within the first two groups the higher composite ranks higher; within the last, the product whose score covers more of the weighting ranks higher, and the higher composite breaks a tie. Every row prints how much of the weighting its score covers.

Below that, no score is printed at all. Redistribution has a limit. A product on the vitamin D, omega-3, NAC & milk thistle shelf measurable only on label disclosure — a tenth of that category’s weighting — was published as a 10.00, because the arithmetic handed one component the entire score. It ranked where it belonged, near the bottom, while printing the largest number on the page. “We could not measure this” is not the same statement as “this is perfect”, so where too little of the weighting could be computed the number is withheld and the coverage is shown in its place. A score is printed only where at least 60% of the weighting was computed, counted both over the whole weighting and over the part a label and a price can answer for. The ranking order does not change; the figure simply is not published, for the same reason a dose or a price is not published when the label does not support it.

Ties. A composite is worked out to 2 decimal places, the precision every ranking prints, so two products that print the same composite have the same score: there are no hidden digits to separate them. Where the rule above leaves two such products side by side, they are listed in the order they appear in our product records, which is not a judgement of either. On a head-to-head page, two figures that print the same are a tie, and the page names no winner on that measure: composites are compared to 2 decimal places, costs to 3 and dose integrity to 1, as printed.

A ranking needs enough published scores. A category is published only once at least 15 of its products carry a published score (8 for saw palmetto, where fewer scoreable products exist to be bought). Products whose score is withheld do not count toward that, because a ranking is only as large as the scores it prints. Not published today: saw palmetto (0 of 8).

A dose threshold never comes from people with a diagnosed condition. Where the only published dose for an active was studied in people with a medical condition, it is not a threshold for anyone else, so the active is recognised and named on the product page but not scored. That applies today to 7 actives: Huperzine A, Niacinamide, Minoxidil, Biotin, Silymarin (Milk Thistle), Saw Palmetto, Ketoconazole.

The word beside a composite score names its band, nothing more. Each product page prints one of 5 words next to its composite score. The bands are fixed ranges of the score out of ten, the same on every shelf; a withheld score prints no word.

WordComposite score
Exceptional8 to 10
Strong6 to under 8
Middling4 to under 6
Weak3 to under 4
Poor0 to under 3

Cost per gram of protein

Price divided by the total protein in the package. Nothing more sophisticated, and that is the point — it is the number value comes down to, and a shelf price does not show it.

A 12-count box at $24.99 with 21 g of protein per bar holds 252 g of protein, which is $0.099 per gram. Rankings are percentile-based within a category, so a product is measured against its actual competition rather than an arbitrary threshold.

Where a product is sold in several sizes, its value is the lowest cost per unit among the sizes on sale up to 5 kg (11 lb). Larger bags — some powders are sold in 20 and 25 kg bags — are wholesale quantities: the product page lists them, marked as not used in the value score.

A size’s servings are the count its store option prints, where it prints one (“60 Servings”): a flavour’s scoop can differ from the scoop on the label we read. Where the option names no count, they come from the package’s weight over the serving weight on that label, from its count of bars, cans or capsules, or from the label’s own servings per container. A printed count more than twice or less than half what the package itself gives (its weight over that serving weight, or its capsules over the capsules a serving takes) is taken for a listing error and not used; that figure is used instead.

A package’s weight is the net weight printed on its label when two independent readers transcribed the same metric figure from the same brand image of that size; otherwise it is the weight the store’s size option names (“2LB” is read as 907 g). So far 70 products have a size weighed this way, and the product page prints that weight beside the size.

Where a package size cannot be determined with confidence, the result is left empty rather than estimated. An empty cell is honest; a plausible wrong number is not.

Reference doses

Each audited active is measured against a reference dose, not against a marketing claim: the dose a cited trial gave, one end of a range a position stand states, or a population intake a guideline body sets. Each threshold’s type is printed beside it. The thresholds and their sources:

