How we audit nootropics

How we calculate dose fit

For each active ingredient, we divide the amount in one standard serving by the study dose we use as that ingredient's benchmark, up to full credit. If the benchmark is 250 mg and a serving has 125 mg, the ingredient gets half its dose-fit points. Most benchmarks are the lowest dose that helped in a controlled human study. Some are deliberately conservative, set above the smallest dose any study tried, and every dose card names the study its benchmark comes from. Every active ingredient counts equally, so one large ingredient can't hide several small ones. Vitamins and minerals aren't scored, unless a product is sold to correct a deficiency. Magnesium L-threonate is the one exception, because its trials measure thinking rather than a deficiency. Absorption aids such as black pepper extract, bromelain and AstraGin stay out of the score too, because brands add them to help other ingredients get absorbed. You'll still find them listed in the label audit.

We count the active compound

Some ingredients come as a salt, and part of the weight is the salt. We record the amount the label prints. Where the studies dosed the active compound, we score the active amount the salt supplies. Vyvamind lists 200 mg of citicoline sodium, which supplies about 191 mg of citicoline, so its score uses 191 mg against the 250 mg the studies started from, about 59 mg short. N-acetyl-L-tyrosine is counted the same way, by the tyrosine it carries. Where the studies dosed the salt itself, as the PQQ trials did with PQQ disodium salt, the label figure is the one we score.

We score the label's standard serving

We use the serving the label states. Some brands suggest taking more to reach a studied dose, as Mind Lab Pro does with four capsules instead of two. We score the standard serving and report what the larger one would cost and change.

The form has to match the research

When the research depends on a particular extract or strength and the label doesn't establish it, the ingredient gets at most half credit, even if the milligrams look right. Thesis Stress Reset's 28 mg of saffron matches the trials' amount, but its certificate shows a different extract from the one they tested.

We don't guess what's inside a blend

When a label hides amounts inside a proprietary blend, we record the blend's total and score only the amounts printed outside it. An amount printed for something inside a blend, such as a caffeine total, isn't scored either. Where a review shows an estimate, as Alpha Brain's does from its older labels, the estimate is used only to find similar products and is never scored, and the missing amounts cost the product transparency points instead.

Ingredients with no studied dose

Some ingredients have no dose we can compare with. Dihydrohonokiol-B has no human research at all. Velvet bean has human trials, but in Parkinson's disease, with none showing a benefit for memory or focus. Both count in the average, earn nothing for dose and get the same small evidence credit as anecdotal research.

How we grade the evidence

Each ingredient gets one of four grades for the human research behind the effect it's sold for, such as memory, focus or calm. A strong grade for one effect doesn't carry over to another.

• Strong means several trials against a placebo (dummy pill), or reviews pooling many of them, that mostly agree. Caffeine and bacopa are graded strong. It earns full evidence points.
• Moderate means some placebo-controlled trials point the same way, but too few or too small to settle the question, as with L-theanine and rhodiola. It earns 70 percent of the points.
• Preliminary means a few small human studies, or research mostly in animals and the lab, as with lion's mane and citicoline. It earns 40 percent.
• Anecdotal means only user reports or traditional use, with no controlled human study for that effect. It earns 15 percent.

When the research falls between two grades, we give the lower one. The points then scale with dose fit, so an ingredient at half its studied dose earns half its evidence points.

Trials of the finished product

A peer-reviewed human trial of the finished product adds 10 points to the evidence factor, but only when its main result for the whole group favours the product. A benefit found only in a subgroup or a side measure doesn't count. The brand can have funded the trial or matched its groups rather than fully randomizing them, as long as the review says who funded it and how it was run. Alpha Brain earns the 10 this way, because its whole trial group beat placebo on delayed recall and executive function. Trials by independent researchers that repeat the result add 20 instead, and several brand-funded trials add 10 between them, not 10 each. The evidence factor can't go above 100, and an unpublished trial or a customer survey adds nothing.

How we verify testing

A “third-party tested” line on a label earns nothing on its own. We look for the report behind it, and what a buyer can actually see decides which of four levels the product gets.

• Fully verified, 100 points. A report from an independent lab that anyone can open, for one batch, less than 12 months old, that measures the ingredients against the label. A current listing with a certification program such as NSF counts too. A lab the brand founded itself counts here if it's accredited and publishes batch reports, and the review names the relationship and the lab's accreditation number.
• Partially verified, 70 points. A check you can only half see, such as a batch lookup that works only after you've bought the bottle, a named program that doesn't publish results, a batch report over 12 months old, or an outside report that confirms an ingredient is there without measuring how much.
• Process disclosed, 30 points. A believable testing process the brand describes, naming the lab or method, with no results to see. A report only from the company that makes the product for the brand sits here too.
• Unverified, 0 points. Nothing we could check. It says nothing about whether the product is safe, only that we couldn't confirm what's in it.

A report counts in full only when it measures every active ingredient the score counts. One that measures some of them earns at most the 70 points of partial verification, and the review says which ingredients it measured. A batch report only earns 100 or 70 points when it names an independent lab, because a report that doesn't name its lab can't show it's independent, so it counts as a disclosed process. A batch lookup tool stays at 70 either way.

The words we use for a brand's claims

Each review quotes the sales page's main claims and gives each one a verdict. These four words mean the same thing on every page.

• Holds up. The research or documents we found support the claim as the brand words it, at the amounts in the product.
• Partly. Some of it is supported, but the wording goes further than the evidence, or the support depends on something we couldn't confirm.
• Doesn't hold up. What we found contradicts the claim, or the product's amounts fall short of what it would need.
• Unverified. We found nothing to check it against. That isn't the same as finding it false.

How we choose our pick

In each category, our pick is the product with the highest audit rating. When two products are within 0.3 of each other, the one that costs less for the studied dose at the brand's own store takes the spot, and if that's level too, the one with the better lab report. Affiliate offers don't decide a pick, and price never changes a score.

What isn't in the score

Price, buyer ratings, outside reviewers' verdicts and affiliate commissions all sit outside the rating. Each review shows them separately, so a cheap formula that scores poorly still reads as poor, and an expensive formula that scores well can still be poor value. Commission is never an input to a score, and the scoring script has no field for it. First-hand testing stays out of the score too, and how we test explains what it will add.

Before a review goes live, a separate review pass checks its numbers, citations and claims against the sources. Anything it finds wrong is fixed before the page is published.