IngredientReference doseSource
Alpha-GPC250 mg
trial dose
Marcus et al., J Int Soc Sports Nutr 2017;14:39 (PMID 29042830) — 250 mg and 500 mg daily for 7 days in 48 healthy college-aged men; the threshold uses the lower dose (peer-reviewed trial)
Population: 48 healthy college-aged men across four arms (250 mg, 500 mg, caffeine, placebo)
Ashwagandha (KSM-66)600 mg
trial dose
Wankhede et al., J Int Soc Sports Nutr 2015;12:43 (PMID 26609282) — 300 mg KSM-66 root extract twice daily (600 mg/day) for 8 weeks (peer-reviewed trial)
Population: 57 young men with little resistance-training experience (29 on ashwagandha)
Beta-Alanine4 g
position-stand range
Trexler et al., ISSN Position Stand 2015, J Int Soc Sports Nutr 2015;12:30 (PMID 26175657) — 4-6 g/day; the threshold uses the low end, 4 g (position stand)
Population: position stand; healthy, exercising populations
Betaine Anhydrous2.5 g
trial dose
Cholewa et al., J Int Soc Sports Nutr 2013;10:39 (PMID 23967897) — 2.5 g/day for 6 weeks in strength-trained men (peer-reviewed trial)
Population: 23 experienced strength-trained men (11 on betaine)
Caffeine200 mg
position-stand range
Guest et al., ISSN Position Stand 2021, J Int Soc Sports Nutr 2021;18:1 (PMID 33388079) — 3-6 mg/kg; the threshold is derived from the low end, 3 mg/kg, for a 67 kg adult: 3 x 67 = 201 mg, rounded to 200 mg (position stand)
Population: position stand; trained and untrained individuals
Choline Bitartrate2 g
trial dose
Naber, Hommel & Colzato, Sci Rep 2015;5:13188 (PMID 26271904) — a single 2 g dose of choline bitartrate in 28 healthy young adults (peer-reviewed trial)
Population: 28 healthy young adults; placebo-controlled, double-blind, two sessions
Citrulline Malate 2:18 g
trial dose
Pérez-Guisado & Jakeman, J Strength Cond Res 2010;24:1215-1222 (PMID 20386132) — a single 8 g dose of citrulline malate in 41 men (peer-reviewed trial)
Population: 41 men doing bench-press training sessions; randomised double-blind crossover
Collagen Peptides (audited on Collagen only)2.5 g
trial dose
Proksch et al., Skin Pharmacol Physiol 2014;27:47-55 (PMID 23949208) — randomized double-blind placebo-controlled trial dosing 2.5 g and 5.0 g/day for 8 weeks in 69 women aged 35-55 (peer-reviewed trial)
Population: 69 women aged 35-55
Creatine HCl5 g
trial dose
Londoño-Velásquez et al., J Int Soc Sports Nutr 2025;22(Suppl 1):2533658 (doi 10.1080/15502783.2025.2533658; ISSN Conference abstract, not yet full peer review) — 5 g/day for 8 weeks, matched head-to-head against creatine monohydrate at the same dose, with a placebo arm, in elite team-sport athletes (conference abstract, not yet peer reviewed)
Population: 31 elite handball and softball players, men and women aged 18-28
Creatine Monohydrate5 g
position-stand range
Kreider et al., ISSN Position Stand 2017, J Int Soc Sports Nutr 2017;14:18 (PMID 28615996) — 3-5 g/day to maintain muscle creatine stores; the threshold uses the high end, 5 g (position stand)
Population: position stand; exercising adults
Digestive Enzyme Blend150 mg
trial dose
Majeed et al., Sports Nutr Ther 2016;1:113 (doi 10.4172/2473-6449.1000113) — randomised double-blind placebo-controlled trial, 20 healthy men, 50 mg of DigeZyme, a branded multi-enzyme complex, three times daily (150 mg/day) for 3 days (peer-reviewed trial)
Population: 20 healthy men
EGCG (Epigallocatechin Gallate)400 mg
trial dose
Roberts et al., J Int Soc Sports Nutr 2015;12:1 (PMID 25650043) — 571 mg/day decaffeinated green tea extract providing 400 mg/day EGCG, 4 weeks, recreationally active healthy men (peer-reviewed trial)
Population: 14 healthy, recreationally active men (7 on extract)
EPA + DHA (combined)250 mg
intake guideline
EFSA Panel on Dietetic Products, Nutrition and Allergies, EFSA Journal 2010;8(3):1461 (doi 10.2903/j.efsa.2010.1461) — adequate intake 250 mg/day EPA+DHA for healthy adults; FAO/WHO 2010, Fats and fatty acids in human nutrition (FAO Food and Nutrition Paper 91) — 0.250 g/day EPA+DHA for adults (guideline)
Population: adults (population reference intake)
HMB3 g
trial dose
Wilson et al., ISSN Position Stand, J Int Soc Sports Nutr 2013;10:6 (PMID 23374455) — 3 g/day, the daily dose of calcium HMB (HMB-Ca) in most of the trials it tabulates; its stated position is 38 mg/kg body mass daily (position stand)
Population: position stand; trained and untrained exercising adults
L-Carnitine (audited on Caffeine, Green Tea, Carnitine & Fiber only)2 g
trial dose
Koozehchian et al., J Exerc Nutrition Biochem 2018;22:7-19 (PMID 30661327) — 2 g/day L-carnitine for 9 weeks in healthy resistance-trained men (12 on L-carnitine, 11 on placebo) (peer-reviewed trial)
Population: 23 healthy resistance-trained men aged 18-40
L-Carnitine L-Tartrate2.931 g
trial dose
Volek et al., Am J Physiol Endocrinol Metab 2002;282:E474-482 (PMID 11788381) — 2 g/day L-carnitine, given as L-carnitine L-tartrate, in 10 resistance-trained men (peer-reviewed trial)
Population: 10 resistance-trained men; balanced crossover
L-Citrulline6 g
trial dose
Bailey et al., J Appl Physiol 2015;119:385-395 (PMID 26023227) — 6 g/day for 7 days in 10 healthy men (peer-reviewed trial)
Population: 10 healthy adult men; randomised, placebo-controlled crossover
L-Leucine (audited on BCAA & EAA only)3 g
position-stand range
Jäger et al., ISSN Position Stand 2017 (J Int Soc Sports Nutr 14:20, PMID 28642676) — acute doses should contain 700-3000 mg leucine alongside a balanced array of the essential amino acids (position stand)
Population: position stand; healthy, exercising individuals
L-Theanine100 mg
trial dose
Owen et al., Nutr Neurosci 2008;11:193-198 (PMID 18681988) — 100 mg L-theanine together with 50 mg caffeine, against 50 mg caffeine alone and placebo, in 27 healthy volunteers (peer-reviewed trial)
Population: 27 healthy volunteers; placebo-controlled, with a 7-day washout between treatments
L-Tyrosine2 g
trial dose
Colzato et al., Front Behav Neurosci 2013;7:200 (PMID 24379768) — a single 2.0 g oral dose in 22 healthy young adults (peer-reviewed trial)
Population: 22 healthy female undergraduates aged 18-25; double-blind, placebo-controlled crossover
N-Acetylcysteine1.2 g
trial dose
Leelarungrayub et al., Oxid Med Cell Longev 2011 (PMID 21904641) and Slattery et al., Med Sci Sports Exerc 2014 (PMID 24576857) — 1,200 mg/day in healthy adults; Fernández-Lázaro et al., Nutrients 2023 (PMID 37299425), systematic review: 1,200 mg/day the most common oral dose (peer-reviewed trial)
Population: healthy men: sedentary (Leelarungrayub) and well-trained triathletes (Slattery)
Nitrosigine1.5 g
trial dose
Rood-Ojalvo et al., J Int Soc Sports Nutr 2015;12(Suppl 1):P14 (doi 10.1186/1550-2783-12-S1-P14; ISSN Conference abstract, not yet full peer review) — 1,500 mg/day inositol-stabilized arginine silicate for 4 days (conference abstract, not yet peer reviewed)
Population: 16 healthy men aged 19-33 with limited exercise routines; crossover
Retinol0.3%
trial dose
Zasada et al., Skin Pharmacol Physiol 2020;33:102-116 (PMID 32428912) — clinically controlled comparison of 0.3% and 0.5% retinol serums (peer-reviewed trial)
Population: 37 adult volunteers; split-face comparison
Rhodiola Rosea200 mg
trial dose
Koozehchian et al., Nutrients 2025;17(23):3736 (PMID 41374026; PMC12693935) — 200 mg/day for 7 days, the lower of the two doses tested (200 vs 1,500 mg) in resistance-trained adults (peer-reviewed trial)
Population: 27 resistance-trained adults aged 18-40; crossover
Taurine1 g
trial dose
Balshaw et al., Amino Acids 2013;44:555-561 (PMID 22855206) — a single 1,000 mg dose in trained male middle-distance runners, randomised double-blind crossover; one of three healthy-adult trials at 1 g in the Waldron et al. Sports Med 2018 meta-analysis (PMID 29546641; doses 1-6 g) (peer-reviewed trial)
Population: 8 trained male middle-distance runners; randomised double-blind crossover
Theacrine200 mg
trial dose
Taylor et al., J Int Soc Sports Nutr 2016;13:2 (PMID 26766930) — 200 mg/day, the lower of the two doses tested (200 and 300 mg once daily for 8 weeks) (peer-reviewed trial)
Population: 60 healthy men and women
Vitamin D3 (Cholecalciferol)600 IU
intake guideline
Institute of Medicine, Dietary Reference Intakes for Calcium and Vitamin D (2011; doi 10.17226/13050) — RDA 600 IU/day for ages 1-70, 800 IU/day for 71 and older (guideline)
Population: healthy population, ages 1-70 (RDA)

Every threshold is a reference dose, and its type is printed beside it. 21 are a trial dose: the dose a cited trial gave. 4 are a position-stand range: one end of a range a position stand states (Beta-Alanine, Caffeine, Creatine Monohydrate and L-Leucine). 2 are an intake guideline: a population intake a guideline body sets for healthy people, not a dose a trial gave (EPA + DHA (combined) and Vitamin D3 (Cholecalciferol)).

Where a source states a range rather than one dose, the threshold uses one end of it. 4 of the thresholds above are taken that way, 2 from the low end and 2 from the high end:

  • Beta-Alanine, 4 g: the low end of 4-6 g/day (Trexler et al., ISSN Position Stand 2015).
  • Caffeine, 200 mg: the low end of 3-6 mg/kg (Guest et al., ISSN Position Stand 2021); 3 mg/kg for a 67 kg adult: 3 x 67 = 201 mg, rounded to 200 mg.
  • Creatine Monohydrate, 5 g: the high end of 3-5 g/day (Kreider et al., ISSN Position Stand 2017).
  • L-Leucine, 3 g: the high end of 700-3000 mg (Jäger et al., ISSN Position Stand 2017).

Some thresholds only hold for one kind of product, and are only applied there. 3 of the 27 above are scoped that way, marked in the table. Leucine’s range is stated for an acute protein dose alongside a balanced array of the essential amino acids, so it is audited on the amino shelf and not against a pre-workout that happens to contain leucine. Collagen’s is a skin-elasticity endpoint, so it is not used to judge the 17 protein bars and 20 protein snacks that list collagen as a protein source; and it was measured on hydrolysed collagen peptides, so it applies to those only, not to eggshell membrane or undenatured type II collagen, which are sold by the milligram and named on the product page without being scored. Citing a trial at a product it was not measured on would be a wrong number wearing a citation.

An active can be recognised without being scored. Where the research establishes no threshold, where a label states a raw material without the standardised fraction the threshold is written in, or where it names a different extract from the one the threshold was measured on (only a row naming KSM-66 is audited against the ashwagandha reference dose, and only a row naming a rhodiola extract against rhodiola’s), the ingredient is named on the product page and left out of the dose average — rather than being scored zero for containing something nobody has a number for, or credited at a weight that is not the active. These 36 are recognised and never scored:

IngredientWhy it is not scored
Acetyl-L-Carnitinethe 2,000 mg dose-response applies to free L-carnitine, and this is a different material
Ashwagandha (KSM-66 not named)the reference dose is the KSM-66 root extract the trial gave, and this row does not name it
Betaine Nitrate (betaine fraction not declared)the betaine fraction of the nitrate salt is not declared
Biotinno established dose outside diagnosed deficiency
Cod Liver Oil (EPA/DHA not declared)research doses combined EPA+DHA, and this row declares neither
Collagen, not hydrolysed peptides (eggshell membrane, undenatured type II)the reference dose is hydrolysed collagen peptides, and this row names a different collagen material (eggshell membrane or undenatured type II collagen)
Curcuminthe dose trials used depends on the formulation, so one milligram figure would not mean the same thing twice
Glucomannanthe reference amount and the trials behind it are in overweight or obese adults, so it is not a threshold for anyone else
Green Tea Extract (EGCG fraction not declared)the threshold is EGCG content, and this row declares no EGCG figure
HMB Free Acid (HMB-FA)the HMB reference dose is calcium HMB by weight, and free-acid HMB is a different material; no free-acid dose has been chosen
Huperzine Athe published dose comes from studies in people with a diagnosed condition, so it is not a threshold for anyone else
Hyaluronic Acidno concentration-response is established, and molecular weight is not disclosed
Ketoconazolethe published dose comes from studies in people with a diagnosed condition, so it is not a threshold for anyone else
Konjac Root (glucomannan fraction not declared)the threshold is glucomannan, and this row states the root material without declaring how much of it is glucomannan
L-Histidineno per-amino threshold established for supplementation
L-Isoleucineno per-amino threshold established for supplementation
L-Lysineno per-amino threshold established for supplementation
L-Methionineno per-amino threshold established for supplementation
L-Phenylalanineno per-amino threshold established for supplementation
L-Threonineno per-amino threshold established for supplementation
L-Tryptophanno per-amino threshold established for supplementation
L-Valineno per-amino threshold established for supplementation
Magnesiumreplacement need is individual
Milk Thistle Extract (silymarin fraction not declared)research doses silymarin content, and this row declares no silymarin figure
Minoxidilthe published dose comes from studies in people with a diagnosed condition, so it is not a threshold for anyone else
N-Acetyl L-Tyrosine (NALT)a different compound from the L-tyrosine the cited trial dosed
Niacinamidethe published dose comes from studies in people with a diagnosed condition, so it is not a threshold for anyone else
Potassiumreplacement need is individual
Psyllium Huskthe only published amount is the qualifying intake a US regulation sets for a disease-risk health claim, not a dose a trial in healthy adults gave, so it is not a threshold for anyone
Pumpkin Seed Oilno dose-ranging trial establishes a threshold
Rhodiola Rosea (root, extract not declared)the reference dose is a standardized root extract, and this row is not declared to be one
Saw Palmettothe published dose comes from studies in people with a diagnosed condition, so it is not a threshold for anyone else
Saw Palmetto (powder, or material not declared)the trials dosed a standardised lipophilic extract, and this row is not declared to be one
Silymarin (Milk Thistle)the published dose comes from studies in people with a diagnosed condition, so it is not a threshold for anyone else
Sodiumreplacement need is individual
Vitamin C (L-Ascorbic Acid, topical)its only source is an absorption study in pig skin, not a study in healthy adults

Citrulline malate and free L-citrulline are tracked separately. They are not interchangeable, and treating them as one misstates the delivered dose — which would undermine the whole exercise.

Exceeding a reference dose does not raise a score. Meeting it is the bar; megadosing is not a virtue we reward.

How the dose-integrity score is worked out

Each active the reference gives a threshold for is audited once. Rows naming the same active are added together first (a label splitting one active across two forms delivers their sum; a concentration printed twice counts once), and a standardised extract counts only the active fraction its label declares. Creatine is one active however many forms a label lists it in: where a label lists more than one creatine form (monohydrate and HCl), their stated amounts are added together and audited once against the creatine monohydrate reference dose, and a form inside a blend with no amount of its own adds nothing. A caffeine salt or complex (di-caffeine malate, zumXR and the like) weighs more than the caffeine in it, so its row counts only at the caffeine yield its label states; a row that states none is shown as caffeine not stated, and only the caffeine the label does state is counted. Then:

  1. Each audited active’s disclosed dose is divided by its threshold, and the result is capped at 1. A dose at or above the threshold counts 1; a dose below it counts its share, so a dose at 80% of its threshold counts 0.8. An active with no disclosed dose — inside a proprietary blend, in a branded complex under one shared weight, or printed with no amount — counts 0.
  2. Those shares are averaged across the audited actives and multiplied by 10.
  3. A label that carries a proprietary blend loses 2 points from that. The score never goes below 0.

Actives the reference recognises without a threshold are named on the product page and left out of the average. Where fewer than 3 actives can be scored and the reference recognises under 50% of the label’s dosed or blended actives, no dose-integrity score is given: a score for the few actives the audit understood would not be a score for the product.

That share, printed on every product page as its coverage line, counts actives only. A row with a printed weight is left out when it is a figure about other rows rather than one more ingredient: a blend’s header (its members are counted), a sub-total that the dosed rows printed beneath it add up to (“2:1:1 BCAA” over leucine, isoleucine and valine), a total such as “Total Omega-3 Fatty Acids”, a footnote equivalence row (“Herbal Equivalent”), and an acidulant, sweetener, flavouring, flow agent or colour the label happens to weigh.

Dose integrity is part of the score on pre-workouts; creatine; collagen; caffeine, green tea, carnitine & fiber; vitamin D, omega-3, NAC & milk thistle; and BCAA & EAA. It is not part of the score on protein powders, protein bars, protein snacks, energy drinks and electrolytes, and their head-to-head pages print no dose-integrity figure.

Proprietary blends

An ingredient inside a proprietary blend has no disclosed dose, so it is scored as undosed rather than assumed adequate, and on a shelf scored on dose integrity the product loses a further 2 points of its dose-integrity score, out of 10. Disclosure is the one thing on a label a brand fully controls: in every category scored on dose, a hidden amount earns nothing for dose integrity, and nothing for label disclosure where that is scored too.

Spiking flags

On protein powders, protein bars and protein snacks, the check looks for glycine, taurine or creatine among the first 5 entries of the label’s printed ingredient list, which runs in descending order by weight. Each ingredient it flags lowers the product’s formula score by 1.5 points, out of 10 (the formula score does not go below 0). The product page names the ingredient, its position and why that placement matters, and links the panel it came from. A flag reports where an ingredient sits on the list, not how much was added or why, and it is not an accusation of fraud.

What the check cannot see. It reads only the ingredient list. On a Supplement Facts label an ingredient already named in the panel need not be repeated in that list, so glycine, taurine or creatine declared only in the panel is not seen and raises no flag. Where the order of the list cannot be established reliably (a panel parsed from a brand’s machine-readable data, or a label image with no ingredient list), the check is not run and the product page says so, rather than reporting a clean result it did not earn.

Rules we do not break

Every published number traces to a captured source. A product page renders only after its panel has been read and stored, so a rating can always be checked against the label it came from.

A panel is captured one of two ways, and every product page names which. Most are read from the label image by two independent readers, and nothing publishes unless two independent readings recorded the same value — where the first two disagree, even on one ingredient, the product is held back until a third independent reading agrees with one of them, rather than published with a number we are unsure of. A minority come from brands that publish their panel as machine-readable data on their own product page, which is parsed directly; there is no second reading to disagree with, because there is no image being interpreted. Each product page says how its panel was read, and how many readings agreed, under Provenance, so the distinction is never hidden behind an average.

What “verified” means. Wherever this site counts “verified” products, verified means: two independent readers agreed on the label, or the figures were parsed from the brand’s own machine-readable nutrition table. Of the 669 verified products today, 598 were read from the label by two independent readers who agreed, and 71 were parsed from a brand’s own table (Quest Nutrition’s).

No rating, review count, or score is ever invented. Where evidence is thin the page says not enough data, which is always an acceptable answer here. Star ratings are never fabricated to decorate a listing.

A reformulation silently invalidates an audit. When a changed label comes to light — through a correction, or the daily check that the brand’s page still shows the label image our readers transcribed — and a reviewer confirms its figures differ, the product page says the brand has since updated the label and that its figures are from the label read on the date shown, until two independent readers have re-read the new label. The product stays published, and its score does not change, in the meantime. Every price and buyer rating prints the date it was captured, so its age is never hidden, and a label image the brand has since taken down is linked from an archived copy or marked as removed.

When a figure we published turns out to have been wrong, we correct it and list it in the corrections log: the date, what was published, what it was corrected to, why it was wrong and how we found it. If you find a number that does not match its source, please tell us through the contact page.

Commission does not buy a rating. Some links earn us a commission and those pages say so. No brand can pay for a score, raise one, or remove one, and products are rated the same whether or not there is money attached.

Method changes

Changes to how products are scored, newest first. A changed threshold or rule can move a product’s score with no change to its label or price. Dose figures below are the threshold before and after each change; the table of reference doses above shows every one in use today.

· Collagen cost per gram counts collagen peptides only

Eggshell membrane and undenatured type II collagen were counted as stated collagen in the cost per gram of collagen, though they are dosed in milligrams and are no longer audited against the collagen peptide dose. They are now left out of cost per gram too; a product whose only stated collagen is one of them has no collagen cost figure, and the page says why.

· Creatine forms on one label are audited as one active

A label that lists creatine in more than one form, such as monohydrate and HCl, was audited as two actives, each against the full creatine reference dose, so adding a second form of creatine could lower its dose score. Their stated amounts are now added together into one creatine active, audited against the creatine monohydrate reference dose. A form inside a blend with no amount of its own still counts as undisclosed. A label with one form is audited exactly as before.

IngredientBeforeAfterWhy
Creatine Monohydrate5 g5 gSame figure; where a label lists more than one creatine form, their stated amounts are added and audited once against it.

· Topical vitamin C is no longer scored

Its only source is an absorption study in pig skin, which breaks the rule that thresholds come from studies in healthy adults. A label that states topical vitamin C now has it named on the product page with that reason, and it is left out of the dose average. The source is still cited. No product on record states it, so no score moved.

IngredientBeforeAfterWhy
Vitamin C (L-Ascorbic Acid, topical)20%not scoredNo longer scored: the only source is an absorption study in pig skin, not a study in healthy adults.

· The collagen reference dose applies to hydrolysed collagen peptides only

The collagen reference dose comes from a trial that gave hydrolysed collagen peptides by the gram. Eggshell membrane and undenatured type II collagen are different materials sold by the milligram, and were audited against it. They are now named on the product page and not scored against it. They still count as the collagen a label states when a collagen product is priced per gram.

IngredientBeforeAfterWhy
Collagen Peptides2.5 g2.5 gSame figure; it now applies to hydrolysed collagen peptides only, not eggshell membrane or undenatured type II collagen.

· Value uses only a recent brand-store price

A buyer rating stops counting once it is too old; a price had no limit, so a price kept from an earlier read could score value for weeks. A value score is now worked out only from a brand-store price read within 7 days of the scoring run. An older price is still shown with the date it was read, but the product gets no value score, is left out of its shelf’s value comparison, and its page gives the date of the newest price. Because value is ranked against the whole shelf, other products’ value scores can move when one leaves the comparison.

· A missing sugar alcohol figure is not read as none when the label lists one

On protein bars and protein snacks, a panel figure for sugar alcohols that was not recorded counted as none. Where the ingredient list names a sugar alcohol, such as maltitol, erythritol, sorbitol, xylitol, isomalt, lactitol or mannitol, that term cannot be measured, so the formula score is now withheld and the page says which sugar alcohol the label lists. Glycerol is not counted. A label that names none is scored as before.

· The certification term in the formula score is published

The formula score on the protein shelves adds one point when a product’s record lists a third-party certification, still capped at ten. The term was applied but not stated; the method page now states it, and where the list comes from. No step of the pipeline collects certifications yet, so no record lists one and no score moved.

· A value comparison that cannot separate products is withheld

Where fewer than five products on a shelf had a cost figure, or every one cost the same, each scored a neutral five on value, and that counted as a measured component. Value is now withheld there, with the reason on each page, and the remaining components are reweighted.

· Ashwagandha and rhodiola are audited only on rows naming the extract their trial gave

The ashwagandha reference dose comes from a trial of KSM-66, one branded root extract, and was applied to any ashwagandha row. Only a row that names KSM-66 is now audited against it; another ashwagandha row is named on the product page and not scored. A rhodiola row is audited only where it declares a rhodiola extract, not where another extract is named before rhodiola. No product on record was affected.

IngredientBeforeAfterWhy
Ashwagandha (KSM-66)600 mg600 mgSame figure; only a row that names KSM-66 is audited against it.
Rhodiola Rosea200 mg200 mgSame figure; only a row declaring a rhodiola extract is audited against it.

· Creatine with caffeine or electrolytes is ranked on the creatine shelf

A powder whose label doses at least three grams of creatine in a row of its own is a creatine product, priced per gram of creatine, whatever else it adds, such as caffeine or electrolytes. A pre-workout and a protein powder keep their shelves. Two creatine and caffeine powders moved from the energy ranking, where they were priced per serving against cans and shots, to the creatine ranking. Their pages keep their addresses.

· Collagen is priced per gram of collagen only

Cost per gram on the collagen shelf divided the price by the grams of every scored active a label stated outside a blend. On a few labels that included an amino acid profile row matched to leucine, whose reference dose belongs to the amino acid shelf. Collagen is now priced on the label's collagen rows only, as creatine is priced on creatine. Those labels' cost per gram rose slightly, and a label that states no collagen amount of its own no longer has a cost per gram. Because value is ranked against the whole shelf, other collagen products' value scores moved slightly.

· A product can be shown but not ranked

A product the owner judges unlike the rest of its shelf is shown on its own page and left out of the ranking, its facets and its comparisons. It is not scored on value and is not part of the shelf's value comparison, and its page and the ranking say why, with the label figure that decided it. The first is a collagen capsule whose label states a collagen amount far below the rest of the collagen shelf.

· A bare vitamin D row is credited when the panel names its form elsewhere

A panel that prints a nutrient as a bare row with an amount, such as vitamin D, and names its form elsewhere without an amount, such as cholecalciferol inside a blend, is now audited on the stated row, counted once as that form. The unquantified mention beside it is not audited separately. Before, the stated amount was not matched and the form was shown as undisclosed in a blend. This follows the earlier rule that counts choline once.

· Rhodiola is audited only when the row declares an extract; HMB's reference dose is calcium HMB

The rhodiola reference dose comes from a trial that gave a standardized root extract. A rhodiola row that declares an extract is audited against it as before. A row that names the root, or rhodiola alone, without declaring an extract is now recognised and reported by name but not scored, as saw palmetto powder is. The HMB reference dose is now stated as calcium HMB by weight, the form its source's trials gave. A free-acid HMB row is recognised under its own name and not scored until a free-acid dose is chosen; no product on record prints one, so no HMB score moved.

IngredientBeforeAfterWhy
Rhodiola Rosea200 mg200 mgSame figure; the material is now stated: a root extract standardized to 3% salidroside and 1% rosavin, per the trial's Methods. A row that declares no extract is no longer audited against it.
HMB3 g3 gSame figure, stated as 3 g of calcium HMB by weight. Free-acid HMB is a separate, unscored material.

· The coverage line counts actives only

Every product page prints how many of the label's dosed actives the dose reference recognises, and where fewer than three actives can be scored that share decides whether a dose-integrity score is given at all. Its count of dosed rows included rows that are not one more ingredient: a blend's header, whose members were already counted; a sub-total that the dosed rows printed beneath it add up to, such as a BCAA line over leucine, isoleucine and valine; a total such as total omega-3 fatty acids; a footnote equivalence row; and acidulants, sweeteners, flavourings, flow agents and colours a label happens to weigh. Those rows are no longer counted, so the line counts actives only. No threshold changed. Two products are now given a dose-integrity score that was withheld before. On one of them the cost per fully dosed serving can now be worked out too, and because value is ranked against every product on its shelf, the value scores of the other products on that shelf moved slightly.

· Thresholds are called reference doses, each with its type

The method called every threshold a clinical dose, including two that are population intake guidelines rather than doses a trial gave, and several taken from one end of a range a position stand states. Every threshold is now called a reference dose, and each is printed with its type: a trial dose, a position-stand range, or an intake guideline. No threshold or score changed.

· Caffeine salts and complexes count only at the caffeine yield their label states

A label row that is a caffeine salt or complex, such as di-caffeine malate or zumXR, prints the weight of the salt or complex, not of the caffeine in it, and the dose audit counted that whole weight as caffeine. Such a row now counts toward caffeine only at the caffeine yield its label states. A row that states none is shown as caffeine not stated, is named on the product page, and adds nothing; only the caffeine the label does state is counted. Plain caffeine, and a caffeine row that names a salt only as what the stated caffeine is supplied as, count as before. Four published pre-workouts carry such a row without a stated yield, and one of them now sits below the caffeine threshold.

· Buyer ratings from the review service first where a store has one

For a store whose product pages draw their stars through a review service we read, the capture took the rating from the page's own structured data whenever the page carried one, and asked the service only when it did not. Some pages' structured data trails the stars the page displays, so the service was never asked. The review service is now read first, and the page's structured data is used only when the service gives no rating for the product. Stores read through no review service are read from the page as before. Three more stores are now read through the review service their pages use, each checked against its pages as displayed first.

· A size's servings from the count its option prints

Value divides a price by the servings a size holds. Where the store's size option names a weight, that weight was divided by the serving weight on the label we read, even when the option also printed its own servings count; a flavour whose scoop differs from that label's was then priced at the wrong number of servings. A size option that names its own servings count is now priced per that printed count, and the label's serving weight is used only when the option names none. The amount of protein or active ingredient per serving still comes from the label. A printed count more than twice or less than half what the package's own weight gives is taken for a listing error and not used.

· L-theanine threshold re-cited to a trial that gave that figure

The L-theanine threshold cited a trial whose dose was close to the threshold but not equal to it. It now cites a trial in healthy volunteers that gave the threshold's own figure, together with caffeine, as the earlier trial also did. The figure is unchanged, so no score moved.

IngredientBeforeAfterWhy
L-Theanine100 mg100 mgSame figure, new source: a trial that gave 100 mg together with 50 mg caffeine. The previous source gave 97 mg with 40 mg caffeine.

· Printed weights within 5 percent of their own size are kept

The check that keeps a size from being priced at another size's label refused correct readings where a store lists sizes only fractions of a pound apart. A printed net weight within 5 percent of the weight its own size is sold as is no longer refused by that check; it still needs two readers to agree on it.

· Package weights from the net weight printed on each size's label

Value divides a price by the package's weight, and that weight came from the store's name for the size, which can differ from what the label prints. A size is now priced at the net weight printed on its own label when two independent readers, reading the same brand image, agree on that figure; otherwise the store's size name is used as before. A printed weight nearer another listed size than its own is not used.

· Choline, N-acetyl-L-tyrosine, citrulline DL-malate and graded alpha-GPC rows in the dose audit

Four label forms are now matched to the dose reference as follows. Where a label prints a choline amount beside the choline bitartrate it comes from, choline is counted once: the amount is not added to the bitartrate row. N-acetyl-L-tyrosine is recognised as its own compound and is not scored, because the cited trial gave L-tyrosine. L-citrulline DL-malate is audited as citrulline malate. An alpha-GPC row whose name states its grade is read at that grade, as half its printed weight.

· Graded alpha-GPC, branded complexes, betaine nitrate and branded theacrine rows in the dose audit

Four label forms are now read as follows. An alpha-GPC row whose name opens with its grade is read at that grade, as half its printed weight, when nothing else is declared. An active named only as one part of a branded complex, with no amount of its own, gets no dose credit, as in a proprietary blend. Betaine nitrate is recognised as its own compound and is not scored, because a label giving the salt's weight does not state how much of it is betaine; a row that states its betaine is audited as betaine. A theacrine row that names theacrine first, with its branded, standardized source after it in brackets, is read at the amount printed.

· Two readers must have read the same label image

Label figures publish only when two independent readers agree. Their agreement now counts only when both read the same image file: a reading of another flavour's or size's label, or of a marketing image, is not paired with it. Products that had been verified across two different images were read again from the published image.

· Dose thresholds checked against the population and figures of their cited sources

Every scored threshold was checked against its source. A threshold must come from research in healthy adults, and its figure must be one the source states. Where a source pooled trials in people with diagnosed conditions it was replaced; where the figure did not match the source it was corrected; one threshold with no healthy-adult source is no longer scored.

IngredientBeforeAfterWhy
L-Carnitine2 g2 gSame figure, new source: a trial in healthy men. The previous source was a meta-analysis that pooled trials in people with diagnosed conditions.
EGCG (Epigallocatechin Gallate)300 mg400 mgNow from a trial in healthy, recreationally active men. The previous source was a meta-analysis that pooled trials in people with diagnosed conditions.
L-Tyrosine1 g2 gNow from a trial in healthy adults. The previous source was a review that mixed clinical and healthy populations.
Glucomannan3 gnot scoredNo longer scored: the reference amount and the trials behind it are in overweight adults, not a general healthy-adult population.
Beta-Alanine3.2 g4 gThe cited position stand states 4-6 g a day; the threshold now uses its low end.
Alpha-GPC300 mg250 mgNow cites a trial that dosed 250 mg and 500 mg daily in healthy men, and uses the lower dose. The previous figure was not a dose its cited trial used.
L-Carnitine L-Tartrate2 g2.931 gThe cited trial gave 2 g of L-carnitine as the tartrate salt. Labels state the salt weight, which is 68.23% L-carnitine, so the threshold is now the salt weight that carries 2 g.
Caffeine200 mg200 mgSame figure; its derivation is now stated: the low end of the position stand's 3-6 mg/kg, for a 67 kg adult.

· N-acetylcysteine and EPA + DHA scored again, from healthy-adult sources

Both had been unscored since 2026-10-04, when their earlier thresholds were withdrawn. Each now has a threshold from research in healthy adults: N-acetylcysteine from trials in healthy exercising adults, EPA + DHA from the EFSA and FAO/WHO adequate intake for healthy adults. With these scored, the vitamin D, omega-3, NAC & milk thistle ranking began publishing.

IngredientBeforeAfter
N-Acetylcysteinenot scored1.2 g
EPA + DHA (combined)not scored250 mg

· Buyer ratings read from the review service a brand's page uses

Some brand product pages do not include their star rating in the page itself; it is drawn in the browser by a review service. For those stores the rating is now read from the same service the page uses, so the figure scored is the one shown on the brand's own page. A store is added only after its figures are checked against the page as displayed.

· Buyer ratings older than 30 days are no longer scored

A rating is kept when a page stops showing one, so some products were being scored on figures more than a month old. A rating not re-read within 30 days now drops out of the score. The product page still shows it, with the date it was last read and a note that it is not scored.

· Packages over 5 kg are left out of value pricing

Value took the lowest unit price across every size on sale, which for some powders was a 20 or 25 kg wholesale bag. Sizes over 5 kg (11 lb) are now listed on the product page but not used in the value score.

· Thresholds from studies in people with a diagnosed condition withdrawn

A dose studied only in people with a diagnosed condition is not a threshold for anyone else. The thresholds below came from such studies, so these ingredients are now named on product pages and left out of the dose average.

IngredientBeforeAfter
EPA + DHA (combined)1 gnot scored
Huperzine A0.4 mgnot scored
Ketoconazole1%not scored
Minoxidil2%not scored
N-Acetylcysteine1.2 gnot scored
Niacinamide4%not scored
Psyllium Husk7 gnot scored
Saw Palmetto320 mgnot scored
Silymarin (Milk Thistle)420 mgnot scored

· Amazon prices and star ratings are no longer used

Amazon's Associates policy allows its prices to be shown, and its star ratings to be used, only on terms the ones held here did not meet: they had been read from listings in a browser, not obtained through Amazon's API. Amazon links remain. Prices come only from brand storefronts, and buyer ratings only from brands' own product pages.